Deployment persistence

Do enterprise AI deployments last?

Vendor case studies never tell you what happened next. This tracks the durability of public evidence — whether a deployment's source still resolves, or the same organization shows up in later AI work — across the cohort first documented in 2023.

Still referenced

97%

of 2023 deployments
Org expanded AI

66.9%

same org, newer evidence
Lost footprint

3%

source no longer reachable
Coverage

66%

166 of 251 judged
Data updated 12 minutes ago

Of the 166 enterprise AI deployments first documented in 2023 that we have standing to judge, 97% are still publicly referenced in 2026 — 66.9% because the organization has been documented doing more AI since, and 30.1% because the original source is still live. Only 3% have lost their public footprint.

Outcome breakdown

What happened to 2023 deployments

Share of the 166 judgeable 2023 cases by outcome. Cases with no checkable source are excluded, not counted as either standing or gone.

Organization expanded66.9%

The same organization is documented in newer AI deployment evidence.

Original source still live30.1%

The case's cited source still resolves; no newer evidence needed.

Lost public footprint3%

The cited source no longer resolves and the organization has no newer case. Not a claim the system was discontinued.

ⓘ How persistence is measured

Still publicly referenced means the original source still resolves, or the same organization has been re-documented in a later case. It is a measure of evidence durability, not whether the system is still in production.

Only cases at least a year old are judged. The denominator excludes cases with no recently-checked source — "we never looked" is never counted as either survival or failure, which is why coverage (66%) is reported alongside the headline. A low reappearance share is never read as decay: reappearance proves continuation, its absence proves nothing.

Cohort 2023 · 166 of 251 cases judged · 85 unverified · as of 2026.

Adoption maturity

Which use cases are adopted deeply, not just widely?

Adoption volume vs. the share of each use case's deployments from organizations documented running multiple AI projects · Click a dot for its deep dive.

Workflow automation is the most-documented use case (135 deployments) yet one of the shallowest — only 36.3% come from multi-project AI adopters. predictive maintenance, by contrast, is a durable staple at 53.3%.

0%25%50%057114171228corpus average · 40.3% repeat adoptersDurable staples: Higher Documented deployments (adoption volume) and higher Share of deployments from multi-project AI adopters.Durable staplesNiche but sticky: Lower Documented deployments (adoption volume) and higher Share of deployments from multi-project AI adopters.Niche but stickyWidely tried, shallow: Higher Documented deployments (adoption volume) and lower Share of deployments from multi-project AI adopters.Widely tried, shallowExperimental: Lower Documented deployments (adoption volume) and lower Share of deployments from multi-project AI adopters.ExperimentalCustomer service automationAI platformPredictive maintenanceWorkflow automationClaims automationAgriculture optimizationRisk assessmentPatient engagementClinical documentationCompliance automationAutomotive operations automationIntelligent document processingDocument automationCloud migrationAI development platformBusiness process automationCustomer support automationCustomer personalizationEnergy operations automationDocumented deployments (adoption volume)Share of deployments from multi-project AI adopters

+1 labels hidden for clarity. Hover or focus a dot for details.

A "repeat adopter" is an organization documented in 2+ AI deployments anywhere in the corpus; the axis is the share of a type's deployments from such organizations — a proxy for durable, program-level adoption vs one-off experiments (public-evidence persistence itself is near-universal, so it can't separate types). Showing the 20 highest-volume types. Volume split at the plotted median (77).