- Home
- AI agents
- Enterprise AI agent examples
- Examples by industry
Plan · Enterprise AI agent examples
AI agents by industry: sourced examples from pharma to the public sector
Industry AI agent examples are deployments a named company, its vendor or a government body has published, with the task, the platform and a figure. Across industries the same patterns repeat: agents that draft for a professional to sign, agents that answer from controlled content and agents that act in a system. Regulation decides which pattern comes first.
In this guide: 83 pages
Key takeaways
- Healthcare, banking and manufacturing have many published examples with figures in our research; pharma, energy and professional services have few measured results.
- Regulated industries keep a person on the decision. Novo Nordisk’s study reports, Banner Health’s oncology summaries, Bradesco’s audit work and Markerstudy’s claim call summaries are all drafts a professional checks.
- Agents that act on their own sit where volumes are high and a wrong action is cheap to reverse: order entry at Danfoss, freight quotes at C.H. Robinson, support tickets at Grupo Falabella and candidate conversations at The Adecco Group.
- Several large figures are calculations or plans. Highmark Health credits its assistant with a calculated USD 27.9 million of AI-enabled value in 2025 (Google Cloud use-case list, 2026), Toyota projects savings per facility and Genentech expects to automate 43,000 hours. Test them before you plan on them.
Which AI agent patterns show up in each industry?
Three patterns cover nearly every published industry example: drafting for a professional to sign, answering from controlled documents, and acting on high-volume transactions. The table maps them to twelve industries, with a named example in each cell.
Each pattern brings its own controls. A draft carries less risk per output because a professional reads every one, so the controls are a source for each claim and a recorded review by a named person. An answer agent adds one hard requirement, permission-aware retrieval, so each user sees only what that user may open. An agent that acts needs its own identity, limits on what it may write and an approval rule for exceptions, which is why the published ones sit on high-volume, reversible transactions.
The industries also differ in who publishes. Banks, insurers and hospitals appear mostly in Microsoft and Google customer stories, life sciences companies often in Anthropic and AWS stories, and the public sector in its own releases with trial results. A vendor story is written to sell the vendor’s platform, so the platform column says more about who wrote the story than about what works.
Publicly reported examples by industry and pattern, opened on 2 November 2026. "None published" means no example with a published result in our research set.
| Industry | Drafts for a person | Answers from content | Acts in a system |
|---|---|---|---|
| Pharma and life sciences | Novo Nordisk clinical study reports; Genentech research summaries | AstraZeneca clinical operations questions | None published |
| Healthcare | Acentra Health appeal letters; Banner Health oncology summaries; Manipal nurse handoffs | Humana call support for member advocates | None published |
| Banking | Tata Capital credit notes; Bradesco audit planning | Wells Fargo procedures; ABN AMRO customer requests | Goldman Sachs trade accounting (in development) |
| Insurance | ICICI Lombard and Markerstudy claim summaries | SIGNAL IDUNA health insurance questions | Hiscox quoting for complex risks |
| Manufacturing | HARTING configuration; Mercedes-Benz quality causes | Textron manuals; Georgia-Pacific operator guidance | Danfoss order entry |
| Retail | Walmart shift planning | Lowe’s associate answers | Grupo Falabella support tickets |
| Logistics | None published | None published | C.H. Robinson quotes, bookings and customs |
| Energy and utilities | None published | Eneco customer questions | AES audit reports |
| Public sector | UK Consult consultation themes | Citizens Advice Caddy; YoungWilliams citizen answers | UK Extract planning records |
| Staffing | None published | None published | The Adecco Group candidate agents |
| Legal services | A&O Shearman draft work product | Epiq claimant questions | None published |
| Accounting and audit | KPMG audit procedures with auditors in control | None published | None published |
Read across a row and the regulation shows. Pharma and healthcare publish drafts for a professional to sign, logistics and retail publish agents that act, and the public sector publishes trials with a method. In a regulated industry the first agent is usually a draft.
What have pharma, healthcare, banking and insurance companies published?
These four industries publish the most figures, and most come from agents that draft, summarize or answer for a professional who decides. The table adds examples beyond the one-per-industry overview, with each figure as published.
Publicly reported agents in regulated industries, as published by each source and opened on 2 November 2026. None of these companies is a QueryNow client.
| Company and industry | What the agent does | Published figure | Source and date | How to read it |
|---|---|---|---|---|
| Genentech (pharma) | Searches literature and internal repositories, plans multi-step research and returns cited summaries | Expected to automate over 43,000 hours of biomarker validation work | AWS case study, undated; built on what AWS now calls Bedrock Agents Classic | A projection |
| Banner Health (healthcare) | Drafts oncology history summaries and flags inconsistencies such as records from another patient | 1,400 clinical notes in the pilot since June 2025; 85% of users report time savings | Anthropic customer story, undated | A user survey; halving administrative burden is a goal |
| Manipal Hospitals (healthcare) | Generates nurse handoff reports: medication changes, lab results, vital signs, concerns | Handovers from 90 minutes to around 20 | Google Cloud customer story, undated | Company and vendor claim |
| Kuwait Finance House (banking) | RiskGPT runs risk models across branches; risk executives decide | Credit case evaluation from four to five days to under an hour | Microsoft customer story, 3 December 2024 | Older source; the 96% headline is not explained in the body |
| Banco Bradesco (banking) | AILA supports internal audit planning, reporting and root-cause analysis; auditors own the work | 65% less time on planning; 55% less time on reports | Microsoft customer story, 24 June 2025 | No baseline or method |
| SIGNAL IDUNA (insurance) | Co SI helps service agents answer complex health insurance inquiries | For less experienced agents, searches 30% faster and escalations from 27% to 3% | Google Cloud use-case list, updated 22 April 2026 | Google’s wording in a roundup |
| Markerstudy Group (insurance) | Summarizes claims calls; a handler checks each summary | About four minutes saved per call; 56,000 hours a year is calculated from 840,000 calls | Microsoft UK Stories, 26 November 2024 | Older source; the yearly total is a calculation |
Model risk, clinical safety and claims regulation explain the shape. A draft that a credit manager, nurse or adjudicator signs fits the controls these industries already run, while an agent that decides alone would pull in validation and oversight duties first.
What have manufacturing, retail, energy and public sector organizations published?
Outside the regulated four, more agents act, and the public sector publishes trials with a method. The table adds examples the overview does not carry.
Publicly reported agents in other industries, as published by each source and opened on 2 November 2026. None of these organizations is a QueryNow client.
| Organization and industry | What the agent does | Published figure | Source and date | How to read it |
|---|---|---|---|---|
| HARTING (manufacturing) | An assistant works with Siemens NX and Simcenter to configure custom prototypes | Configuration from 15 to 20 minutes to about 1 minute | Microsoft Cloud Blog roundup, 24 July 2025 | One roundup line |
| Walmart (retail) | A task tool in the associate app prioritizes and recommends tasks for team leads | Shift planning from 90 minutes to 30 | Walmart release, 24 June 2025 | An early estimate by managers |
| The Home Depot (retail) | A phone agent identifies what callers need; Magic Apron advises customers | In a 50-state pilot the phone agent identified caller needs within 10 seconds | Google Cloud use-case list; Home Depot release, 11 January 2026 | Pilot figure from Google; the company release gives none |
| Eneco (energy) | A website agent answers customers and hands unsure cases to live chat with a summary | 24,000 chats a month against 10,000 before; 67% handled without handoff against 40% | Microsoft customer story, 5 March 2025 | Company claim through the vendor |
| YoungWilliams (public services) | Answers citizens on state health and human services programs and summarizes cases for staff | 41% higher satisfaction; wait times halved; first response in 3 seconds against 3.95 minutes | Microsoft customer story, 19 May 2025 | Contractor and vendor claim; the states are not named |
| UK Government, Consult (public sector) | Finds themes in consultation responses and sorts each response; officials review | First live trial on 2,000+ responses: F1 score of 0.76 and theme rankings "nearly identical" to human reviewers | GOV.UK, 14 May 2025 | A measured trial; time and cost savings are projections |
The UK Incubator for AI also reports that Caddy, built with Citizens Advice to find guidance for advisers, halved response time in a randomised controlled trial, and that Extract has turned historic planning records into data for more than 120 councils since its launch to all councils in England on 4 June 2026.
How should you use industry examples when you plan an agent?
Use them to choose the pattern and the controls and to decide what to measure in your own baseline. Set the target from your own numbers, because the published figures are claims measured on someone else’s process.
- Match on the pattern before the industry. A claims summary in insurance and an appeal letter in healthcare have more in common with each other than either has with an autonomous order agent.
- Check the date and the source type. Pre-2025 stories describe older models and platforms, and a one-line roundup entry is weaker than a customer story with a method.
- Separate measured, estimated and planned. "Expected", "potentially", "up to" and "calculated" mark claims you have to test yourself.
- Take the controls from a well-documented example in your pattern: who reviews, what the agent may write and how the result is measured.
When no example exists for your industry and pattern, look one column over. An energy company planning an answer agent for field crews can learn more from Textron’s maintenance manuals than from any energy example in the set.
How do Microsoft, AWS and Google implement it?
Which cloud each industry example ran on, as the sources state. Vendor roundups supplied many entries, so each vendor appears most often among its own customers.
Capability, Microsoft, AWS, Google
| Capability | Microsoft | AWS | |
|---|---|---|---|
| Pharma and healthcare | Acentra Health on Azure OpenAI | Genentech and Novo Nordisk on Amazon Bedrock; Banner Health in its AWS environment | Manipal Hospitals and Highmark Health on Google Cloud; Merck & Co. with Gemini Enterprise |
| Banking and insurance | Tata Capital, Kuwait Finance House, ABN AMRO, Wells Fargo, Bradesco, ICICI Lombard, Markerstudy | No banking or insurance example with figures in our research set | United Wholesale Mortgage, SIGNAL IDUNA, Hiscox underwriting |
| Manufacturing, retail and energy | Textron Aviation, HARTING, Eneco | Georgia-Pacific on Amazon Bedrock | Danfoss, The Home Depot, Grupo Falabella; AES with Claude on Google Cloud |
| Public sector | YoungWilliams on Microsoft Foundry Agent Service | No public sector example in our research set | The UK planning prototype with Google DeepMind and Google Cloud |
Several organizations built their own agents: C.H. Robinson in logistics, Walmart in retail and the UK Incubator for AI for Consult, Extract and Caddy.
Frequently asked questions
Which industries use AI agents most?
We know of no reliable count. In the sourced examples we checked, healthcare, banking and manufacturing have more named deployments with figures than other industries, and pharma, energy and professional services have few. Vendor roundups skew the picture toward each vendor’s own customers.
Are AI agents allowed in regulated industries?
Yes, with a person on the regulated decision. Most published pharma, healthcare, banking and insurance agents draft, summarize or answer, and a clinician, nurse, credit manager or adjudicator decides; Hiscox’s automated quoting for complex risks is the main exception. Regulation adds logging and oversight duties that depend on the use and the jurisdiction.
Why do public sector examples publish trials?
Government bodies publish their method. Citizens Advice and the UK Incubator for AI ran a randomised controlled trial of Caddy, and Consult’s first live trial reports an F1 score against human reviewers. Few private-sector stories give that much detail.
Do the same controls apply in every industry?
The core set does: an owner, an identity, defined data access, an approval rule and monitoring. Regulated industries add duties that depend on the use and the jurisdiction, such as clinical review in healthcare or the EU AI Act’s high-risk duties for credit decisions, and we write those into the acceptance criteria before the build.
How current are these examples?
We opened every source on 2 November 2026. Some are older: Kuwait Finance House and Markerstudy are from late 2024, and we mark them. Vendor roundups are updated in place, so an entry can be older than the date of the list.
More in Enterprise AI agent examples
Sources
Each source was read on 2 November 2026. The date shown is the date on the page, or the page’s ms.date for Microsoft Learn; an undated page shows the date it was read.
- Genentech case study: generative AI to get lifesaving medicines to patients faster
- Banner Health Claude Platform case study
- Novo Nordisk Claude Code case study
- Manipal Hospitals case study
- Kuwait Finance House cuts risk processing time by 96% with Microsoft AI solutions
- Audit of the future: Bradesco increases efficiency by 65% with Microsoft Azure
- Markerstudy Group embraces AI to drive innovation and excellence
- 1,302 real-world gen AI use cases from the world’s leading organizations
- AI-powered success with more than 1,000 stories of customer transformation and innovation
- Walmart unveils new AI-powered tools for 1.5 million associates
- The Home Depot and Google Cloud launch agentic AI tools to help customers and associates
- Eneco moves to Microsoft Copilot Studio to better support over 1.5 million customers
- YoungWilliams customer story (Azure AI Agent Service)
- Government-built Humphrey AI tool reviews responses to consultation for first time in bid to save millions
- Building better frontline services (Caddy)
- Building better planning (Extract)
- Danfoss and Go Autonomous: reimagining order management at scale with Google Cloud
- Toyota scales enterprise AI with Deep Agents and LangSmith
- C.H. Robinson debuts Lean AI for ocean shipping, air freight, and customs
- Highlights from the 2025 AWS Life Sciences Symposium’s Clinical Trials track (AstraZeneca)
- Acentra Health boosts employee productivity with generative AI and Azure OpenAI Service
- Tata Capital: automating credit appraisal
- ABN AMRO Bank migrates to Microsoft Copilot Studio to support over 3.5 million customer conversations
- Goldman Sachs taps Anthropic’s Claude to automate accounting, compliance roles
- ICICI Lombard: increasing productivity using a claims copilot
- Mercedes-Benz accelerates the transformation of its production network through the use of AI and humanoid robots
- Textron Aviation enhances maintenance efficiency with Azure AI
- Bridging the knowledge gap: using generative AI on AWS to preserve critical expertise (Georgia-Pacific)
- Lowe’s virtual assistants boost satisfaction and sales
- AES customer story
- A&O Shearman and Harvey to roll out agentic AI agents targeting complex legal workflows
- Epiq Global customer story (Microsoft Power Platform)
- KPMG advances AI integration in KPMG Clara smart audit platform
- Adecco Group announces Salesforce’s Agentforce Coworker rollout across 40 countries
Planning an AI program?
Talk to the team that builds it and runs it after launch.
- Built to SOC 2, HIPAA and GDPR standards
- EU AI Act-aligned delivery
- Microsoft Solutions Partner