Data report

State of AI Use Cases

What 666 documented AI deployments across 201 use cases say about where AI runs, how far it acts on its own, and how often organizations prove the result.

Computed live from the library on 28 September 2026. By Len Debets.

666
Deployments
521 organizations in 46 countries
57%
Disclose an outcome
Of 666 public deployments, a measurable result
70%
Agentic
Of 666 public deployments, the AI completes work itself
201
Use cases
Canonical pages, each with evidence and a playbook

Finding 1: most AI results are claimed, few are proven

57% (n=666) of public deployments disclose a measurable outcome. Of the 439 reported metrics, 48% come from vendors and 52% from the deploying organization itself.

Who made the claim (reported metrics)
  • Organization
    227
  • Vendor
    211
  • Independent
    1
Evidence grade of deployment records
  • Grade B
    350
  • Grade C
    314
  • Grade D
    2

Finding 2: AI is moving from answering to acting

70% (n=666) of public deployments belong to use cases where the AI completes work itself (a supervised or autonomous agent, or an agentic workflow), rather than only assisting a person.

Use cases by autonomy level
  • Supervised agent
    89
  • Copilot
    72
  • Assist
    29
  • Autonomous
    11
Deployments by stage
  • Production
    388
  • Scaled
    166
  • Pilot
    65
  • Announced
    39
  • Paused or rolled back
    8

Finding 3: where the evidence is

Public deployments by industry of the deploying organization, with the share that disclosed an outcome, and by region.

Public deployments by industry
Public deployments by region
  • North America
    299
  • Europe
    189
  • Asia Pacific
    74
  • Global
    73
  • Latin America
    13
  • Middle East
    12
  • Africa
    6

Finding 4: the building blocks

The AI patterns use cases are built from, and how the EU AI Act classifies them.

The benchmarks with the most evidence

KPIs with at least three pooled public values across the whole library ("up to" values are listed on each KPI page, not pooled).

Benchmarks with the most public data points
KPIMedianReported rangeData pointsClaimed by
Productivity gain50%
8.7% to 113%
Not pooled: up to 90%, up to 87.5%, up to 80%
23plus 3 up to10 organization, 13 vendor
Cycle time reduction59%
20% to 95%
Not pooled: up to 70%, up to 30%
22plus 2 up to8 organization, 14 vendor
Accuracy91%
42% to 100%
Not pooled: up to 99%
21plus 1 up to11 organization, 10 vendor
Handling time reduction60%
8.2% to 97%
Not pooled: up to 50%, up to 20%, up to 5%
17plus 3 up to5 organization, 12 vendor
Containment rate68.5%
20% to 96%
1810 organization, 8 vendor
Automation rate76%
25% to 99%
Not pooled: up to 70%, up to 50%
15plus 2 up to8 organization, 7 vendor
Cost reduction45%
7% to 95%
Not pooled: up to 70%
10plus 1 up to4 organization, 6 vendor
Time saved per task7.5 minutes
2 minutes to 120 minutes
Not pooled: up to 225 minutes, up to 120 minutes, up to 18 minutes
8plus 3 up to5 organization, 3 vendor

About this report

Every figure is computed from the public records in the library at the date above, using the rules on the methodology page. The underlying data is available as open data under CC BY 4.0. The sample reflects what organizations choose to publish, which skews toward large organizations and successful projects.