The Agentic Enterprise  ·  Companion to Chapter 21

The Case Atlas

Twenty-five enterprise agentic deployments — sortable by industry, phase, and ROI tier

25
deployments documented
9
industries represented
11
at enterprise scale today

Chapter 21 argues that use case selection is a portfolio management problem, not an optimization problem. This companion makes the argument concrete. We catalog twenty-five enterprise agentic deployments — every one cited to a primary or near-primary source — across nine industries, three deployment phases, and four ROI tiers. The atlas is meant to be read sideways, not top-to-bottom: filter for your industry, your stage, or your tolerance for risk, and look at what comparable organizations are actually shipping.

Why this exists

Most published agentic AI case lists are vendor-curated and optimistic. They feature pilots that were announced but never reached production, conflate copilots with autonomous agents, and report return-on-investment figures lifted from press releases without source attribution. Chapter 21 asks the reader to assemble a portfolio. We owe the reader the raw material to do that honestly.

The atlas is the raw material. Every entry names the deployer, the vendor or build path, the launch year, the deployment phase as we read it from public reporting, and the most defensible ROI metric we could find. Where the metric is a vendor claim, we say so. Where it is a deployer-confirmed number — JPMorgan's CTO citing 360,000 attorney-hours saved, Klarna's CEO disclosing the equivalent of 700 agents — we cite the disclosure.

How to read the atlas

Three filters matter. Industry tells you whether anyone shaped like you has shipped this work. Deployment phase tells you whether the case is a real reference or an aspiration. ROI tier — Transformative, High, Medium — tells you what good looks like for your sector, not what is theoretically possible.

Three traps to avoid. First, the atlas over-indexes on financial services and retail because those sectors disclose more aggressively, not because they are universally further ahead. Second, "Enterprise" phase means the system is in production at scale today; it does not mean the system has been running long enough to validate ROI claims. Third, deployer-named ROI metrics are themselves a published artifact — they have an audience and a purpose. Read them with the discount you would apply to any earnings-call number.

Three companion chapters in Report №03 go deeper. The Klarna, JPMorgan, and McDonald's rows in the atlas link directly to chapters that examine each case in narrative depth — including, in the McDonald's case, what it means when a much-lauded pilot is rolled back.

What the shape of the data shows

Three views of the same 25 cases. Each is a count, not a forecast. Read each chart as a description of what disclosed deployments look like — not as a base rate for your sector.

Use cases by industry

Distribution across 25 disclosed deployments

Observation. Financial Services (5), Retail & E-commerce (4), and the next industry account for 52% of disclosed cases. Interpretation. These sectors disclose aggressively because AI is competitive narrative. Caveat. Absence in this chart is not absence in the field — manufacturing, defense, and government are deploying agents but disclose less.

Deployment maturity

Pilot → Scaling → Enterprise

0 3 6 9 12 2 Pilot 12 Scaling 11 Enterprise
Observation. 23 of 25 systems (92%) are past pilot. Interpretation. The named-deployer set we tracked is selection-biased toward live production — vendors and CIOs publicize what works, not what stalled. Read accordingly. A 12-of-25 "Scaling" count is not an industry base rate; it is a ceiling on what disclosed wins look like.

ROI tier distribution

Impact magnitude across deployments

25 deployments
Medium2
High18
Transformative5
Observation. 23 of 25 disclosed ROI metrics sit in High or Transformative tiers. Interpretation. Deployers do not publish modest wins. Recommendation. Treat tier labels as shape, not magnitude. Validate the underlying number — hours saved, dollars avoided, latency cut — before benchmarking your pilot against any of them.

The atlas

Bank of America
Internal (BofA AI/ML)
Enterprise Transformative
Erica AI-Driven Virtual Financial Assistant
Financial Services · Customer Service / Retail Banking · launched 2018

ROI 3 billion total client interactions since 2018; 50 million users; 58 million interactions/month as of Aug 2025 (Bank of America press release)

BlackRock
Internal + Microsoft Azure OpenAI (LangChain/LangGraph)
Scaling High
Aladdin Copilot AI Agent Platform
Financial Services · Investment Management / Operations · launched 2024

ROI Increased productivity and alpha generation for investment professionals across $11T AUM platform; democratizes expert Aladdin access to all users (per BlackRock engineering leads, LangChain 2025)

Goldman Sachs
Cognition (Devin)
Pilot High
Devin Autonomous Software Engineer
Financial Services · IT / Software Engineering · launched 2025

ROI Estimated 3–4x productivity improvement vs. earlier AI tools (per Goldman CIO Marco Argenti, CNBC July 2025); code migration tasks run 10x faster than human engineers

JPMorgan Chase
Internal (JPMorgan AI/ML team)
Enterprise Transformative
COIN Contract Intelligence Platform
Financial Services · Legal / Operations · launched 2017

ROI 360,000 attorney-hours saved annually; loan service error rate reduced by ~90% vs. a decade prior (per JPMorgan CTO Sri Shivananda, Q3 2024)

Morgan Stanley Wealth Management
OpenAI
Enterprise High
AI @ Morgan Stanley Assistant & Debrief
Financial Services · Finance / Client Advisory · launched 2023

ROI 98% of Financial Advisor teams adopted the Assistant; significant reduction in post-meeting administrative time (per Morgan Stanley press release, June 2024)

Amazon
Internal (Amazon AI)
Enterprise Transformative
Rufus AI Shopping Assistant
Retail & E-commerce · Customer Service / Marketing · launched 2024

ROI $10B+ in projected annualized incremental sales; customers using Rufus are 60% more likely to complete a purchase; 250M shoppers used it in 2025 with 140% YoY user growth (Amazon CEO Andy Jassy, Q3 2025 earnings)

Shopify
Internal (Shopify Magic)
Scaling Medium
Sidekick AI Commerce Co-Pilot
Retail & E-commerce · Operations / Marketing · launched 2024

ROI Merchants use Sidekick to onboard seasonal staff 50% faster; automated Flow workflows deployed without coding; multi-step root-cause analysis delivered in seconds vs. hours of analyst work (Shopify Winter 2026 Edition)

Walmart
Pactum AI
Scaling High
Autonomous Supplier Contract Negotiation Agent
Retail & E-commerce · Supply Chain / Procurement · launched 2021

ROI 3% average cost savings across negotiated contracts; 75% of suppliers preferred negotiating with the AI vs. humans; negotiations across 2,000 suppliers simultaneously (Fox Business / Pactum, 2023)

Walmart
Internal (Walmart AI, powered by Wallaby LLM)
Scaling High
Sparky AI Shopping Agent
Retail & E-commerce · Customer Service / Marketing · launched 2024

ROI 35% higher average order value for Sparky users vs. non-Sparky customers; eliminated 30M unnecessary delivery miles via AI route optimization (Walmart AI Strategy 2025 / Constellation Research)

Bayer
Thoughtworks (co-developed with Bayer R&D)
Scaling High
PRINCE Multi-Agent Drug Discovery System
Healthcare & Pharma · R&D · launched 2024

ROI 90% reduction in manual effort for legacy study report retrieval; regulatory document drafting reduced from weeks to minutes; Bayer targets 40% R&D productivity increase by 2030 via AI (Bayer / Pistoia Alliance 2025)

Epic Systems (deployed at 70+ health systems)
Epic + Microsoft Azure OpenAI
Scaling High
Epic Agent Factory — Clinical Workflow AI Agents
Healthcare & Pharma · Operations / Clinical · launched 2024

ROI ~2/3 of Epic providers have used AI features; ambient documentation reduces clinical note time significantly; 100+ AI features live or in development across Epic's EHR platform (MedCity News, March 2026)

Moderna
OpenAI (ChatGPT Enterprise)
Enterprise High
mRNA Research & Operations AI Agent Suite
Healthcare & Pharma · R&D / Operations / Legal · launched 2023

ROI 750+ custom GPTs deployed in first 2 months; near-universal employee adoption within 6 months; Dose ID GPT reduces vaccine dose optimization from weeks to hours (OpenAI / Moderna case study 2024)

Universal Health Services (UHS)
Hippocratic AI (Polaris platform)
Scaling High
Post-Discharge Patient Follow-Up AI Agent
Healthcare & Pharma · Operations / Patient Care · launched 2024

ROI Average patient satisfaction rating of 9.0/10 for AI agent interactions; 115M+ total patient interactions across Hippocratic's platform with no reported safety issues; $3.5B company valuation after 50+ enterprise health system deployments (Hippocratic AI / WellSpan, 2024–2025)

Schneider Electric
Internal (Aveva software) + AI/ML
Enterprise High
AI-Driven Smart Factory & Supply Chain Optimization
Manufacturing & Supply Chain · Supply Chain / Operations · launched 2022

ROI Hundreds of thousands of dollars in labor savings at Kentucky smart factory; eliminated conveyor chain-break downtime entirely via AI predictive maintenance; #1 Gartner Top 25 Supply Chain 2023, 2024, 2025 (Supply Chain Dive / SCMR)

Siemens (deployed at Schaeffler, thyssenkrupp, 100+ customers)
Siemens + Microsoft Azure OpenAI
Scaling High
Siemens Industrial Copilot for Engineering & Operations
Manufacturing & Supply Chain · Operations / Engineering · launched 2024

ROI Panel visualizations generated in 30 seconds vs. hours; generated code needs only 20% adaptation; 100+ companies actively using it; 120,000 engineers have access as of Oct 2024 (Siemens/Microsoft press release, Oct 2024)

Lemonade
Internal (Lemonade AI — Jim & Maya)
Enterprise Transformative
Jim AI Instant Claims Processing Agent
Insurance · Operations / Claims · launched 2017

ROI 55% of claims processed fully autonomously (no human input); Loss Adjustment Expense ratio of 6% vs. ~9% industry average; pet insurance cost-per-claim reduced 68% from $44 to $14 (Lemonade Q4 2025 Shareholder Letter)

Zurich Insurance
Internal (Zurich AI team) + Expert AI
Scaling High
CATIA Catastrophe Claims Intelligence Agent
Insurance · Operations / Claims · launched 2023

ROI CATIA identified 500 additional catastrophe claims during its pilot (5 events), generating $1.4M in savings; Expert AI partnership achieved 58x reduction in claim review time, saving 8 hours/policy and ~$40M/year in underwriting leakage; real-time fraud detection equivalent to £260K suspicious claims identified daily in 2024 (Zurich / Klover.ai analysis)

A&O Shearman (Allen & Overy)
Harvey (OpenAI GPT-4 + Azure)
Enterprise High
Harvey Agentic Legal Research & ContractMatrix
Legal & Professional Services · Legal · launched 2022

ROI 30% reduction in contract review time; 7 hours saved per contract via ContractMatrix; staff save 2–3 hours/week on routine tasks; 40,000 queries processed by Harvey in first rollout phase (A&O Shearman / Klover.ai, 2025)

Deloitte (deployed internally + at HPE)
NVIDIA (Llama Nemotron models via Zora AI by Deloitte)
Pilot High
Zora AI Finance Agent — Expense & Invoice Automation
Legal & Professional Services · Finance / Procurement · launched 2025

ROI Deloitte targets 25% cost reduction and 40% productivity increase in its own finance operations; HPE projects 50% reduction in reporting production time; Deloitte plans Zora deployment to thousands of internal users by end of 2025 (Deloitte press release, March 2025)

Klarna
OpenAI
Enterprise Transformative
OpenAI-Powered Customer Service AI Agent
Customer Service / BPO · Customer Service · launched 2024

ROI Equivalent of 700 FTE agents replaced; 2.3M conversations in month 1; handle time: 11 min → <2 min; 25% drop in repeat inquiries; $40M estimated annual profit improvement (Klarna press release, Feb 2024)

Telstra
Microsoft (Azure OpenAI Service)
Enterprise High
Ask Telstra & One Sentence Summary AI Agent Tools
Customer Service / BPO · Customer Service · launched 2024

ROI 90% of agents using One Sentence Summary reported time savings and increased effectiveness; 20% reduction in follow-up contacts; 84% of agents said Ask Telstra positively impacted customer interactions (Microsoft case study / Telstra Annual Report 2024)

Wiley (John Wiley & Sons)
Salesforce (Agentforce + Einstein AI)
Scaling High
Agentforce Customer Service AI Agent
Customer Service / BPO · Customer Service · launched 2024

ROI 213% ROI with $230K in savings; 40%+ improvement in case resolution vs. previous chatbot; onboarding of seasonal agents 50% faster (Salesforce customer story, 2024)

Cisco (internal + enterprise customers)
Internal (Cisco AI) + reasoning LLMs
Scaling Medium
Cisco AgenticOps — Autonomous Network Operations
IT Operations / Software Engineering · IT / Network Operations · launched 2025

ROI Configuration automation and network monitoring automation address the top-two priorities for Cisco's 69% of DevNet engineers; reduces MTTR for network incidents from hours to minutes (per Cisco AgenticOps announcement, Feb 2026)

GitHub / Microsoft (77,000+ enterprise customers)
GitHub / Microsoft (OpenAI GPT-4 Turbo)
Enterprise High
GitHub Copilot Enterprise — AI Coding Agent
IT Operations / Software Engineering · IT / Software Engineering · launched 2023

ROI 55% faster task completion (controlled study: 1h 11m vs. 2h 41m); 84% PR build success rate; 77,000 enterprise customers as of 2025; 20M+ users (GitHub/Gartner 2025 Magic Quadrant)

Workday (enterprise customers globally)
Internal (Workday Illuminate platform)
Scaling High
Workday Illuminate HR & Finance AI Agents
HR / Talent · HR / Finance · launched 2024

ROI 65% reduction in contract execution time; 90% reduction in time spent on staffing changes; 900 hours/year saved by Financial Audit Agent in early access; Payroll compliance 4x faster (Workday Rising 2025 press release, Sept 2025)

What the shape tells us

Observation. Eleven of twenty-five deployments are at enterprise scale today; twelve are scaling but not yet enterprise; only two remain in pilot. Five qualify as Transformative on ROI evidence; eighteen qualify as High; two as Medium. Financial services, retail, and healthcare account for thirteen of the twenty-five.

Interpretation. The "agents are still pilots" narrative is increasingly stale at the front of the curve. Where deployers have explicit governance, named owners, and the integration spine described in Pillar IV, agents are crossing into production at meaningful scale. The concentration in three industries is not a coincidence — they share three traits that Pillar III flagged as load-bearing: structured high-volume transactions, strong evals from regulatory pressure, and tolerance for human-in-the-loop on high-stakes outputs.

Recommendation. Use the atlas as a calibration tool, not a copy-from-here menu. Find the three rows closest to your industry and stage. Read the source. Ask whether the deployer's preconditions — data readiness, identity model, governance maturity — match yours today. If they do not, the gap is the work; the use case is the reward.

Provenance and limits

Every row carries one source URL — the strongest public artifact we found at the time of writing. Where a primary source exists (annual report, earnings call, deployer-published case study), we cite it. Where the strongest public artifact is a vendor case study or a major-outlet news report, we cite that and accept the discount. Two specific limits apply:

  • Selection bias. The atlas reflects what is publicly disclosed. Many enterprise agent deployments are intentionally not disclosed, particularly in regulated sectors where competitive advantage and disclosure liability point in opposite directions. The atlas should not be read as a representative sample of all enterprise agentic activity.
  • Time decay. Deployment phase and ROI tier are read at a point in time. Cases move. The McDonald's row in Report №03's narrative chapter is the canonical example: a much-cited pilot, then a rollback. The atlas will be re-tagged as deployers update their disclosures.

For deeper treatment of three of the entries in this atlas, see Report №03 — chapter 15 (Klarna), chapter 16 (JPMorgan), chapter 17 (McDonald's). For the framework that this atlas instantiates, return to chapter 21.