Luciano Ayres

The Enterprise AI Casebook

Enterprise AI Case Studies: Results, Evidence and Lessons

Real organizations and what they did with AI: what it measurably changed, how strong the evidence is, what went wrong, and how to adapt it in your organization.

Curated by Luciano Ayres · 4 cases · Updated

Showing all 4 cases

  1. Operations · Healthcare & Life Sciences

    How Kaiser Permanente Used AI Scribes to Save Physicians Nearly 16,000 Hours of Documentation

    Kaiser Permanente physicians used ambient AI scribes in 2.5 million visits over 14 months, saving nearly 16,000 documentation hours. What worked, what to watch.

    ProvenEvidence B: Disclosed

  2. Operations · Public Sector

    What the UK Government Learned Giving an AI Assistant to 20,000 Civil Servants

    The UK government's Copilot trial reported 26 minutes saved a day per user, but one department found no evidence of higher productivity. What leaders can learn.

    Mixed resultsEvidence B: Disclosed

  3. Software Engineering · Retail & Consumer

    How Airbnb Used AI to Finish a 1.5-Year Code Migration in 6 Weeks

    Airbnb used LLMs to migrate nearly 3,500 test files in 6 weeks instead of an estimated 1.5 years. How it worked, where automation stopped, and how to adapt it.

    ProvenEvidence B: Disclosed

  4. Customer Service · Financial Services

    How Klarna Used AI to Handle 80% of Customer Service Chats, and Why It Brought Humans Back

    Klarna's AI assistant handled 80% of its customer service chats in 2025, its annual report says, after the CEO conceded quality had slipped. What leaders learn.

    Mixed resultsEvidence A: Verified

Browse by business function

Browse by industry

Download the data

Every case as open data: organization, industry, business function, stage, outcome, evidence grade, measured results and sources. Free to reuse under CC BY 4.0 with attribution to The Enterprise AI Casebook.

CSV JSONLicense

How cases are graded

What do the evidence grades mean?

  • A: Verified. The headline result comes from a regulatory filing, an audited report, an earnings call or an independent study.
  • B: Disclosed. The organization itself stated the headline result on the record.
  • C: Reported. The headline result was reported by the press or the technology vendor, without the organization confirming it.

No case rests on vendor sources alone, and a case graded C is never marked as proven.

What do the outcomes mean?

  • Proven. Measured results at scale, backed by grade A or B evidence.
  • Promising. Early or vendor-reported results that still need confirmation.
  • Mixed results. Real gains alongside real costs or limits.
  • Reversed. The organization scaled back or changed course.

Why include cases that went wrong?

Because that is where most of the lessons are. Every case has a section on what didn't go to plan, and cases are updated when the story changes.

How do I use a case in my organization?

Each case ends with "Adapt it in your organization": the preconditions to check, a 90-day pilot with an owner and a success measure, the metric to track, the failure modes to plan for, and questions for your leadership team.