Skip to content

Latest commit

Β 

History

102 Commits

Folders and files

NameName
Last commit message
Last commit date
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

Finance LLMs

Compilation of Real-World LLM & AI Agent Use Cases in Financial Services

Context

Large language models (LLMs) and AI agents are transforming financial services: from streamlining regulatory compliance and accelerating deal analysis to personalizing customer experiences and automating complex workflows across banking, trading, and asset management.

This repository curates real-world LLM and agentic AI implementations in finance, showcasing enterprise platforms, specialized models, and integrated solutions deployed by leading financial institutions worldwide.

Have a great example of LLMs in the financial services sector? Contribute by opening a pull request or submitting an issue.

Contributions Welcome!


πŸ“‘ Table of Contents

  • Retail & Commercial Banking
    • Covers personal and business banking servicesβ€”checking and savings accounts, mortgages, SME financing, and credit solutions. Also addresses central banking operations, digital transformation in traditional banks, and AI-driven advisory for both retail and corporate clients.
  • Wealth & Asset Management
    • Encompasses asset and wealth advisory, mutual funds, private equity, institutional asset management, robo-advisory, and portfolio management for high-net-worth individuals and institutional clients.
  • Capital Markets & Trading
    • Covers investment banking, securities trading, market infrastructure, equity research, credit analysis, and financial market analytics. Includes broker-dealers, exchanges, and financial data platforms.
  • Payments & Transaction Services
    • Focuses on digital payments, card networks, cross-border transactions, BNPL, and payment infrastructure. Also highlights AI-based fraud detection, risk assessment, and automated payment processing innovations.
  • Insurance & Risk Management
    • Addresses life, health, property, and casualty insurance, along with reinsurance, underwriting automation, and actuarial modeling. Features AI-driven claims processing, fraud detection, personalized policy recommendations, and broader risk assessment tools.

Each example can be classified into one of 3 categories based on type of use case:

  • Enterprise Platform: Large-scale LLM deployments spanning multiple lines of business or the entire organization. These implementations typically address a broad range of tasksβ€”such as customer support, internal operations, and complianceβ€”within a unified, centrally managed framework.
  • Fine-Tuned Model: Models either pre-trained or fine-tuned on focused datasets (e.g., regulatory filings, financial news) to achieve deep domain expertise. By narrowing the training data to specific tasks and terminology, they deliver higher accuracy and more relevant insights than general-purpose models.
  • Integrated Solution: Vendor-provided or off-the-shelf AI solutions deployed for specific workflows and use cases. Financial institutions leverage ready-made integrations, cloud services, and prompt engineering to streamline tasksβ€”such as AI-powered chatbots or document processingβ€”without extensive internal development.

Retail & Commercial Banking

Name Type Date Description Site Paper
HSBCnio Integrated Solution Oct 2026 HSBC launched HSBCnio, an AI-enabled corporate and institutional transaction banking platform that exposes authorized account and transaction data via HSBC's Model Context Protocol (MCP). Corporate treasury teams connect internal AI tools on a read-only basis β€” under existing HSBCnet logins and permissions β€” for liquidity forecasting, variance analysis, cross-border cash balance analysis, and reporting, alongside APIs and a planned "Ask HSBC" query feature. Rolled out in Hong Kong (~4,000 multinational regional offices); no external LLM vendor is named. πŸ”— -
NatWest Complaint Investigation Agent Integrated Solution Oct 2026 NatWest AI Research deployed a single-agent complaint investigation system built on the OpenAI Agents SDK, GPT-4.1, and GPT-4.1-nano into UK retail banking production (distinct from Cora). The agent queries 16 read-only, Pydantic-validated tools across call transcripts, KYC, and dispute records to assemble source-cited chronological timelines and policy-aligned summaries for human handler review under FCA rules. Across 1,234 production complaints (~120k median tokens), it matched or exceeded human investigations in 90% of 187 expert-paired cases, cutting median time from 30 minutes to 23 seconds. - πŸ”—
PRAGMA (Revolut) Fine-Tuned Model Oct 2026 Revolut pre-trained PRAGMA, a family of encoder-only Transformer foundation models (10M, 100M, and 1B parameters), on 24 billion banking events (207 billion tokens) from 26 million pseudonymised user records across 111 countries, outperforming task-specific baselines on credit scoring, external fraud, product recommendation, and LTV prediction while eliminating weeks of manual feature engineering. In October 2026, Revolut disclosed it is scaling PRAGMA in live production on NVIDIA Blackwell Ultra GPUs using NVIDIA cuDF and Dynamo-Triton inference software, servicing an initial 80 million customers and engineered to scale 3x. πŸ”— πŸ”—
ATUM (Lloyds Banking Group) Fine-Tuned Model Oct 2026 Lloyds Banking Group built ATUM, a proprietary encoder-only Transformer transaction foundation model that converts billions of sequential banking transactions into reusable customer embeddings across bank-wide workflows. Accelerated on NVIDIA RTX PRO 6000 Blackwell GPUs using self-supervised pre-training, ATUM replaces redundant task-specific feature engineering across fraud detection, loan affordability, attrition, and personalization. Scaled from 10M to 100M parameters (with a 1B model planned) over 18B+ annual events across ~28M customers; downstream lift percentages were not disclosed. πŸ”— -
Huntington Bank The 1866 Studio Integrated Solution Oct 2026 Huntington Bank ($284B in assets) unveiled The 1866 Studio (early access), an AI-powered operating system for small business owners built on Google Cloud's Gemini platform and multimodal Gemini models. Built-in agents guide entrepreneurs through entity selection and registration, assess financing readiness by flagging eligibility gaps, generate marketing collateral, and orchestrate daily financial and operational tasks while keeping owners in control of decisions. Early access opened October 19, 2026; no adoption metrics were disclosed. πŸ”— -
ItaΓΊ Unibanco & OpenAI Enterprise Platform Oct 2026 Brazil's ItaΓΊ Unibanco partnered with OpenAI to launch ia.i, a multimodal text-and-voice banking assistant spanning checking, credit, cards, investments, and insurance, governed by IARA, ItaΓΊ's internal multi-model orchestration and compliance platform. Phase one reached roughly 300,000 retail customers ahead of a planned full-base rollout by end-2026, while an internal OpenAI-powered legal search and workflow engine indexing 90M+ documents cut legal process times by 20%–80%. Specific OpenAI model versions are not named. πŸ”— -
TD Bank Mortgage Agentic AI Integrated Solution Oct 2026 TD Bank Group expanded its in-house agentic AI suite β€” co-developed with its Layer 6 AI lab (specific LLMs unnamed) β€” across Real Estate Secured Lending, extending May 2026's mortgage/HELOC pre-adjudication agent into mortgage funding and discharges. Autonomous agents classify documents, calculate income, validate policy, assemble cross-system funding summaries, and compile discharge statements under mandatory colleague review. Pre-adjudication dropped from 15 hours to under three minutes, while funding summaries cut initial review time by up to half. πŸ”— -
Bradesco & Gemini Enterprise Enterprise Platform Oct 2026 Banco Bradesco, one of Brazil's largest banking and financial conglomerates, deployed Google Cloud's Gemini Enterprise to automate heavy fiscal, accounting, and contractual analysis workflows. Its finance teams use the platform to accelerate proposal reviews, service classification, and budgetary analysis within governed enterprise controls. The deployment cut manual document review times from one hour to five minutes, reduced risk inconsistencies by 60%, and delivered over 10% in financial efficiencies. πŸ”— -
Lloyds & Google Cloud Enterprise Platform Oct 2026 Lloyds Banking Group migrated 15 modelling systems from on-premise infrastructure to Google Cloud's Vertex AI (now Gemini Enterprise), where 300+ data scientists launched 80+ ML use cases and 18+ GenAI systems β€” including an income-verification model that cut mortgage processing from days to seconds. By October 2026, Lloyds scaled Envoy, a governed Gemini-powered engineering pipeline featuring a reusable agent marketplace, "golden path," and regulatory CI/CD guardrails, onboarding nearly 1,000 engineers building production agents for fraud detection and commercial onboarding. πŸ”— πŸ”— -
Commerzbank & Google Integrated Solution Oct 2026 Commerzbank first adopted Google's multimodal Gemini 1.5 Pro to automate financial advisory documentation and analyze lengthy client interactions and financial filings. By October 2026, the bank expanded onto Google Cloud's Gemini Enterprise to automate document quality assurance, cutting manual document review work from 20 hours to 1 hour, and began scaling a multi-agent system for AI-driven quality assurance. πŸ”— πŸ”— -
BBVA The Frame Enterprise Platform Oct 2026 BBVA introduced "The Frame," a group-wide engineering and governance standard for creating, deploying, and monitoring AI agents across its 80M-customer bank on a model-agnostic stack (specific LLM vendors unnamed), alongside a 24/7 multimodal client adviser ("Blue") slated for 2027. Agents run within scoped data permissions and explicit human-oversight boundaries; existing production deployments filter over 50% of Spanish RM app chats, resolve ~90% of voice-pilot calls in Peru and Mexico, cut complaint registration nearly 80%, and lift developer productivity ~28%. πŸ”— -
Barclays & Anthropic Enterprise Platform Oct 2026 Barclays expanded its collaboration with Anthropic to scale Claude across UK retail servicing, Global Markets, and software engineering under human oversight. Its Claude-powered RAG Colleague Knowledge Assistant supports 16,000+ UK employees across 1M+ searches serving 20M+ retail customers, while Global Markets uses Claude to classify, enrich, and route ~120,000 incoming client emails daily. Barclays is also scaling Claude Code for legacy modernization, targeting 50% developer adoption by end-2026. πŸ”— -
Lloyds Credit Policy RAG Integrated Solution Oct 2026 Lloyds Banking Group deployed a production Gemini 2.5 Flash RAG system (live since April 2026) for lending and compliance staff querying quarterly-revised credit policy manuals. Structure-aware indexing parses policy tables into structured rows and enforces section citations on every asserted condition, monitored across document revisions via SME-calibrated provision alignment and a Gemini 2.5 Pro statement-level judge. Serving 600+ monthly queries at 17-second median latency and ~$0.03 per query, its longitudinal evaluation eliminates up to 25.7 points of false recall-regression drift. - πŸ”—
Brex Assistant & MCP Connectors Integrated Solution Oct 2026 Brex built an AI-powered financial assistant on Amazon Bedrock and Claude models to automate expense management, achieving 75% workflow automation and raising compliance rates from 70% to the mid-90s. In October 2026, Brex launched official Model Context Protocol (MCP) connectors (early access) for Claude and ChatGPT, exposing 19 role-governed actions so admins and cardholders can issue, lock, edit, or cancel corporate cards and query spend limits and policies in natural language. πŸ”— πŸ”— -
BNP Paribas & Google Cloud Enterprise Platform Sep 2026 BNP Paribas signed a five-year partnership with Google Cloud to integrate Gemini models and Gemini Enterprise into its multi-cloud architecture (distinct from its Mistral AI deployment). In Corporate & Institutional Banking, Gemini is being integrated into LLM@CIB β€” its internal GenAI assistant serving 65,000+ employees β€” alongside authenticated, least-privilege agents for corporate credit memo drafting, research, sales, trading, and structuring. Gemini also powers Nickel Assist in production, helping 200 payment-account advisers answer inquiries from operating guidelines. πŸ”— -
Citi Stylus Workspaces & Arc Enterprise Platform Sep 2026 Citi scaled Citi Stylus Workspaces β€” integrating Anthropic Claude, OpenAI, and Google Gemini models β€” alongside Arc (launched April 2026), an enterprise platform for building and governing multi-step, agent-to-agent workflows under human oversight. Stylus Workspaces reaches 180,000+ employees across 87+ jurisdictions (>87% adoption, 65M+ interactions), cutting U.S. Services client-onboarding document reviews from ~1 hour to 15 minutes, while Cognition's Devin coding agent remapped 30+ years of legacy code in two days. πŸ”— -
ItaΓΊ Unibanco & Cognition Devin Integrated Solution Sep 2026 Over 75% of ItaΓΊ Unibanco's technology teams deployed Cognition's autonomous coding agent Devin under engineer review to document 300,000+ code repositories, migrate legacy .NET services to Java 6x faster, and automatically remediate roughly 70% of security vulnerabilities, lifting bank-wide engineering throughput by 20%–30%. In the same SΓ£o Paulo disclosure, Brazilian digital bank Nubank used Devin to compress a multi-million-line monolith migration from projected years into weeks at over 20x lower cost. Underlying LLMs are not disclosed. πŸ”— -
Deutsche Bank & Google Cloud (Financial Research Agent) Enterprise Platform Aug 2026 Deutsche Bank served as design partner for Google Cloud's Financial Research Agent, the core agent in Gemini Enterprise for Financial Services (preview, also adopted at launch by CME Group; distinct from DB Research's DB Lumina). Built with 50+ finance-specific skills, Agent2Agent (A2A) APIs, and 13 MCP/data connectors, the agent synthesizes market, company, and internal data into client briefings and credit analyses with confidence scores, data snapshots, and source citations, rolling out first to Corporate Bank German MidCorp teams. No time-savings metrics disclosed. πŸ”— -
BMO Aura Integrated Solution Aug 2026 BMO deployed Aura, an internal GenAI commercial credit underwriting assistant and agent (no model or vendor named), into selected Commercial Banking underwriting workflows. Aura pulls and synthesizes borrower data across approved internal systems to generate structured draft commentary for commercial credit submissions, while underwriters validate outputs and retain final credit accountability. Alongside 95% daily AI usage across commercial teams, Aura anchors BMO's target of ~50% faster credit underwriting by FY2028. πŸ”— -
Westpac Agentic Data Ecosystem Enterprise Platform Aug 2026 Westpac unveiled an agentic data ecosystem and three banker AI platforms β€” Start Your Day Informed, Create a Better Day, and Sleep Tight at Night β€” integrating its 285-source intelligence layer with Snowflake's CoWork and CoCo assistants (distinct from its AWS lending agents). The platforms synthesize daily market signals, test commercial underwriting assumptions, and monitor credit and risk controls overnight, speeding AI model builds 5x, customer insight-to-engagement 6x, and campaign creation from months to days; underlying LLMs were unnamed. πŸ”— -
DBS Agentic AI for Corporate Bankers Enterprise Platform Aug 2026 DBS scaled an agentic AI solution that streamlines credit memo preparation for corporate bankers serving large and mid-sized clients, expanding from a 150-participant pilot to around 1,500 relationship managers and credit risk managers globally. Specialised agents tackle more than 70 tasks; with credit memos and related credit activities taking up to 40% of relationship managers' time, DBS targets a reduction of at least 30%. No model or vendor named. πŸ”— -
Westpac Lending AI Agents Integrated Solution Aug 2026 Westpac put five specialist AI agents built on Amazon Bedrock AgentCore into production home loan and credit card assessment. An account scrutiny agent classifies transactions from manual bank statements; others extract payslip data and verify it against lending policy, with bankers retaining oversight. They process 32,000+ payslips and 1.5M+ transactions weekly, saving 12,500 banker hours in their first month β€” over 150,000 annualised. πŸ”— -
CIBC AI 2.0 (CAI 2.0) Enterprise Platform Jul 2026 CIBC unveiled CIBC AI 2.0 (CAI 2.0, in pilot), billed as Canadian banking's first enterprise-wide agentic AI workspace, built in-house on a custom agentic harness (no underlying LLMs named). Staff connect internal data and tools to delegate multi-step research, pitchbook drafting, side-by-side credit agreement comparison, and compliance anomaly reporting for human review, compressing half-day financial analyses into minutes. CAI 2.0 builds on CIBC's chat-based CAI platform, used by 50,000+ employees globally (20,000 daily active users). πŸ”— -
Scotiabank Scotia Intelligence Agents Enterprise Platform Jul 2026 Scotiabank rolled out governed internal knowledge agents β€” Delivery Navigator, Travel and Expense Knowledge Agent, and Procurement Knowledge Agent β€” on its Scotia Intelligence enterprise data and AI platform. Grounded in approved internal repositories and governed via the AI Consortium's Agentic Control Plane, the agents guide staff through delivery standards, expense eligibility, and procurement workflows. Scotia Intelligence reaches 71,000+ employees and has logged 14 million actions since March 2026; no LLM vendor was named. πŸ”— -
Monzo Fincrime Agent Integrated Solution Jul 2026 Monzo built Fincrime Agent, a reusable Python framework pairing multimodal chain-of-thought reasoning LLMs (via Monzo's multi-vendor AI gateway; specific models unnamed) with supervised ML across four production financial-crime agents. On high-risk transfers, Fraud Prevent evaluates customer questionnaires, screenshots, invoices, and voice notes to feed structured semantic features into a meta-stacking ML ensemble, escalating uncertain cases to human investigators. Running live in the payment path in under 10 seconds, it boosted prevented authorized fraud (Β£ value and count) by >20%. πŸ”— -
Bank of America EricaAssist Enterprise Platform Jul 2026 Bank of America enhanced EricaAssist, its internal human-assisted AI agent for consumer servicing (distinct from commercial-banking AskGPS), with generative AI capabilities (no external LLM vendor named). Working alongside employees during live calls, EricaAssist summarizes why a client is calling, pulls relationship context, and recommends role-specific next steps while keeping representatives responsible for client guidance. Used by 18,000+ customer service representatives, it surfaces guidance in under three seconds and cuts average call duration by nearly one minute. πŸ”— -
Commonwealth Bank AI Orchestration Agent Enterprise Platform Jul 2026 Commonwealth Bank of Australia and Microsoft co-developed a central retail-support AI orchestration agent built on Microsoft Copilot Studio, Dynamics 365, and Microsoft Foundry (distinct from AWS-based CommBiz Gen AI). The agent interprets inbound customer intent across digital messaging, routing routine queries to conversational AI and escalating sensitive cases to human specialists with live summaries and suggested responses. Roughly 84.6% of May 2026 self-service messaging interactions resolved end-to-end in-channel, with bank-wide expansion underway. πŸ”— -
ACRM (Ant International) Enterprise Platform Jul 2026 Ant International deployed ACRM in production to automate Out-of-Time credit risk model refreshing when borrower score distributions drift. A central Qwen-32B-based Orchestrator coordinates six specialist agents across upstream cohort selection, observation windowing, and feature screening using DPO alignment on expert trajectories, RAG over historical drift playbooks, and hard PSI/KS compliance guardrails before human review. Live for three months across six credit scenarios, ACRM cut average model refresh cycles from weeks to 1.1 days and reduced iteration rounds by 65%. - πŸ”—
DBS Joy Agentic AI Enterprise Platform Jul 2026 DBS made DBS Joy, its corporate virtual assistant (launched 2018, given GenAI capabilities in 2025), fully agentic for 350,000 corporate and SME clients in Singapore, letting them retrieve and analyse account and transaction data within a single conversation rather than searching the platform manually. No model or vendor named. In H1 2026 chats rose 22% and active users 61%, cutting calls/emails to customer service by 7%. Another 100,000 Hong Kong users follow in September, with DBS Joy and the retail DBS digibot together expected to handle 1M+ chats monthly across 10M customers. πŸ”— -
Nubank Support AI Agents Enterprise Platform Jun 2026 Nubank published its evaluation-driven framework (KDD '26) for building and validating LLM customer-support agents across its 100M+-user base, deployed in production across card delivery, debt management, credit-limit support, card management, and product explanation (specific LLMs unnamed). The pipeline pairs structured context engineering and human-in-the-loop prompt iteration with GEPA-optimized LLM-as-a-judge offline simulations. Large-scale A/B testing on card delivery lifted AI transactional NPS by 37 percentage points and self-service rates by 29 percentage points. - πŸ”—
YouZhi-LLM (PSBC & Huawei) Fine-Tuned Model Jun 2026 Postal Savings Bank of China and Huawei built YouZhi-LLM (7B and 14B), high-concurrency financial LLMs adapted from OpenPangu-7B and Qwen2.5 bases and served via vLLM-Ascend on Huawei Ascend NPUs. A layer-adaptive GQA-to-MLA attention conversion cuts KV-cache memory by 72% and perplexity loss up to 35% versus uniform conversion, followed by distillation, financial SFT, and compliance refusal tuning. YouZhi-7B lifts financial benchmark scores by 12.3% at 2.69x max concurrency, scoring 96.0–100% across PSBC mobile banking intent, loan, and wealth tasks. - πŸ”—
HSBC & Google Cloud Enterprise Platform Jun 2026 HSBC and Google Cloud announced a multi-year AI partnership using Gemini models, the Gemini Enterprise Agent Platform and Google DeepMind. Applications include financial-crime monitoring across the near one billion transactions HSBC screens monthly (intervening twice as fast), a relationship-manager decision assistant cutting prep from hours to minutes, and hyper-personalised wealth advisory. The deal targets more than 200 new AI use cases over two years. πŸ”— -
Santander AI-First Enterprise Platform Jun 2026 Santander extended AI access to all 185,000 employees under its "AI-first" strategy, using a multi-provider stack (Microsoft Copilot, OpenAI ChatGPT, Anthropic Claude, Google Gemini, G42). Openbank's models process ~100,000 AML alerts/year (hours to minutes); UK self-service handles ~240,000 calls/year, freeing ~45,000 service-team hours. AI delivered €35M in Q1 2026, on track to exceed €200M in 2026, targeting €1B over 2026-2028. πŸ”— -
BNP Paribas & Mistral AI Enterprise Platform May 2026 BNP Paribas signed a three-year renewal with Mistral AI in May 2026 (expanding their 2023–2024 Group-wide LLM agreement) covering on-premises Mistral models, software, and joint engineering co-development under European data-residency controls. Live deployments span Global Markets, Commercial & Personal Banking β€” including Hello bank!'s HelloΓ―Z virtual assistant serving 1M+ clients β€” customer support, IT, and proactive cybersecurity vulnerability detection. Specific Mistral model versions and bank-wide efficiency metrics were not disclosed. πŸ”— πŸ”— -
Revolut AIR Integrated Solution Apr 2026 Revolut launched AIR (AI by Revolut), an in-app AI assistant that lets customers manage money through natural conversation β€” answering questions on spending, budgeting and investing, pausing recurring payments, managing card controls, checking live exchange rates, and helping with trip budgets and in-app eSIM purchases. Sensitive actions require biometric approval, and Revolut says personal data is never stored by third-party AI partners or used for model training. Rolling out from April 2026 to its ~13 million UK customers first, with other markets to follow. πŸ”— -
Starling Assistant Integrated Solution Mar 2026 Starling Bank launched Starling Assistant, billed as the UK's first agentic AI financial assistant, built on Google Gemini running inside Starling's own Google Cloud environment. Rolled out to personal current account holders from 20 March 2026 (business and joint accounts to follow), it uses voice and natural language to set up savings transfers, organise bill payments, create budget spaces, analyse spending and guide accessibility support. πŸ”— -
NatWest & OpenAI Enterprise Platform Feb 2026 NatWest became the first UK bank to sign a strategic collaboration with OpenAI, embedding OpenAI models (version unspecified) into its Cora customer chatbot and AskArchie employee assistant. Customers can ask natural-language spending questions and report fraud in real time; around 25,000 customers gain the new agentic financial assistant by end of Q1 2026, with voice-to-voice capability piloting later in 2026. πŸ”— -
HSBC & Harvey AI Enterprise Platform Jan 2026 HSBC partnered with Harvey AI to deploy AI across its global in-house legal function, streamlining workflows in contract analysis, due diligence, compliance, and litigation. πŸ”— -
BNY Mellon - Eliza Enterprise Platform Dec 2025 BNY developed Eliza, an enterprise AI platform originally powered by OpenAI models, enabling 20,000+ employees to build agents across 125+ live use cases in legal, risk, and client management β€” cutting legal contract review time 75% with 99% workforce adoption. In December 2025, BNY partnered with Google Cloud to integrate Gemini Enterprise (plus Gemini 3 and Veo 3) into Eliza for agentic market research. πŸ”— πŸ”— -
HSBC & Mistral AI Enterprise Platform Dec 2025 HSBC partnered with Mistral AI in a multi-year strategic collaboration to deploy self-hosted GenAI models across the bank, powering productivity tools for tailored client communications, financial analysis, multilingual translation, and faster development cycles. πŸ”— -
Lloyds AI Financial Assistant Enterprise Platform Nov 2025 Lloyds Banking Group unveiled what it calls the UK's first large-scale, multi-feature agentic AI financial assistant, built on its proprietary Generative AI and agentic framework combining curated bank data with LLMs (no external model or vendor named). Launching in the mobile app in early 2026 for 21M+ customers, it offers 24/7 financial coaching with memory, expanding to mortgages, car finance and protection. πŸ”— -
Macquarie Bank & Gemini Enterprise Enterprise Platform Oct 2025 Macquarie Bank is rolling out Google Cloud's Gemini Enterprise to all employees β€” not just technical teams β€” for personal agents (document summarization, research, content drafting) and enterprise agents spanning system-design documentation, code generation, conversation summaries, and customer support. 99% of employees completed generative-AI training, with a target of every employee integrating AI into daily workflows within six months. πŸ”— -
Bank of America's AskGPS Enterprise Platform Sep 2025 BofA built AskGPS (Ask Global Payments Solutions), an in-house GenAI assistant trained on 3,200+ internal documents to help employees serving 40,000+ business clients answer complex inquiries. It reduces response times from hours to seconds, saving tens of thousands of employee hours annually. πŸ”— -
Wells Fargo & Google Cloud AI Enterprise Platform Aug 2025 Wells Fargo uses Google Agentspace (powered by Gemini models) to deploy agentic AI across its workforce, enabling employees to find and synthesize information faster, automate tasks and workflows, and increase organizational agility. πŸ”— -
U.S Bank & Amazon Integrated Solution Aug 2025 U.S. Bank deployed a real-time agent-assist system using Amazon Q in Connect together with Amazon Bedrock and Claude Haiku model to transcribe calls, detect customer intent, and surface relevant knowledge-base recommendations during voice interactions. πŸ”— -
PayU Integrated Solution Jul 2025 PayU implemented a secure enterprise-AI assistant using Amazon Bedrock, deploying multi-agent workflows (via RAG and text-to-SQL) to enable employees to query business metrics, HR policies and operations dataβ€”all within a private VPC to satisfy regulatory data-residency and compliance needs. πŸ”— -
Intuit TurboTax Integrated Solution Jul 2025 Intuit built Intuit Assist, an LLM-powered AI assistant within TurboTax that helps millions of users understand their tax situations, deductions, and refunds using a hybrid Claude–GPT system with RAG, fine-tuning, and strict regulatory compliance πŸ”— -
Nubank Integrated Solution May 2025 Nubank built an AI β€œprivate banker” using GPT-4 and agentic frameworks like LangChain and LangGraph to automate customer support and money-transfer workflows. The system handles millions of user queries and transactions autonomouslyβ€”reducing transfer times and improving accuracyβ€”through coordinated LLM agents and a production-grade LLMOps platform. πŸ”— -
Apoidea Group Fine-Tuned Model May 2025 Apoidea Group, in collaboration with Amazon SageMaker HyperPod and using the LLaMA‑Factory framework, built a solution called SuperAcc that is based on a fine-tuned multimodal model Qwen2‑VL‑7B‑Instruct to enhance visual information extraction from banking documents, boosting table-structure recognition accuracy substantially and reducing processing times from hours to minutes. πŸ”— -
Capital One Chat Concierge Fine-Tuned Model Feb 2025 Chat Concierge, built on a fine-tuned Llama LLM, helps car buyers compare vehicles, explore financing, estimate trade-in values, and schedule test drives. Fine-tuned on Capital One's proprietary data, it also utilized an agentic approach to understand preferences, tailor recommendations, ensure policy compliance, and engage in human-like interactions. πŸ”— -
Deutsche Bank & Google Enterprise Platform Feb 2025 Deutsche Bank leverages Google Cloud’s Vertex AI and Gemini LLMs to streamline banking operation. It involves automating document processing for regulatory compliance, enhancing customer support with AI-powered assistants, and accelerating software development for financial services. πŸ”— -
CommBiz Gen AI Integrated Solution Jan 2025 Together with AWS, the Commonwealth Bank of Australia (CBA) rolled out a Gen-AI powered messaging service to assist tens of thousands of business customers with inquiries, facilitating quicker payments and efficient transactions. They leveraged Amazon Bedrock Knowledge Bases, Claude 3 and Cohere LLMs, and Amazon OpenSearch as the vector database. πŸ”— -
North for Banking Enterprise Platform Jan 2025 RBC and Cohere co-developed and securely deployed an enterprise generative AI (genAI) solution optimized for financial services, building upon Cohere's proprietary foundation models πŸ”— -
Amazon Finance Enterprise Platform Dec 2024 Amazon Finance Automation built a generative-AI Q&A chat assistant on Amazon Bedrock to empower analysts to quickly answer customer queries by retrieving and synthesising policy knowledge, accelerating response times and reducing manual research πŸ”— -
Banestas & Google Enterprise Platform Dec 2024 Banestes, a Brazilian bank, used Gemini in Google Workspace to streamline work dynamics, such as accelerating credit analysis by simplifying balance sheet reviews and boosting productivity in marketing and legal departments. πŸ”— -
OCBC ChatGPT Integrated Solution Nov 2024 OCBC Bank became the first Singapore bank to deploy a generative AI chatbot, OCBC ChatGPT, to all 30,000 employees across 19 countries, aiming to assist with tasks such as writing, research, and ideation. Developed with Microsoft Azure and following a successful six-month trial, the chatbot operates securely on the bank's private cloud to ensure data security. πŸ”— -
BBVA & OpenAI Enterprise Platform Nov 2024 Global financial institution BBVA signed an agreement with OpenAI for 3,000 ChatGPT Enterprises licenses, leading to increased productivity and creativity. Staff across various departments have developed over 2.9k specialized GPTs that boost efficiency, spark creativity, and share expert knowledge across their organization of 125,000 e.g., tasks like translating risk-specific terminology and drafting responses to client inquiries. πŸ”— -
PennyMac & Google Enterprise Platform Oct 2024 Pennymac Financial Serives, a national home loan lender and servicer, integrates Google's Gemini LLMs across various departments to enhance efficiency and reduce costs while ensuring security and compliance. HR uses it for job descriptions and policy drafting, while the underwriting team uses Gemini to analyze proprietary data, improving regulatory understanding and best practices. πŸ”— -
Belfius & Mistral AI Enterprise Platform Sep 2024 Belfius, a leading Northern European digital bank, deployed Mistral Large to power its customer-facing chatbot in Dutch, French, and English, migrating from a scripted bot platform to an AI-powered solution that delivers personalized banking experiences and reduces manual operator workload. πŸ”— -
DBS GPT Enterprise Platform Jul 2024 DBS Bank developed an in-house GenAI co-pilot, the CSO Assistant, to support 500 customer service officers in Singapore. It delivers real-time call transcription, solutioning, and auto-filling of service requestsβ€”achieving near 100% accuracy and reducing call handling time by up to 20%. πŸ”— -
Bitext Mistral-7b-Banking Fine-Tuned Model Jun 2024 Fine-tuned version of the Mistral-7B-Instruct-v0.2, specifically tailored for the banking domain. It is optimized to answer questions and assist users with various banking transactions πŸ”— πŸ”—
Scotiabank & Google Integrated Solution May 2024 Scotiabank uses LLMs via Google Cloud to enhance customer service and automate document processing. It features LLM-powered chatbot summarization for seamless handoffs and faster resolutions, plus enhanced Q&A and search for streamlined employee workflows. πŸ”— -
ING Bank Integrated Solution Feb 2024 ING implemented a generative AI-powered customer-facing chatbot to enhance customer service efficiency. This chatbot utilizes a multi-step process that retrieves information from data stores, ranks potential answers by relevance, and applies strict guardrails to ensure accurate and appropriate responses. πŸ”— -
Citi & GitHub Copilot Integrated Solution Feb 2024 Citi has focused on improving developer productivity by embracing LLM-based coding assistants by rolling out GitHub Copilot (powered by OpenAI’s Codex) to all of its 40,000 software developers enterprise-wide​. It serves as an AI pair-programmer, suggesting code snippets, functions, or fixes inside developers’ code editors. πŸ”— -
Akbank & Azure OpenAI Integrated Solution Jan 2024 Akbank, one of TΓΌrkiye's largest banks, uses Azure OpenAI Service to power an AI chatbot that automates customer support, improving accuracy to 90% and cutting response times by three minutes per interaction. This enhances efficiency and allows agents to focus on proactive service. πŸ”— -
Ally Financial - Ally.ai Integrated Solution Dec 2023 Ally Financial, the largest digital-only bank in the US and a leading auto lender, launched Azure OpenAI LLM-powered Ally.ai to more than 700 customer care associates in summarizing conversations between them and Ally customers. This automation of post-call documentation for customer service associates is done through LLM summarization of customer call transcripts. πŸ”— -
Westpac & KAI-GPT Fine-Tuned Model Jun 2023 Australian lender Westpac is using KAI-GPT, a banking industry-specific LLM, to help bankers locate, interpret and understand information from policies, regulatory filings, procedures, and complex financial products. KAI-GPT is based on Pythia-Chat-Base-7B, fine-tuned on banking-related dataset comprising 24k question-answer pairs from Common Crawl, 18k questions from Kasisto's own conversational data, and 245 million words from 44k banking-related documents πŸ”— πŸ”—
XuanYuan 2.0 Fine-Tuned Model May 2023 Chat model (built upon the BLOOM-176B architecture) trained by combining general-domain with domain-specific knowledge and integrating the stages of pre-training and fine-tuning, It is capable of providing accurate and contextually appropriate responses in the Chinese financial domain. - πŸ”—
BBT-FinT5 Fine-Tuned Model Feb 2023 Chinese financial pre-training language model (1B parameters) based on the T5 model, and pre-trained on the 300Gb financial corpus called FinCorpus - πŸ”—

Wealth & Asset Management

Name Type Date Description Site Paper
Pictet & Anthropic Enterprise Platform Oct 2026 Swiss private bank and asset manager Pictet (~CHF 800B AUM) rolled out Anthropic's Claude Code and Claude Cowork to ~700 employees across engineering, product, compliance, and investment teams behind a custom API gateway enforcing EU/Swiss data residency and zero model-training retention. Connected via MCP to GitHub, Jira, Figma, and internal wikis, staff build internal tools (such as PicAccess for access provisioning) and run regulatory reviews, cutting a 50+-directive compliance gap analysis from two weeks to hours. πŸ”— -
Schwab Advisor Center & Claude Integrated Solution Oct 2026 Charles Schwab integrated Anthropic's Claude for Financial Advisors into Schwab Advisor Center (announced September 2026, live October 2026), becoming the first RIA custodian to connect advisor-specific Claude MCP workflows directly to custody data across 17,000+ independent RIAs. Advisors query client accounts, positions, and history alongside planning and CRM tools to assemble pre-meeting briefs, tax-transition checks, and follow-ups under explicit advisor approval and admin audit logs, alongside Wealth.com's Ester estate-review tool. πŸ”— πŸ”— -
S&P Global iLEVEL Agent & MCP Enterprise Platform Oct 2026 S&P Global embedded a native iLEVEL Agent and Model Context Protocol (MCP) server into its iLEVEL private markets portfolio-monitoring platform serving 700+ general partners and institutional allocators (distinct from Capital IQ Pro ChatIQ). The agent automates unstructured financial document ingestion, natural-language portfolio search, integrated valuations, and private-credit covenant tracking against credit agreements, while the MCP server exposes governed iLEVEL data inside Anthropic Claude and Microsoft Copilot under user-level permissions. No usage metrics disclosed. πŸ”— -
Balyasny Asset Management BAMAgent Enterprise Platform Oct 2026 Multi-strategy investment firm Balyasny Asset Management (~$38B AUM) integrated Google's multimodal Gemini models and Anthropic's Claude Fable 5 into BAMAgent, its proprietary multi-model research platform isolated inside VPC Service Controls with least-privilege permissions and mandatory human sign-off. Connected to 80+ internal databases across 200+ investment teams, agents synthesize filings, central-bank releases, and charts, while ~30-minute merger-arbitrage runs cut initial deal-package prep from 3–5 days to under one day (89.4% internal benchmark score). πŸ”— πŸ”— -
Morningstar Direct AI Enterprise Platform Oct 2026 Morningstar launched Morningstar Direct AI, an agentic interface for its flagship institutional platform (distinct from Direct Advisory Suite's assistant and Mo), introducing three purpose-built agents grounded in Morningstar data, ratings, and analyst frameworks across public and private securities. The Product Development Agent analyzes category flows for fund launches, the Distribution Agent identifies model-portfolio gaps, and the Manager Research Agent vets strategy behavior across market regimes. Live globally alongside Claude, Copilot, and ChatGPT integrations; underlying LLMs are unnamed. πŸ”— -
Deutsche Bank Private Bank Agentic KYC Integrated Solution Sep 2026 Deutsche Bank Private Bank launched an agentic AI Source of Wealth (SoW) solution integrated directly into its digital Know Your Customer (KYC) platform (no LLM vendor named). The agent automates research, extraction, and SoW assessment preparation across client documentation and approved external sources, flagging gaps or inconsistencies while human staff retain review and final accountability. Live across Singapore, Hong Kong, and Dubai booking workflows, it supports a projected ~30% increase in 2026 Emerging Markets client onboarding. πŸ”— -
Millennium Digital Risk Analyst Integrated Solution Aug 2026 Multi-strategy investment firm Millennium (340+ investment teams) and Anthropic are co-developing a Digital Risk Analyst inside Millennium's AI lab on Claude Enterprise, building on firmwide Claude and Claude Code adoption across trading desks and engineering. Connected to proprietary portfolio data and persistent memory, the agent investigates daily cross-asset risk-position shifts, tests candidate actions in sandboxed environments, logs full reasoning traces, and routes outputs to human risk managers for review. No quantitative performance metrics were disclosed. πŸ”— -
OCBC HELIOS Enterprise Platform Jul 2026 OCBC launched HELIOS (Holistic wEalth Lifecycle Insights & Ongoing Surveillance), an agentic AI due-diligence and risk-surveillance platform for private banking (distinct from the client-facing OCBC WoW app; no LLM vendor named). HELIOS front-loads KYC before a relationship manager engages a prospect β€” automating intelligence gathering, relationship-network mapping, credit-risk profiling, and compliance-originated lead generation β€” under RM and compliance accountability. Rolling out across Bank of Singapore in Singapore, Hong Kong, and Dubai, it halves account opening to 15 business days. πŸ”— -
LPL Financial Latitude & Cyan Enterprise Platform Jul 2026 LPL Financial unveiled LPL Latitude, a unified advisor platform anchored by Cyan, a proprietary generative and agentic AI assistant embedded directly inside advisor workflows across its ~$2B three-year technology overhaul. Cyan provides conversational workflow support, executes agentic routine account maintenance, drafts financial-planning summaries, and generates practice-growth recommendations from performance data while keeping human advisors as the sole source of investor advice. Neither underlying foundation models nor initial advisor adoption metrics were disclosed. πŸ”— -
OCBC WoW Integrated Solution Jul 2026 OCBC launched WoW (Whole-of-Wealth), billed as the region's first AI-native banking app, using two GenAI avatars (Wendy and Wayne) to deliver hyper-personalised, 24/7 wealth adviceβ€”portfolio analysis, market news, and investment ideasβ€”via voice or text to Premier Private Client customers. A related agentic-AI source-of-wealth (SOW) tool cut SOW report preparation from 10 days to one hour. πŸ”— -
Morgan Stanley at Work (MCP) Enterprise Platform Jun 2026 Morgan Stanley at Work is opening its Shareworks and Equity Edge stock-plan administration platforms to external AI agents via the Model Context Protocol (MCP), letting clients' agents pull equity-compensation data and insights directly in an agent-to-system manner. The platforms hold roughly $7.35 trillion in assets across about 3,400 administration clients, including nearly half the S&P 500; full rollout is targeted for 2027. πŸ”— -
Charles Schwab Portfolio Insights Integrated Solution May 2026 Schwab launched Portfolio Insights, its first generative-AI capability for retail investors (no LLM vendor named). It produces narrative daily performance summaries for the up-to-five holdings with the largest moves, recaps market and industry news on clients' S&P 500 positions, and adds Schwab Center for Financial Research commentary. Rollout to eligible self-directed U.S. clients began May 2026. πŸ”— -
Broadridge Agentic Operations Enterprise Platform May 2026 Broadridge deployed production agentic AI capabilities across wealth management and capital markets middle- and back-office operations, anchored by its normalized Financial Services Data Ontology ($15T+ daily trading activity; specific LLMs unnamed). Under human-supervised, auditable controls, agents autonomously analyze, prioritize, and resolve trade fails and breaks, account opening and maintenance, real-time valuation exceptions, customer inquiries, and email workflows (with DeepSee). Shaped across 40+ managed-services clients processing millions of monthly transactions, it delivers up to 30% Day-1 operational cost reductions. πŸ”— -
Citi Sky Enterprise Platform Apr 2026 Citi Wealth unveiled Citi Sky, an always-on conversational AI "team member" built on Google Cloud's Gemini Enterprise Agent platform, Google DeepMind's real-time avatar technology, and the Gemini Live API. It delivers Chief Investment Office market insights and financial guidance (e.g., CD-maturity alerts) via audio/video in English and Spanish, with a phased rollout to U.S. Citigold clients from summer 2026. πŸ”— -
Vanguard Expert Insights Integrated Solution Apr 2026 Vanguard launched Expert Insights, a generative-AI portfolio-analysis tool (no model or vendor disclosed) that lets financial advisors generate instant, methodology-aligned, client-ready guidance and pressure-test portfolios. Currently piloting with select advisors, it will be embedded on-demand in Vanguard's Portfolio Analytics Tool later in 2026. Vanguard's annual advisor portfolio-analysis engagements have roughly quadrupled over the past six years. πŸ”— -
Morningstar Advisory AI Assistant Integrated Solution Mar 2026 Morningstar embedded a new AI assistant in its Direct Advisory Suite (distinct from 2023's "Mo" chatbot), grounded in Morningstar's independent data and research. Advisors use natural-language requests to surface meaningful rating changes, prepare meeting briefs, and turn client statements into actionable proposals; it is also reachable through other AI tools via Morningstar's MCP connections. Initial U.S. rollout, with broader U.S./Canada availability through 2026. πŸ”— -
Franklin Templeton & Wand AI Enterprise Platform Nov 2025 Franklin Templeton (~$1.66 trillion AUM) formed a multi-year partnership with Wand AI to deploy agentic AI in live production across investment research, operations, and transformation initiatives, moving beyond pilots to enterprise-wide automation. Underlying LLMs were not disclosed; wide-scale rollout is targeted for 2026. πŸ”— -
Ardian GAIA Enterprise Platform Sep 2025 Ardian developed GAIA, an internal GenAI platform (powered by Azure OpenAI and Mistral AI) that transforms how investment teams analyze and extract insights from complex documents, enabling 500+ weekly active users to analyze due diligence documents, financial reports, and legal filings. The platform has processed 18,000+ documents and generated 280,000+ AI-powered queries, achieving 2X growth in documents processed while accelerating deal execution and freeing teams to focus on strategic analysis. πŸ”— -
Fidelity International Freya Integrated Solution Sep 2025 Fidelity International launched Freya, a conversational LLM-based assistant on its Personal Investing platform (500,000 customers), answering natural-language queries with reasoning-based responses and built-in guardrails that stop short of regulated advice. The underlying model/vendor is unnamed. A phased rollout will expand Freya to Workplace Investing and Fidelity Adviser Solutions. πŸ”— -
AlphaAgents (BlackRock) - Aug 2025 BlackRock developed a role-based multi-agent system powered by GPT-4o to enhance wealth management in equity portfolio constructions, enabling LLM agents to collaborate autonomously on equity research, stock selection, and portfolio optimization. - πŸ”—
Nippon India Mutual Fund Integrated Solution Jul 2025 Nippon Life India Asset Management Ltd. (Nippon India Mutual Fund) used Amazon Bedrock Knowledge Bases to deploy a RAG AI assistant, leveraging Claude 3 Sonnet on Amazon Bedrock to reformulate queries and rerank responses. The solution increased answer accuracy by over 95% and reduced hallucinations by >90%, cutting report generation time from 2 days to about 10 minutes. πŸ”— -
Alpian & Google Cloud Integrated Solution Jun 2025 Alpian, Switzerland's first fully cloud-native private bank, built a Gemini-powered platform engineering tool on Google Cloud (Vertex AI, BigQuery) that lets developers deploy services and interact with infrastructure through a conversational interface within a deliberately bounded scope that preserves security and compliance guardrails. It cut deployment times from days to hours, sped up feature deployment by 25%, and automates compliance workflows aligned with Swiss FINMA standards. πŸ”— -
BlackRock Aladdin Integrated Solution Jun 2025 BlackRock’s Aladdin Copilot uses generative AI to let investors query portfolio and risk data in plain language, surfacing precise, contextual insights. It enhances productivity, decision-making, and personalization across the Aladdin platform while meeting strict compliance and accuracy standards. πŸ”— -
FinE5 Fine-Tuned Model Feb 2025 Fin-E5, the finance-adapted embedding model in FinMTEB, is built on the E5 model and trained using a persona-based data synthesis method to enhance performance across financial embedding tasks. - πŸ”—
PIMCO & Azure Enterprise Platform Feb 2025 PIMCO developed ChatGWM, an AI-powered search tool built on Azure AI, to enhance client service by streamlining information retrieval for its associates. ChatGWM utilizes retrieval-augmented generation (RAG) to search across approved structured and unstructured data sources, and then process them with Azure OpenAI LLM to provide accurate and up-to-date information swiftly. πŸ”— -
Aditya Birla Capital Integrated Solution Jan 2025 Financial services provider Aditya Birla Capital implemented SimpliFi, a generative AI chatbot built on Azure OpenAI Service, to enhance customer engagement and streamline financial services. SimpliFi assists users in navigating financial solutions independently, aligning with the preferences of their 25 to 35-year-old demographic. πŸ”— -
Touchstone-GPT Fine-Tuned Model Nov 2024 Open-source financial LLM trained through continual pre-training and financial instruction tuning, which demonstrates strong performance on the financial bilingual (English and Chinese) Golden Touchstone benchmark. πŸ”— πŸ”—
Banca Investis Integrated Solution Nov 2024 Banca Investis launched NIWA, a GenAI-powered investment advisory platform designed to enhance customer engagement through hyper-personalized financial services. Serving as a "digital junior banker," NIWA analyzes over 500 pieces of information dailyβ€”including clients' financial assets, preferences, and market researchβ€”to provide tailored financial insights and real-time responses to investment-related inquiries. πŸ”— -
FinTral Fine-Tuned Model Aug 2024 Suite of multimodal LLMs built upon the Mistral-7b model and tailored for financial analysis. FinTral integrates textual, numerical, tabular, and image data, and is pretrained on a 20 billion token, high quality dataset - πŸ”—
Goldman Sachs & Meta Enterprise Platform Aug 2024 Goldman Sachs introduced the GS AI Platform, a GenAI tool designed to enhance employee productivity across various use cases. Built on Meta's LLaMa models, the assistant aids in tasks such as summarizing and proofreading emails, extracting information from documents, as well as translating code between programming languages. πŸ”— -
Nomura & Meta Enterprise Platform Aug 2024 Leading Japanese financial institution Nomura uses Meta's LLaMa models on Amazon Bedrock to democratize generative AI by driving faster innovation, transparency, bias guardrails, and robust performance across text summarization, code generation, log analysis, and document processing. πŸ”— -
JPMorgan Chase IndexGPT Enterprise Platform Jul 2024 JPMorgan Chase launched a generative AI-based tool (via AWS Bedrock) called IndexGPT, designed to serve as a 'research analyst' for over 50,000 employees, aiding in various tasks that enhance productivity and decision-making within the firm. It is able to generate and refine written documents, provide creative solutions and summarize extensive documents. πŸ”— -
IDEA-FinQA Fine-Tuned Model Jun 2024 Financial question-answering system based on Qwen1.5-14B-Chat, utilizing real-time knowledge injection and supporting various data collection and querying methodologies, and comprises three main modules: the data collector, the data querying module, and LLM-based agents tasked with specific functions. πŸ”— πŸ”—
BCI & Azure Enterprise Platform Mar 2024 British Columbia Investment Management Corporation (BCI) integrated Microsoft 365 Copilot and Azure OpenAI Service to enhance productivity and streamline operations. This implementation has led to a 10%-20% productivity boost for 84% of initial users, saving over 2,300 person-hours through automation. Notably, the time required to write internal audit reports decreased by 30%. πŸ”— -
RAVEN Fine-Tuned Model Jan 2024 Fine-tuned LLaMA-2 13B Chat model designed to enhance financial data analysis by integrating external tools. Used supervised fine-tuning with parameter-efficient techniques, utilizing a diverse set of financial question-answering datasets, including TAT-QA, Financial PhraseBank, WikiSQL, and OTT-QA - πŸ”—
InvestLM Fine-Tuned Model Sep 2023 Financial domain LLM tuned on LLaMA-65B, using a carefully curated instruction dataset related to financial investment. The small yet diverse instruction dataset covers a wide range of financial related topics, from Chartered Financial Analyst (CFA) exam questions to SEC filings to Stackexchange quantitative finance discussions. πŸ”— πŸ”—
CFGPT Fine-Tuned Model Sep 2023 Financial LLM based on InternLM-7B that is designed to handle financial texts effectively. It was pre-trained on 584 million documents (141 billion tokens) from Chinese financial sources like announcements, research reports, social media content, and financial news, and then fine-tuned on 1.5 million instruction pairs (1.5 billion tokens) tailored for specific tasks of financial analysis and decision-making. πŸ”— πŸ”—
FinMA Fine-Tuned Model Jun 2023 Comprehensive framework that introduces FinMA (Financial Multi-task Assistant), an open-source financial LLM fine-tuned (7B and 30B versions) from LLaMA using a diverse, multi-task instruction dataset of 136,000 samples. The dataset encompasses various financial tasks, document types, and data modalities. πŸ”— πŸ”—
Fin-Llama Fine-Tuned Model Jun 2023 Specialized version of LLaMA 33B, fine-tuned (with QLoRA and 4-bit quantization) for financial applications using a 16.9k instruction dataset. πŸ”— -
Morningstar - Mo chatbot Integrated Solution May 2023 Morningstar introduced Mo, an AI chatbot powered by the Morningstar Intelligence Engine, which combines Morningstar's extensive investment research library with Microsoft's Azure OpenAI Service. Mo is designed to provide investors and financial professionals with concise, conversational insights by processing natural language queries and summarizing relevant information from over 750,000 investment options. πŸ”— -
Cornucopia-LLaMA-Fin-Chinese Fine-Tuned Model Apr 2023 Open-source LLaMA-based model fine-tuned for Chinese financial applications. It uses instruction tuning with Chinese financial Q&A datasets to enhance domain-specific performance. πŸ”— -
Morgan Stanley & OpenAI Pre-trained Mar 2023 Morgan Stanley Wealth Management announced a partnership with OpenAI to develop an internal-facing GPT-powered assistant (AI @ Morgan Stanley Assistant), allowing financial advisors to query the bank’s vast research repository and internal knowledge base in natural language πŸ”— πŸ”—
FLANG-ELECTRA Fine-Tuned Model Oct 2022 Domain specific Financial LANGuage model (FLANG) which uses financial keywords and phrases for better masking, and built on the ELECTRA-base architecture. Note: Considered a smaller LM as it has fewer than 1B params πŸ”— πŸ”—
FinBERT-21 Fine-Tuned Model Jul 2020 FinBERT (BERT for Financial Text Mining) is a domain specific language model pre-trained on large-scale financial corpora, allowing it to capture language knowledge and semantic information from the finance domain. Note: Considered a smaller LM as it has fewer than 1B params - πŸ”—

Capital Markets & Trading

Name Type Date Description Site Paper
IBKR Autonomous Trading Integrated Solution Oct 2026 Interactive Brokers launched IBKR Autonomous Trading, letting eligible clients authorize third-party AI platforms β€” including ChatGPT, Claude, Claude Code, Grok, and Perplexity β€” connected via its public MCP server to place orders without per-trade approval. Agents operate solely inside a ring-fenced, cash-funded AI Partition across 75,000+ global stocks, ETFs, and listed options. Safeguards block penny stocks, short sales, leveraged products, and margin-borrowing options, require 10 prior manually approved $100+ trades, and provide a Client Portal kill switch. πŸ”— -
FinDialogLens (JPMorgan Chase) Fine-Tuned Model Oct 2026 JPMorgan Chase's ML Center of Excellence developed FinDialogLens, a hybrid event-extraction pipeline recovering missed Request-for-Quote (RFQ) trades across interleaved multi-party sales-and-trading chatrooms. Three fine-tuned RoBERTa-base classifiers detect RFQ triggers, prices, and hit/miss cues to segment event windows for a Trade Engine using GPT-4o or fine-tuned 3B–8B open models (Flan-T5-XL, Phi-3-mini, Mistral-7B, Llama-3-8B), paired with a difficulty-aware router. At a 70,000-RFQ/day scale, it reaches 92.1% final-price and 94.3% trade-outcome accuracy, while routing cuts LLM calls by 85%. - πŸ”—
FLINT (Kensho / S&P Global) Integrated Solution Oct 2026 S&P Global's Kensho unit deployed FLINT (Financial LINking Text-to-SQL) in production within its financial data retrieval service, using Claude 4.5 Opus to query normalized databases with opaque integer foreign keys. A lookup agent and Kensho Link disambiguate natural-language mentions across 20M+ entities into WHERE-clause IDs, combined with vector-retrieved expert SQL templates, bridge-table schema pruning, and execution-error reflection. Across 359 production-schema questions averaging 5–6 joins, FLINT hits 68.8% strict accuracy on Financials and 65.2% on Transactions (~85% audited), whereas seven baselines stay below 50%. - πŸ”—
Nasdaq Calypso Agentic AI Enterprise Platform Sep 2026 Nasdaq introduced a governed agentic AI operating environment inside Nasdaq Calypso (used by 270+ institutions and 60,000+ users globally) to automate workflows across capital markets, risk, collateral, and treasury operations. An integrated Model Context Protocol (MCP) orchestration layer lets institutions run Calypso's native agents alongside their own proprietary AI infrastructure across core trade-lifecycle, market, and reference data. Guardrails enforce strict sandboxing, operational boundaries, live human oversight, and zero external data retention; underlying LLMs and rollout counts were not disclosed. πŸ”— -
Bloomberg Enterprise MCP Enterprise Platform Sep 2026 Bloomberg launched Enterprise MCP as an AI access layer for Data License Plus (DL+), connecting clients' internal LLM applications and AI agents to licensed datasets via Model Context Protocol (distinct from Terminal-based ASKB and BloombergGPT). The service combines natural-language semantic field search, entity resolution, and Automated Entitlements checks against Bloomberg's Order Management and Data Distribution Systems with reusable workflow Skills for price-deviation trade review, outlier detection, and sanctions screening across 50,000+ data fields and 100M+ securities. πŸ”— -
Robinhood Agentic Trading Integrated Solution Sep 2026 Robinhood launched Agentic Trading via MCP servers in May 2026 and expanded at HOOD Summit in September 2026 with native in-app Robinhood Agents and third-party Agent Apps. Users select foundation models β€” including OpenAI's GPT-Luna β€” and connect external data apps such as Nasdaq and Unusual Whales to research portfolios, configure recurring "Loops," and execute trades inside a ring-fenced account where manual trade confirmation defaults on. By September 2026, over 150,000 customers had opened agentic trading accounts, generating nearly 30 million daily tool calls. πŸ”— πŸ”— -
Moody's Credit MCP & AI Skills Enterprise Platform Sep 2026 Moody's released portable AI Skills built on the open SKILL.md standard alongside federated Model Context Protocol (MCP) connectors across Microsoft 365 Copilot (Researcher, Word, Excel, Teams) and Google Cloud's Gemini Enterprise for Financial Services (distinct from Moody's AI Studio). Without copying underlying datasets, customers run multi-step earnings summaries, peer comparisons, rating pitches, and counterparty credit reviews grounded in entitlement-controlled Moody's Ratings, corporate financials, ownership structures, and physical risk models with source citations, compressing research from days or hours to minutes. πŸ”— πŸ”— -
ChatGPT for Financial Services (OpenAI, Morgan Stanley & Evercore) Enterprise Platform Sep 2026 OpenAI launched ChatGPT for Financial Services, an institutional capital-markets suite co-designed with Morgan Stanley and Evercore on GPT-6 Astra. It pairs pre-indexed SEC filings, Daloopa fundamentals, PitchBook, LSEG News, Quartr transcripts, and 50+ entitlement-respecting connectors (including S&P Capital IQ, FactSet, MSCI, Moody's, and Datasite) to build auditable LBO and comparable-company models, screen M&A buyers, and generate firm-templated pitchbooks with inline cell-level citations and compliance audit exports. No seat counts or productivity metrics were disclosed at launch. πŸ”— -
FactSet & Google Cloud (Workstation & MCP) Enterprise Platform Aug 2026 FactSet partnered with Google Cloud to embed Gemini models, Google Search grounding, and the Gemini Enterprise Agent Platform into FactSet Workstation for multimodal deep research (distinct from FactSet AI for Banking), and launched its AI-Ready Data and Portfolio Analytics MCP connectors inside Gemini Enterprise for Financial Services. Analysts, portfolio managers, and risk teams run plain-language queries over live FactSet book-of-record holdings, fixed-income and derivatives tools, scenario and hedge analytics, and bond total-return calculations under enterprise entitlements. No adoption metrics disclosed. πŸ”— -
MSCI MCP Connector Integrated Solution Aug 2026 MSCI launched its Model Context Protocol (MCP) Connector inside Google Cloud's Gemini Enterprise for Financial Services, giving institutional investors entitlement-controlled, natural-language access to MSCI datasets within a single conversation. Clients query MSCI index constituents and factor analytics, benchmark private-capital funds (IRR, TVPI, DPI), inspect multi-asset portfolio liquidity and cash flows, evaluate real-assets indices, and screen sustainability and climate exposures grounded in MSCI AI Portfolio Insights. No client adoption metrics were disclosed at launch. πŸ”— -
Abu Dhabi Securities Exchange (ADX) MCP Integrated Solution Aug 2026 The Abu Dhabi Securities Exchange (ADX) launched what it bills as the MENA region's first exchange-operated Model Context Protocol (MCP) server, opening live official market data directly to ChatGPT, Claude, and agentic workflows. Retail and institutional investors, brokers, and developers query real-time Level 1 prices, per-symbol order-book depth, retail-versus-institutional and foreign-versus-local flow statistics, corporate actions, and machine-readable XBRL financial disclosures in plain language across four quota tiers (up to 4,000 calls/month). πŸ”— -
S&P Global Adaptive Retrieval Enterprise Platform Jul 2026 S&P Global and Kensho launched Adaptive Retrieval inside the unified S&P Global AI Data Portal to complement Kensho's Deterministic Retrieval for client LLMs and multi-agent workflows (distinct from Capital IQ Pro ChatIQ and HorizonsAgents). Accessible via Model Context Protocol (MCP) connectors, plugins, and reusable skills, the service dynamically decomposes natural-language financial queries, routes lookups across multiple licensed S&P Global datasets, and returns cited, auditable responses enforcing user entitlements. Routing models and latency benchmarks were not disclosed. πŸ”— -
Hebbia Matrix (OpenAI & Anthropic) Integrated Solution Jul 2026 Serving over a third of the top 50 asset managers and investment banks, Hebbia orchestrates OpenAI (o1, o3-mini, GPT-4o) and Anthropic (Claude Fable 5, added July 2026) models in its multi-agent Matrix diligence platform, alongside direct Intercontinental Exchange (ICE) equity and fixed-income pricing feeds (May 2026). Agents decompose data-room filings and credit agreements to draft investment memos, extract covenants, and flag legal risks with inline cell-to-source citations, lifting internal financial Q&A accuracy ~20% with Fable 5 and cutting research time up to 85%. πŸ”— πŸ”— -
Broadridge LTX Agentic BondGPT Integrated Solution Jun 2026 Broadridge-backed corporate bond electronic trading venue LTX added trader-configured agentic execution capabilities to its BondGPT application across 40+ liquidity providers (including Goldman Sachs, J.P. Morgan, Morgan Stanley, BofA, and TD Securities) and 100+ buy-side firms. Traders use natural-language instructions to configure agents that monitor live Liquidity Cloud conditions, trigger alerts, populate trade tickets, select dealers, launch RFQs, and auto-execute within policy-driven size limits, pre-action explainability, audit logs, and human-in-the-loop approvals. Underlying LLMs were unnamed. πŸ”— -
Tradeweb TARA Integrated Solution Jun 2026 Tradeweb launched TARA (Tradeweb AI Research Assistant), a conversational AI assistant embedded directly inside its institutional U.S. credit electronic trading platform. Integrating intraday and historical Tradeweb trading activity, FINRA TRACE prints, and Tradeweb Ai-Price fixed-income analytics, TARA answers natural-language trader queries on market flows, liquidity conditions, and client-specific execution performance. Piloted with asset managers including T. Rowe Price ahead of broader U.S. rollout, Tradeweb did not disclose the underlying language models or latency metrics. πŸ”— -
Intercontinental Exchange & Anthropic Enterprise Platform Jun 2026 Intercontinental Exchange (ICE) joined Anthropic's Project Glasswing to deploy Anthropic's frontier Claude Mythos Preview model for defensive software security across its financial market infrastructure. Security engineering teams run the model across codebases underpinning the New York Stock Exchange (NYSE), global derivatives clearinghouses, fixed-income data services, and mortgage technology platforms to autonomously surface and remediate software vulnerabilities prior to production release. Vulnerability-detection counts and remediation-time reductions were not disclosed. πŸ”— -
LSEG Everywhere MCP Connector Enterprise Platform May 2026 Under its LSEG Everywhere program, London Stock Exchange Group rolled out managed Model Context Protocol (MCP) connectors linking licensed LSEG financial data to Gemini Enterprise, Claude, ChatGPT, Microsoft Copilot, AWS, Snowflake, and Databricks (distinct from its internal Amazon Q post-trade tool). The connectors stream entitled market pricing, macro series, company fundamentals, IBES estimates, and quantitative analytics into institutional agents; by Q3 2026, over two-thirds of LSEG's top 50 clients were actively using LSEG MCP connections. πŸ”— -
S&P Global Energy HorizonsAgents Enterprise Platform May 2026 S&P Global Energy launched HorizonsAgents, a suite of AI agents converting complex energy and sustainability data into decision-ready, auditable insights for banks, investors, project developers, and corporate finance teams. Four agents β€” Transition Finance (beta since November 2025), Data Center Intelligence, Sustainability Benchmarking, and Net Zero Investment β€” deliver comparable outputs in minutes, aligned with ISSB frameworks. No underlying models named. πŸ”— -
Obin AI & Google Cloud Enterprise Platform Apr 2026 Obin AI, an enterprise agentic-AI company founded by Google AI, JPMorgan, and Silver Lake veterans, partnered with Google Cloud, integrating Gemini and the Agent Development Kit and listing on Google Cloud Marketplace. Its Obin Financial Agent helps asset management, commercial lending, and insurance teams accelerate complex financial analyses; clients include institutions managing over $1 trillion in assets. πŸ”— -
QRAFTI - Apr 2026 A multi-agent agentic framework for empirical quantitative-finance research that exposes a panel-data toolkit plus data access and factor construction via MCP servers as callable tools. Uses chained tool calls with reflection-based planning to replicate established factors, formulate and test new signals, and generate standardized research reports with narrative analysis. - πŸ”—
FactSet AI for Banking Enterprise Platform Mar 2026 FactSet launched FactSet AI for Banking (alpha), an AI-native workflow-automation platform built with (and backed by an investment in) Finster AI for investment banking and sell-side teams. It automates creation of transaction artifacts across the deal cycle, from origination through execution, within a single governed environment with deep Microsoft Office integration, drawing on leading AI models (specific vendors not named). πŸ”— -
Bloomberg ASKB Enterprise Platform Feb 2026 Bloomberg launched ASKB (beta), an agentic conversational interface for the Terminal running a coordinated network of parallel agents built on a mix of commercial and open-weight LLMs. It answers natural-language queries, auto-generates Bloomberg Query Language, and runs multi-step workflows (earnings prep, post-event analysis), synthesizing Bloomberg data, news, filings, and analytics plus research from 800+ providers with source attribution. Distinct from BloombergGPT. πŸ”— -
AlphaSense Generative Search Enterprise Platform Jan 2026 AlphaSense rebuilt Generative Search as a multi-agent research system with dynamic planning and reasoning over hundreds of millions of premium documents plus structured financial data, with pre-built workflow agents that automate earnings analysis, market scans, and deliverable creation (slides and reports) for banking, PE/VC, and IR teams. A July 2026 update added native PowerPoint and Excel assistants; Blue Deer Capital reports building board-ready decks 60-70% faster. πŸ”— -
S&P Global & Google Cloud Enterprise Platform Dec 2025 S&P Global and Google Cloud struck a multi-year partnership headlined by the S&P Global Data Retrieval Agent, built by S&P's Kensho AI unit, which lets customers query S&P Global data directly inside Gemini Enterprise with citation-backed answers. The deal also unifies S&P's proprietary datasets on BigQuery for AI-ready distribution and deploys Gemini Enterprise internally for workforce productivity. πŸ”— -
FinRpt Fine-Tuned Model Nov 2025 First comprehensive framework for automating Equity Research Report generation using LLMs, featuring a dataset integrating 7 financial data types, an 11-metric evaluation system, and a multi-agent framework with SFT and RL training. AAAI 2026. πŸ”— πŸ”—
LLM Pro Finance Suite Fine-Tuned Model Nov 2025 A suite of five instruction-tuned multilingual finance LLMs (8B–70B) built on Llama, Qwen and Gemma bases via continual pre-training plus SFT on ~2M samples (54% finance; English, French, German). Targets financial reporting analysis, risk, compliance, sentiment, translation and RAG, preserving general capability. Open 8B checkpoints published on Hugging Face under the OVHaiLLM org. πŸ”— πŸ”—
S&P Capital IQ Pro ChatIQ Enterprise Platform Oct 2025 S&P Global upgraded Capital IQ Pro with Multi-Document ChatIQ and Document Intelligence 2.0 (ChatIQ built with S&P's Kensho AI unit), letting banking and buyside analysts query and compare across filings, transcripts, investor presentations, news, and research spanning multiple companies or industries, returning generative-AI insights with precise citations for full auditability. No adoption metrics disclosed. πŸ”— -
Coinvisor Fine-Tuned Model Oct 2025 An RL-enhanced chatbot agent for interactive cryptocurrency investment analysis, using a reinforcement-learning-based tool-selection mechanism for multi-step planning and real-time integration of diverse crypto data sources. Improves tool-orchestration recall by 40.7% and F1 by 26.6% over baselines, with a 20-investor user study rating average satisfaction 4.64/5. - πŸ”—
Deutsche Bank DB Lumina Enterprise Platform Sep 2025 Deutsche Bank Research built DB Lumina, an AI-powered research assistant developed with Google Cloud Consulting, combining Gemini models with a RAG architecture (Gemini Embeddings, Vertex AI Vector Search, pgvector) that grounds responses in internal research and sources like SEC filings with inline citations. In production since September 2024 with ~5,000 users (targeting 10,000+), it saves analysts 30–45 minutes on earnings note templates and up to 2 hours on research reports. πŸ”— -
Trading-R1 Fine-Tuned Model Sep 2025 Tauric Research's 4B financial trading model built on a Qwen3-4B backbone, trained with a multi-stage easy-to-hard SFT-plus-RL curriculum on Tauric-TR1-DB (100k samples over 18 months, 14 equities/ETFs, five data sources). Generates structured, evidence-grounded investment theses with volatility-adjusted decisions, delivering improved risk-adjusted returns and lower drawdowns versus open and proprietary baselines. πŸ”— πŸ”—
Moody's AI Studio Integrated Solution Aug 2025 Moody's built AI Studio, a multi-agent AI platform that automates financial workflows like credit memo generation, cutting analyst time from 40 hours to 2–3 minutes by orchestrating specialized agents using proprietary and third-party data. The platform has been commercialized for financial clients and adopted internally by 40,000 employees, driving large-scale efficiency and competitive advantage. πŸ”— -
Goldman Sachs & Cognition Devin Integrated Solution Jul 2025 Goldman Sachs began piloting Cognition's Devin, an autonomous AI software engineer, across its 12,000-plus developersβ€”initially hundreds of instances, potentially scaling to thousandsβ€”for tasks such as coding, testing, and debugging under a human-supervised "hybrid workforce" model. CIO Marco Argenti estimates three-to-four-times developer productivity gains. Distinct from Goldman's Meta-based GS AI Platform. πŸ”— -
IG Group & Anthropic Integrated Solution Apr 2025 IG Group partnered with Anthropic to deploy Claude for Work, doubling productivity in financial content creation, cutting over 70 hours of manual work weekly, and improving compliance and analytics workflowsβ€”all with ROI achieved in under three months. πŸ”— -
Endex & OpenAI Integrated Solution Feb 2025 Endex, an AI platform for financial firms, uses OpenAI’s advanced modelsβ€”GPT-4o, o1-mini, and o3-miniβ€”to power autonomous agents that retrieve, synthesize, and analyze complex financial data, helping firms automate tasks like earnings summaries, investment memos, and due diligence with greater speed and accuracy. πŸ”— -
FinBLOOM Fine-Tuned Model Feb 2025 FinBloom 7B, built on Bloom 7B, was trained on 14M financial news articles and 12M SEC filings, then fine-tuned with 50K financial queries for enhanced real-time data retrieval. This process ensures strong contextual understanding for financial decision-making. - πŸ”—
MUFG Bank Integrated Solution Feb 2025 MUFG Bank leveraged LLMs to automate data extraction and summarization from corporate reports, enabling faster financial analysis for FX & Derivative Sales. The system, leveraging retrieval-augmented generation (RAG) and fine-tuned prompts, reduces client presentation preparation from hours to minutes, thereby improving client advisory efficiency. πŸ”— -
Rogo & OpenAI Fine-Tuned Model Feb 2025 Rogo's fine-tuned OpenAI models and integration of extensive financial datasets (including S&P Global, Crunchbase, and FactSet) allows it to scale financial analysis and deliver real-time financial intelligence to >5k financial professionals, shifting their focus from manual work to high-value decision making. It uses GPT-4o for in-depth financial analysis, o1-mini for contextualizing financial data, and o1 for evaluations and advanced reasoning. πŸ”— -
TigerGPT Integrated Solution Feb 2025 Tiger Brokers integrated DeepSeek's AI model, DeepSeek-R1, into its AI-powered chatbot, TigerGPT. This adoption aims to enhance market analysis and trading capabilities for its customers through the improved logical reasoning capabilities. πŸ”— -
London Stock Exchange Group & Amazon Q Integrated Solution Jan 2025 London Stock Exchange Group’s post-trade services team deployed Amazon Q Business (running on Amazon Bedrock) with a front-end custom UI, document crawling of FAQs and rulebooks, and retrieval-augmented generation (RAG) plus validation via Claude v2 to deliver accurate answers in seconds for complex member queries. πŸ”— -
FinGPT Fine-Tuned Model Jun 2023 Open-source financial LLM (FinLLM) using a data-centric approach (based on Llama 2) for automated data curation and efficient adaptation, aiming to democratize AI in finance with applications in robo-advising, algorithmic trading, and low-code development. πŸ”— πŸ”—
Ask FT Integrated Solution Mar 2024 LLM tool by Financial Times (FT) that enables subscribers to query and receive responses derived from two decades of published FT content. πŸ”— -
BloombergGPT Fine-Tuned Model Mar 2023 50-billion-parameter LLM specifically designed for financial applications and the industry's unique terminology, trained on a 363-billion-token dataset sourced from Bloomberg’s proprietary data, complemented with 345 billion tokens from general-purpose datasets πŸ”— πŸ”—

Payments & Transaction Services

Name Type Date Description Site Paper
Chime & Decagon Integrated Solution Oct 2026 Chime deployed Decagon's unified AI agent platform across both chat and voice member support using natural-language Agent Operating Procedures (AOPs) and cross-channel shared memory. Deterministic technical guardrails enforce core business logic as agents autonomously execute lost-or-stolen card replacements, track pending transactions and direct deposits, and update account preferences. Within two weeks of voice launch, the agents scaled past 1M monthly calls, resolving ~70% of voice and 70%+ of chat inquiries while cutting support costs by 60%; underlying LLMs are unnamed. πŸ”— -
Visa Commercial Agentic B2B Pilot Integrated Solution Oct 2026 Visa launched a live end-to-end B2B agentic payments pilot with accounts-payable platform BILL and accounts-receivable platform Billtrust, built on Visa Intelligent Commerce and Visa's Trusted Agent Protocol (distinct from consumer Visa & OpenAI). Verified buyer-side AI agents log into supplier portals without shared human credentials, reconcile approved invoices, and settle transactions using scoped Visa tokenized commercial virtual cards alongside Visa Commercial Fraud Solutions. Underlying foundation models and pilot transaction volumes were not disclosed. πŸ”— -
FIS AML Investigations Agent Integrated Solution Oct 2026 FIS launched its commercially available AML Investigations Agent for U.S. banks, co-engineered with Anthropic using Claude models inside FIS's private cloud under the FIS Control Plane governance layer and integrated with FIS Horizon and Systematics cores. The agent gathers KYC profiles, transaction histories, and check images, evaluates typologies, and drafts FinCEN SAR narratives while linking every finding to source records for human compliance sign-off. Deploying with design partner Amalgamated Bank, FIS projects time reductions exceeding 90% on L1 alerts, 98% on L2 investigations, and 80% on SAR drafting. πŸ”— -
Airwallex AgentOS Enterprise Platform Oct 2026 Airwallex launched AgentOS in July 2026 (allowing finance teams to run accounts through AI agents like Claude and Cursor via CLI or MCP) and expanded in October 2026 with agentic Business Accounts and early access to its Agentic Commerce Suite. Finance teams authorize external AI agents to provision multi-entity accounts, move funds, and execute FX conversions within role-based approvals, while merchants expose catalogs, accept agent-led payments, and manage disputes alongside multi-dimensional AI token metering. Underlying LLM vendors and adoption metrics were not disclosed. πŸ”— πŸ”— -
American Express Intent Intelligence & ACE Enterprise Platform Oct 2026 American Express published its open Intent Intelligence Framework inside the Amex Agentic Commerce Experiences (ACE) Developer Kit to secure autonomous AI shopping transactions prior to checkout execution. Through a Create Intent API, an AI agent records merchant, product, expiration, and budget constraints for passkey-based cardmember sign-off, generating a cryptographic Intent ID and JWT Proof of Intent Token that mints a scoped digital payment credential enforced during issuer authorization and dispute resolution. Live transaction volumes were not disclosed. πŸ”— -
Block Buzz & Anthropic Enterprise Platform Oct 2026 Block deployed Anthropic's Claude Fable 5 and Claude Opus 5 inside Buzz, its open-source multi-agent engineering workspace extending its internal goose agent platform. Engineers steer a Fable 5 orchestrator that designs data models, API specifications, and migration plans before dispatching up to 45 concurrent Opus and Sonnet worker agents to edit code and run test suites within strict permission boundaries. Single orchestrated runs execute company-wide codebase migrations merging roughly 1,000 pull requests across millions of lines of code. πŸ”— -
Mastercard Agent Pay for Machines Enterprise Platform Sep 2026 Mastercard launched Agent Pay for Machines (AP4M) in June 2026 across 30+ partners (including Adyen, Stripe, Checkout.com, and Coinbase) for autonomous machine-to-machine payments with agent credentialing, spend limits, and multi-rail card/account/stablecoin settlement. In September 2026, Mastercard added Trust and Intelligence Services (entering U.S. issuer testing) alongside Cloudflare and KYA provider Skyfire, fusing web telemetry, agent identity, verifiable consumer intent, and fraud models into a real-time AI-agent transaction probability score. πŸ”— πŸ”— -
SAGA (PayPal) Fine-Tuned Model Sep 2026 PayPal AI developed SAGA, a 37M-parameter, 12-layer decoder-only Transformer foundation model (GQA, SwiGLU, 1,024-token context) that encodes two years of chronological user action sequences across branded checkout, P2P transfers, app engagement, email/push, and account management. Extending PRAGMA-style per-field tokenization to autoregressive modeling, it decomposes each event into seven round-robin field tokens trained jointly with per-field next-token prediction and event-boundary InfoNCE contrastive heads, outperforming single-token baselines on downstream recommender conversion lift. - πŸ”—
Stripe Checkout WebMCP Integrated Solution Sep 2026 Stripe instrumented its hosted Checkout UI across more than 7.8 million merchants with progressive-disclosure client-side WebMCP tools (distinct from Stripe Minions) so browser-based AI shopping agents transact via structured tool calls rather than fragile DOM scraping or screenshots. Exposing staged actions β€” get_order_summary, select_payment_method, fill_payment_form, and submit_payment β€” with zero merchant code changes and intact human confirmation boundaries, WebMCP reduced browser-agent token consumption by 42% and cut tool calls per checkout by 38%. πŸ”— -
Ant International Antom Transformer & FalconTST Fine-Tuned Model Sep 2026 Ant International unveiled two domain-specific Transformer foundation models powering nearly 100 payment, FX, and treasury products across Alipay+, Antom, WorldFirst, and Bettr (distinct from its AMP protocol): the 10B+-parameter Antom 3-in-1 Transformer (processing 90T of sequential, tabular, and graph payment data annually) and the 8.5B+-parameter FalconTST time-series transformer. Paired with Account for Agent (AFA) and AgentSafePay KYA controls, Antom cut subscription-LLM merchant chargebacks by up to 87%, while FalconTST exceeded 93% cashflow-forecasting accuracy and cut corporate FX hedging costs 30%–60%. πŸ”— -
PicPay Fraud Triage Agents Integrated Solution Sep 2026 Brazilian digital bank and wallet PicPay deployed a coordinated four-agent fraud pre-triage architecture built on Databricks Genie Agents (governed by Unity Catalog; underlying LLMs unnamed) to evaluate statistical transaction-anomaly alerts in near real time. Parallel agents inspect flagged-window transactions, query historical user behavior, consolidate a risk score, and standardize analyst handoffs (~12,000 tokens/case) with automatic fallback to human analysts after three retries, cutting false positives reaching humans by 60%, speeding peak triage up to 10x, and catching 93% of disputed value early. πŸ”— -
PayPal SCOUT (MCP Gateway) Enterprise Platform Aug 2026 PayPal deployed SCOUT (Selective Context Optimization for Universal Tooling) in production behind its enterprise Model Context Protocol (MCP) proxy gateway aggregating 200+ MCP servers and 2,000+ internal tools. Instead of preloading every tool schema into agent prompts, SCOUT exposes two model-agnostic MCP meta-tools (tool_search and execute_tool) backed by zero-downtime catalog updates and hybrid BM25-plus-dense vector retrieval fused via Reciprocal Rank Fusion, cutting per-query MCP tool-schema tokens by 99% (140.2k to 1.3k tokens). - πŸ”—
Amazon Bedrock AgentCore Payments (AWS, Stripe & Coinbase) Enterprise Platform Aug 2026 AWS launched general availability of Amazon Bedrock AgentCore Payments, developed with Coinbase and Stripe, enabling autonomous AI agents to pay for metered APIs, MCP tools, LLM inference, and paywalled web content via USDC stablecoin or card-funded Coinbase and Stripe Privy wallets. Supporting x402 (including dynamic ceiling pricing and an AgentCore Gateway MCP discovery server) and Stripe/Tempo's Machine Payment Protocol (MPP), it enforces short-lived credential tokens, deterministic infrastructure-level session spend/expiry caps, and CloudWatch audit trails. πŸ”— -
Alipay Agentic Commerce Platform & Ah Bao Enterprise Platform Aug 2026 Ant Group's Alipay launched what it bills as China's first full-stack agentic commerce platform alongside consumer AI super-agent Ah Bao and the Agent Hub Access (AHA) interoperability protocol (distinct from cross-border Ant International AMP). Merchants convert web catalogs and service workflows into MCP tools and agent Skills connected through AHA to Alipay's payment, identity, and risk rails; Ah Bao spans 10,000+ everyday services across five smartphone OEMs (>70% China market share) and 16 automakers. πŸ”— -
TransactionGPT (Visa) Fine-Tuned Model Aug 2026 Visa Research presented TransactionGPT at KDD 2026, a ~56M-parameter 3D-Transformer consumer payment foundation model pre-trained on billion-scale real-world Visa transaction trajectories. A Virtual Token Layer fuses intra-transaction tabular metadata, optional LLM embeddings, and sequential history while weight-tied heads jointly predict future time gap, dollar-cent amount, merchant, and MCC alongside supervised objectives, beating production anomaly models and fine-tuned LLMs at roughly 300x faster inference; exact proprietary AUC figures are withheld. πŸ”— πŸ”—
Stripe Compliance Agents Integrated Solution Jun 2026 Stripe built a production ReAct agent system on Amazon Bedrock to assist financial-crime compliance investigations across its $1.4T annual payment volume in 50 countries (distinct from Stripe Minions). Complex reviews are decomposed into a directed acyclic graph of vetted sub-tasks served via an internal LLM Proxy and dedicated Agent Service (hosting 100+ agents), injecting tool outputs as closed-loop observations with immutable audit logs while requiring human reviewers to decide each sub-task. Median review handling time fell 26% with >96% reviewer helpfulness ratings. πŸ”— -
Adyen Agentic Enterprise Platform Jun 2026 Adyen launched Adyen Agentic (in limited U.S. enterprise availability), a modular three-layer API suite β€” Agentic Feed, Agentic Cart, and Agentic Payments β€” that translates a single merchant integration across Google's UCP and AP2, OpenAI's ACP, and Meta AI checkout protocols (distinct from Adyen's 2023 internal support copilot). Merchants preserve merchant-of-record status, fraud rules, and customer relationships while accepting tokenized agent-initiated payments backed by Visa, Mastercard, American Express, and Salesforce; conversion metrics were not disclosed. πŸ”— -
Mastercard Merchant Extraction SLMs Fine-Tuned Model Jun 2026 Mastercard R&D and Economic Intelligence deployed a rank-32 LoRA-fine-tuned Llama 3.1-8B model reaching 96.95% F1 for production merchant entity extraction from noisy payment transaction descriptors, and benchmarked 23 compact replacements (270M–8B parameters) across Gemma 3, Qwen 3.5, and Aya 3.35B. Qwen 3.5-4B with JSON-only prompting achieves 96.60% F1 and 91.67% strict record exact match at 1.2x higher batched throughput, while Qwen 3.5-0.8B hits 94.75% F1 across 14 Databricks endpoints. - πŸ”—
Visa & OpenAI Enterprise Platform Jun 2026 Visa partnered with OpenAI to embed secure Visa payments into ChatGPT and OpenAI agent experiences, letting AI agents shop and pay on a user's behalf at Visa-accepting merchants. Built on Visa Intelligent Commerce with tokenized credentials, real-time authorization, fraud monitoring and user spend controls. Announced at the Visa Payments Forum in San Francisco. πŸ”— -
Fiserv agentOS Enterprise Platform May 2026 Fiserv launched agentOS, a four-layer agentic operating system for financial institutions built on Amazon Bedrock AgentCore in strategic collaboration with OpenAI and AWS. Integrating directly with Fiserv core banking, card issuing, and payment processing systems, the platform routes tasks across frontier OpenAI models and specialized SLMs under unified identity permissions, policy guardrails, audit logs, and kill switches. Co-designed in bank pilots for commercial loan onboarding and operational reporting; quantitative benchmark metrics were not disclosed. πŸ”— -
Klarna Shopping Search (ChatGPT) Integrated Solution May 2026 Klarna launched an AI-powered Shopping Search app inside ChatGPT, powered by its Product Search MCP server, surfacing real-time prices, availability and offers from over 100 million products and 400 million merchant listings across 13 markets within a single conversation before redirecting users to the merchant's site to check out. πŸ”— -
Affirm & Google Integrated Solution May 2026 Affirm partnered with Google to embed its buy-now-pay-later options into Google Search, AI Mode, the Gemini app and Google Pay, giving shoppers transparent pay-over-time financingβ€”with real-time eligibility checks and full cost/schedule disclosureβ€”inside agentic-commerce checkouts. The integration follows the Google- and Shopify-developed Universal Commerce Protocol. Announced May 12, 2026. πŸ”— -
Ant International AMP Enterprise Platform Apr 2026 Ant International launched the Agentic Mobile Protocol (AMP), an open-sourced payment framework for AI agents transacting through mobile interfaces, letting LLMs, platforms and merchants connect to 4.4 billion digital wallet users via the Alipay+ ecosystem (40+ wallet partners, 1.8 billion accounts, 150 million merchants). It halves integration steps versus traditional card binding and adds a "Know Your Agent" identity framework, money-back protection against account takeovers, cross-device support (smartphones, smartwatches, AR glasses, in-car systems) and real-time agent-to-agent settlement of payments as small as US$0.000001. πŸ”— -
Stripe Minions Integrated Solution Feb 2026 Stripe is deploying β€œMinions,” LLM-powered one-shot end-to-end coding agents that turn a Slack prompt into a fully formed, CI-passing pull request autonomously. The agents now generate and merge over 1,300 PRs per week, materially expanding engineering throughput πŸ”— -
PayPal & Google Cloud Integrated Solution Oct 2025 PayPal and Google Cloud introduced a joint agentic commerce solution that lets merchants deploy AI-powered shopping experiences on their own digital surfaces, combining Google Cloud's Conversational Commerce Agent (or custom agents built with the Agent Development Kit) with PayPal's payment capabilities while merchants keep control of the agent's tone, look, and customer relationships. Transactions run on open protocols β€” Agent2Agent (A2A) and the Agent Payments Protocol (AP2, built on A2A and MCP) β€” using cryptographically signed cart and payment mandates for trust and fraud controls. Announced the same day as PayPal's Agent Ready and Store Sync services. πŸ”— -
PayPal Agentic Commerce Integrated Solution Oct 2025 PayPal launched agentic commerce services: "Agent Ready" lets PayPal merchants accept payments on AI shopping surfaces such as Perplexity with built-in fraud detection, buyer protection and dispute resolution, and "Store Sync" makes merchant catalogs discoverable to AI agents while keeping them the merchant of record. Agent Ready rolls out in early 2026. πŸ”— -
Modern Treasury AI Integrated Solution Jun 2025 Modern Treasury built Modern Treasury AI (chat, task management, monitoring, and an AI agent) for payment operations, using LangChain's LangGraph to create controllable, auditable agents with traceable logic, granular permissioning, and human-in-the-loop approvals. Its first production agent automates complex payments, reconciliation, and ledger workflows, accelerating time-to-market versus traditional process automation. πŸ”— -
Mercado Libre & OpenAI Integrated Solution Sep 2024 Mercado Libre uses GPT-4o to power Verdi, an internal development platform that enables 17,000 engineers to rapidly build AI toolsβ€”including for its fintech arm, Mercado Pago. Verdi accelerates application development and automates tasks like customer service mediation, supporting financial operations at scale while improving efficiency and consistency across its digital payments ecosystem. πŸ”— -
FinQuery & Google Integrated Solution Jun 2024 FinQuery, a fintech company, is using Gemini as a productivity and collaboration tool to help in brainstorming sessions, draft emails 20% faster, manage complex cross-organizational project plans, and aid engineering teams with debugging code and evaluating new monitoring tools. πŸ”— -
Discover Financial Services & Google Enterprise Platform Apr 2024 Discover Financial partnered with Google Cloud to utilize LLMs to helps its 10,000 contact center representatives to search and synthesize information across detailed policies and procedures during calls. πŸ”— -
Adyen Integrated Solution Nov 2023 Adyen, a publicly-traded financial technology platform, uses LLMs to improve support operations, and enhance efficiency and response times, with smart ticket routing (which assigns tickets based on sentiment and content) and support agent copilot (which help agents answer tickets faster and more accurately) πŸ”— -
Stripe & OpenAI Integrated Solution Mar 2023 Stripe integrates OpenAI’s GPT-4 to enhance its payment platform by analyzing business websites for better support, improving developer assistance through technical documentation processing, and detecting fraud by analyzing community interactions. πŸ”— -

Insurance & Risk Management

Name Type Date Description Site Paper
WTW Radar Vision Integrated Solution Oct 2026 WTW embedded Radar Vision, a cloud-native AI business-performance and model-monitoring module built on Microsoft Foundry, Azure OpenAI in Foundry Models, and Azure Cosmos DB, into its Radar insurance pricing platform. An insurance-specific AI assistant automates diagnostics across pricing, underwriting, claims, and portfolio workflows, flagging where emerging experience diverges from expectations and explaining outcomes while keeping actuaries and underwriters in control of decisions. Radar is used by 23 of the world's 25 largest insurers; Radar Vision-specific adoption metrics were not disclosed. πŸ”— -
Manulife Enterprise AI Platform Enterprise Platform Oct 2026 Manulife launched its global Enterprise AI Platform on Microsoft Azure and Microsoft Foundry, paired with Microsoft Agent 365 as a centralized control plane for agent registry, observability, and policy governance across 30,000+ employees (distinct from its 2025 call-center RAG pilot and John Hancock Quick Quote). Reusable orchestration components power advisor sales enablement, underwriting and claims document processing, North American contact-center summarization across 110M+ annual interactions, and engineering tools (+30% developer productivity), building on $300M in realized AI value. πŸ”— πŸ”— -
SOMPO & Gemini Enterprise Enterprise Platform Oct 2026 Japanese insurer SOMPO deployed Google Cloud's Gemini Enterprise across its 34,000-person workforce to scale employee-built agentic workflows and engineering automation. Employees have built more than 10,000 custom AI agents to automate internal document search, synthesis, and summarization across insurance operations, while SOMPO's engineering teams cut development time for new models from one week to a single day. πŸ”— -
Aviva GenAI Protection Underwriting Integrated Solution Sep 2026 Aviva completed rolling out an in-house generative AI medical-report summarisation tool across its entire UK individual protection underwriting suite β€” spanning Life Insurance, Critical Illness, and Income Protection (completed September 2026). The system ingests lengthy GP medical files exceeding 90 pages and extracts structured clinical histories for human underwriters, who retain sole authority over coverage and pricing decisions. Across active deployments, it cut medical-file assessment time by roughly half while achieving 99.7% extraction accuracy; underlying LLMs were not disclosed. πŸ”— πŸ”— -
AXA Global AI Hub Enterprise Platform Sep 2026 AXA partnered with Publicis Sapient to scale its vendor-agnostic Global AI Hub (distinct from AXA Secure GPT), whose first version went live in July 2026 across five entities β€” France, Germany, Switzerland, the UK, and AXA XL. Incorporating centralized FinOps, SafetyOps, regulatory compliance checks, and mandatory human oversight on business-critical decisions, the platform orchestrates multiple LLMs and agents across motor claims automation, inbound customer email processing, and underwriting knowledge management. Specific LLM vendors and claim volumes were not disclosed. πŸ”— -
FinixDoc-VL (Ant Group) Fine-Tuned Model Aug 2026 Ant Group built FinixDoc, an agentic financial and claims document parsing system centered on FinixDoc-VL, a 4B vision-language model adapted from Qwen3-VL-4B for degraded mobile-captured receipts, medical claim files, IDs, and ultra-large tables. Training combines a confidence-routed human-in-the-loop Data Factory, homoglyph-aware contrastive learning, and multi-stage GRPO reinforcement learning targeting amounts, dates, IDs, and reading order, scoring 81.43 on the open-sourced FinixDocBench suite (+5.13 over the next-best open VLM) and 84.08 versus 78.73 on internal Ant workflows. πŸ”— πŸ”—
Zurich Insurance New Business 2.0 Integrated Solution Jul 2026 Zurich Insurance launched New Business 2.0 in Japan, a multi-agent generative AI life-underwriting platform that replaces sequential manual case reviews with 12 domain-specialized AI agents running in parallel (distinct from its 2024 Azure OpenAI P&C underwriting tools). Coordinated by an orchestration layer, individual agents simultaneously evaluate medical disclosures, occupational hazards, document completeness, cross-field consistency, and sanctions screening to draft structured underwriting rationales and follow-up tasks for human underwriters. Specific LLM weights and turnaround percentages were not disclosed. πŸ”— -
TravelersLLM (Travelers) Fine-Tuned Model Jun 2026 Travelers developed and deployed TravelersLLM, a proprietary property-and-casualty insurance large language model trained in-house on millions of internal policy, claims, and actuarial documents (distinct from its OpenAI-powered AI Claim Assistant). Operating alongside frontier commercial models as a foundation for Travelers' enterprise agentic ecosystem, it powers underwriting risk analysis, research, and institutional knowledge retrieval, outperforming off-the-shelf commercial models in accuracy, speed, and cost across tens of thousands of insurance benchmark questions; parameter count was not disclosed. πŸ”— -
Prudential Just Ask Integrated Solution May 2026 Prudential's Individual Life Insurance business introduced Just Ask, a generative AI tool for third-party advisors and distributors that surfaces product information, generates preliminary underwriting quick quotes and locates new-business forms, with book-of-business insights planned. It generates certain underwriting rates in minutes versus several business days; about 85% of first-time users reported it was helpful. πŸ”— -
Ping An AI Express Service Enterprise Platform May 2026 Ping An Group deployed AI Express Service, an agentic virtual assistant embedded across more than 10 subsidiary mobile applications serving over 250 million retail customers (~90 million monthly active users). Driven by proprietary Ping An foundation models, the assistant interprets a single spoken or typed instruction to orchestrate cross-subsidiary workflows spanning policy servicing, auto roadside rescue, motor claims filing, telehealth consultations, and retail banking, contributing to AI systems reviewing or processing 84% of Ping An's operational business volume. πŸ”— -
INS-S1 Fine-Tuned Model Mar 2026 Ant Group's industrial-scale insurance LLM family, built via a Verifiable Data Synthesis System and a Progressive SFT-RL curriculum blending verified-reasoning RL (RLVR) and AI feedback (RLAIF) for actuarial reasoning and regulatory compliance. Achieves a record-low 0.6% hallucination rate (HHEM) while outperforming DeepSeek-R1 and Gemini-2.5-Pro. Released with the 39k-sample INSEva benchmark. - πŸ”—
Travelers AI Claim Assistant Integrated Solution Feb 2026 Travelers launched an agentic AI Claim Assistant developed with OpenAI, using OpenAI's models and Realtime API for a fully agentic voice service that handles inbound auto damage claim calls end-to-end β€” consulting, advising and supporting customers, and escalating to live specialists when needed. Initially for auto damage claims, with expansion to additional lines of business planned. πŸ”— -
Humana Agent Assist Integrated Solution Feb 2026 Humana launched Agent Assist, built with Google Cloud's Vertex AI, Gemini and Gemini Enterprise for CX, to support member-service advocates with real-time conversation summaries, benefit details, automated call summaries and compliance guidance under a human-in-the-loop model. It supports 20,000+ advocates handling up to 80 million calls a year; piloted October 2025, with full rollout in 2026. πŸ”— -
LLM Claim Automation Fine-Tuned Model Feb 2026 A locally deployed, domain-adapted LLM that generates structured corrective-action recommendations from unstructured claim narratives, fine-tuned from pretrained models with LoRA on millions of historical warranty-claim records. About 80% of outputs are near-identical matches to ground-truth corrective actions, outperforming general-purpose and prompt-based LLMs; positioned as a viable approach for insurance claim processing. - πŸ”—
John Hancock Quick Quote Integrated Solution Jan 2026 John Hancock (Manulife's US life insurer) launched Quick Quote, a reasoning-based generative AI underwriting-support tool it bills as among the first quote-generating GenAI engines in US life insurance. It returns non-binding preliminary risk assessments for less-complex cases in ~15 minutes versus about a day (up to age 75, $10M face), designed for 4,000+ requests/month, with 20,000+ cases supported since a late-2024 pilot. No specific model or vendor named. πŸ”— -
Oscar Health Oswell Integrated Solution Oct 2025 Oscar Health unveiled Oswell, a member-facing health AI agent powered by OpenAI that explains lab results and medications, checks drug interactions, requests prescription refills, reviews symptoms and helps members prepare questions for doctors. Rolling out in 2026 across 573 counties in 20 states, reaching Oscar's ~2 million members. πŸ”— -
SIGNAL IDUNA & Gemini Enterprise Enterprise Platform Oct 2025 German insurer SIGNAL IDUNA is rolling out Google Cloud's Gemini Enterprise to over 10,000 employees and sales partners under its MOMENTUM 2030 strategy, letting staff build autonomous AI agents for their own job functions integrated with CRM, HR, and marketing systems. Its earlier Co SI Health Assistant made information searches 30% faster and cut escalations from 27% to 3%. πŸ”— -
Manulife & Azure Integrated Solution Apr 2025 Manulife built a RAG-based call center assistant on Microsoft Azure using Azure AI Search with GPT-3.5 for responses and Llama 3/GPT-4 for validation, reducing query time from 21.9 to 7.3 seconds and boosting service accuracy and efficiency. - πŸ”—
Allianz Insurance Copilot Integrated Solution Feb 2025 An internal generative AI system to assist with claims management. Launched in 2024 for auto claims, it leverages cutting-edge LLMs to streamline workflows and automate key tasks. πŸ”— -
AllianzGPT Enterprise Platform Feb 2025 Internal chatbot that leverages capabilities from different LLM providers (e.g., OpenAI, DeepSeek) on Azure to support general productivity tasks, as well as specific functional areas like audit, Risk Consulting actuarial and other business areas. πŸ”— -
Newfront & Anthropic Integrated Solution Dec 2024 Newfront, an insurance platform serving 20% of U.S. startups with unicorn status, integrated Anthropic's Claude AI to enhance efficiency and client service. Claude automates complex insurance tasks, such as answering detailed benefits questions, reviewing contracts, and processing loss run documents. πŸ”— -
Open-Insurance-LLM-Llama3-8B Fine-Tuned Model Nov 2024 Llama 3 (8B model) fine-tuned with LoRA on the InsuranceQA dataset – a corpus of insurance domain Q&A pairs – to specialize it for insurance queries and conversations πŸ”— -
Zurich Insurance & OpenAI Integrated Solution Nov 2024 Zurich Insurance Group turned to Microsoft Azure OpenAI Service to develop AI applications that lead to more accurate and efficient risk management evaluations, accelerating the underwriting process, reducing turnaround times, and increasing customer satisfaction. πŸ”— -
Zurich Insurance & Azure Integrated Solution Nov 2024 Zurich Insurance Group leverages the LLM capabilities in Azure OpenAI Service to enhance its underwriting process by converting unstructured customer dataβ€”such as images, emails, and reports in various languagesβ€”into structured, actionable insights. πŸ”— -
EXL Insurance LLM Fine-Tuned Model Sep 2024 Industry-specific LLM that supports critical claims and underwriting-related tasks, such as claims reconciliation, data extraction and interpretation, question-answering, anomaly detection and chronology summarization. EXL utilized NVIDIA NeMo end-to-end platform for the fine-tuning process on 2 billion tokens of private insurance data. πŸ”— -
Swiss Re & mea Integrated Solution Sep 2024 Swiss Re partnered with mea (a generative AI-powered global platform) for insurance process automation, such as extracting unstructured data from submission-related documents (e.g., Schedules of Value) and convert it into structured, analysable data. πŸ”— -
Bitext Mistral-7B-Insurance Fine-Tuned Model Jul 2024 Fine-tuned version of Mistral-7B-Instruct-v0.2, specifically tailored for the insurance domain. It is optimized to answer questions and assist users with various insurance-related procedures πŸ”— πŸ”—
Five Sigma - Clive Integrated Solution Jul 2024 Insurance tech startup Five Sigma leveraged Google's Gemini models to create an AI claims adjuster engine (Clive) which frees up human claims handlers to focus on areas where a human touch is valuable, like complex decision-making and empathic customer service. πŸ”— -
New York Life Enterprise Platform Jul 2024 New York Life, the largest mutual life insurance company in the US, is actively integrating GenAI across various business functions to enhance efficiency e.g., underwriting, customer service, and hiring. πŸ”— -
Trumble Insurance Agency & Google Integrated Solution May 2024 Trumble Insurance Agency is using Gemini for Google Workspace to significantly improve its creativity and the value that it delivers to its clients with enhanced efficiency, productivity, and creativity. πŸ”— -
AXA Secure GPT Enterprise Platform Apr 2024 AXA developed AXA Secure GPT, a generative AI platform powered by Azure OpenAI Services, to enhance employee productivity while ensuring data security. This platform enables AXA's 140k employees to utilize AI tools within a secure environment, facilitating tasks such as drafting reports, summarizing documents, and generating content. πŸ”— -
Allstate Insurance Integrated Solution Mar 2024 Allstate implemented generative AI models, specifically OpenAI's GPT, customized with company-specific terminology, to enhance customer communications and experience. This system generates approximately 50k daily emails for claims representatives, ensuring messages are clear, empathetic, and free from industry jargon. πŸ”— -
Groupama & Azure OpenAI Integrated Solution Feb 2024 Groupama, a leading French mutual insurance group, uses Azure OpenAI Service to power a virtual assistant that helps customer managers respond to policyholder inquiries with 80% accuracy, improving efficiency and response quality. It is done through providing pre-written responses to policyholder inquiries, drawing from a comprehensive and secure documentation corpus. πŸ”— -
LAQO insurance & Azure OpenAI Integrated Solution Nov 2023 LAQO, Croatia's first fully digital insurer, partnered with Infobip to develop Pavle, a 24/7 AI assistant powered by Azure OpenAI Service. Pavle resolves 30% of customer queries, allowing human agents to focus on complex cases and customer acquisition. πŸ”— -
Roots Automation - InsurGPT Fine-Tuned Model May 2023 Roots Automation released an LLM fine-tuned on insurance-specific documents (ACORD forms, First Notice of Loss (FNOL), loss runs) to automate and cut claims processing time significantly πŸ”— -