How Does $1.8 Trillion Anthropic Actually Create Value? 21 Customer Use Cases

How is AI actually changing the enterprise?

In its SEC S-1 filing, Anthropic presents 21 customer case studies showing how organizations deploy its models and platform. They fall into four functions:

  1. Autonomous Software Engineering & Modernization (Claude Code, Claude Fable, and autonomous coding agents): shipping production code, refactoring and migrating legacy codebases, and reviewing pull requests.
  2. Enterprise Knowledge Work & Cross-Silo Operations (Claude Enterprise, Claude Cowork, Claude Managed Agents, and MCP): linking internal and external data so teams move faster on R&D, underwriting, financial close, procurement, HR and IT questions, and field operations.
  3. Embedded Commercial AI Products (Claude Platform APIs, Claude Sonnet, and custom reasoning layers): building assistants and reasoning into other companies' products, including retail, ERP, CRM, legal tech, and restaurants.
  4. Automated Cyberdefense & Vulnerability Discovery (Claude Opus 4.6, Claude Mythos / Mythos Preview, and Project Glasswing): proactively scanning applications, APIs, open-source browser codebases, and telecommunications networks for security flaws.

The data science use cases come with limited technical detail and report speed rather than accuracy. Bristol Myers Squibb deployed Claude Code to 3,000+ developers and data scientists and used the Model Context Protocol (MCP) to connect Claude to internal sources and 30 to 40 external medical and claims databases. Real-world evidence queries fell from about two months to five minutes. The case study does not mention response quality. Travelers rolled Claude out to nearly 10,000 data scientists, engineers, analysts, and product owners for risk analysis, data prototyping, and model analysis, without reporting results on those tasks. At Jane Street, quantitative researchers use Claude Fable to monitor training runs and process results. They estimate about 50% more analysis than a year earlier. The filing does not say whether the additional analyses were evaluated for quality, or how.

PayPal's product teams built a custom MCP server so non-technical staff can query data, run product analysis, and make dashboards, which takes that work off analysts. AMOS, its Account Management Operating System, replaces manual dashboards with conversational queries over live data and cuts report preparation from days to 5-10 minutes. It also reports detecting subtle merchant growth signals but does not say how those signals are validated.

The tradeoff is between LLM-enabled output growth and validating that output. The open question is whether that growth creates a validation bottleneck.


The 21 case studies

Customers are listed alphabetically, since most of them span more than one function. The table is LLM-generated.

Note: Viewing on a smartphone? Rotate your phone to landscape for the full table.

# Customer Industry Use cases Reported results
1 Accenture (2025) Professional Services Internal Claude Code & Claude Enterprise; client AI transformation; Cyber.AI security scanning; client B2B marketing/sales rebuild 45,000 trained (18,000+ certified); Cyber.AI scan coverage 10% → >80%; 7x AI ROI for US broadband client
2 AIG (2023) Insurance Automated commercial underwriting (Underwriting by AIG Assist); claims triage (Claims by AIG Assist); exploring Claude Code & multi-agent underwriting 8 business lines; >90% faster submission review; Lexington Property: +30% quoted, -55% time-to-quote, +40% bound
3 Bristol Myers Squibb (2023) Pharmaceuticals Querying siloed internal R&D/clinical data via MCP + external medical/claims databases; coding (Claude Code); Claude Cowork & HR/IT self-service 30,000+ employees with access; real-world evidence answers cut from ~2 mo → ~5 min
4 DoorDash (2023) Local Commerce Production coding & migrations (Claude Code, including product managers and designers); company-wide desktop knowledge work (Claude Cowork) 100% corporate access; 7,000+ monthly Cowork users; launch timelines for major products cut by more than half
5 International Rescue Committee (2025) Humanitarian / Nonprofit Real-time frontline health insights; field malnutrition record digitization; portfolio synthesis & grant landscaping (Claude for Nonprofits) 8x user growth (~2,400 seats); 464 staff hours returned/week; IRC's reach: 20.5M+ people in 40 countries
6 Jane Street (2023) Quantitative Trading Autonomous multi-step quant research & training monitoring; codebase refactoring/migration (Claude Fable); design-to-code Fable 5.1 scored 32% higher than Fable 5 on Jane Street's coding evaluation; ~1 week codebase migration (down from months); one researcher estimates ~50% more analysis than a year earlier
7 KBF Design Gallery (2025) Interior Design & Remodeling Custom tools (Claude Code & Claude Cowork) for scheduling, design-to-construction handoff validation, contract scope writing, & client communication 490 hrs/yr recovered for lead designer; 7 hrs/project cut from contract writing; 70 projects/yr (+32% sales)
8 Legora (2024) Legal Tech Embedded reasoning layer for legal drafting, M&A data-room review & research; internal engineering (Claude Code) & Claude Enterprise 1,500+ legal orgs served; 4.3 non-billable hrs/week recaptured; 100% Enterprise access; 14x Claude Code growth
9 Lilly (2024) Pharmaceuticals Enterprise knowledge work (Claude Enterprise); biologist web-app building & chemist real-time visualization; Claude Science pilot; data-loss-prevention controls 50,000+ employees with access (5x expansion); >85% active engagement in 5 core functions
10 Mercado Libre (2023) E-Commerce & Fintech Autonomous coding, PR review & microservice migrations (Claude Code); finance close, variance analysis, & procurement agents (Claude Cowork) Merged code more than doubled year over year; 500,000+ autonomous code reviews a year; 31,000 Cowork users (78% non-tech active weekly)
11 Mozilla (2025) Software / Browser Firefox engineering (Claude Code); large-scale automated security vulnerability discovery (Claude Opus 4.6 & Mythos Preview) 22 vulnerabilities found & resolved (Opus 4.6, Firefox 148); 271 vulnerabilities found & resolved (Mythos Preview, Firefox 150)
12 PayPal (2025) Digital Payments Engineering (Claude Code); custom MCP self-serve product analytics; merchant underwriting/compliance; AMOS analytics & sales plugin 12,000+ monthly active users on Claude Enterprise; 45 min saved/sales member/week; AMOS report prep cut from days to 5–10 min
13 Rakuten (2024) E-Commerce & Tech Autonomous multi-hour coding (Claude Code); cross-company Claude Managed Agents in Slack/Teams (decks, spreadsheets, apps, overnight tasks) 70+ business units; 99.9% accuracy on one complex code change; 7+ hrs of autonomous coding on that change; feature time to market 24 → 5 days
14 Salesforce (2023) Enterprise Software Engineering (Claude Code / Agentforce Vibes); internal Slackbot, Claude Tag & Slack MCP; Agentforce 360 & Claudeforce / Headless 360 +151% effective output/dev YoY; 18x faster API migration; 2x QoQ productivity (8.1M hrs annualized); 99% Slackbot usage
15 SAP (2023) Enterprise Software Embedded reasoning in SAP Business AI / Joule (via MCP and agent-to-agent); Joule for Consultants; 50+ Joule Assistants & 200+ specialized agents Up to 14% faster project delivery (Joule for Consultants); 25 hrs saved per consultant/month
16 Shopify (2024) Commerce Platform Embedded merchant assistant Sidekick (analytics, promos, Flow automations); internal Claude Code tool-building & Fable 5 evaluation Shopify's scale: millions of merchants; 3x YoY daily active merchants on Sidekick; 7,500+ employees on Claude Code
17 Shy Bird (small business) Restaurant / Food Tech Kitchen workflow bottleneck analysis from photos, ticket/sales data, & cook input; built Colette restaurant intelligence platform 60%+ reduction in brunch ticket times (to 12 min); 60 restaurants using Colette; server became back-end engineer in 6 months
18 Travelers (2024) Insurance Full SDLC integration (Claude Code for code gen, docs, testing); enterprise prototyping & risk/business analysis Nearly 25k employees with Claude assistants (up from ~10k); -50% time to merge code
19 UST (2023) IT Services Chip schematic validation & digital-twin regression testing (Claude Code / Opus 4.8); client platforms (CarePath, IntelliOps, FinX) 20,000 associates in training; 50–70% cut in chip validation cycle time; 4 → 2 days standard validation turnaround
20 Verizon (2023) Telecommunications Technology & product engineering (Claude Code); network cybersecurity testing (Project Glasswing / Claude Mythos) 33,000+ engineers with Claude Code
21 Walmart (2024) Retail & E-Commerce Embedded AI shopping agent Sparky (Claude Sonnet 4.6); internal developer productivity agent WIBEY Sparky weekly active users up 110% quarter over quarter; +35% avg. order value on Sparky orders; units sold via Sparky up 2x quarter over quarter

Other popular articles

What if parallel trends are not so parallel?

How to (and not to) log transform zero

Explaining the unexplainable Part II: SHAP and SAGE