Key findings
Organisations with a named AI value owner in each business unit move pilots into production markedly faster than those run from a central lab.
High confidenceData readiness, not model choice, is the most cited reason pilots stall.
High confidenceSpending on AI governance tools is rising, but few firms have tested their controls against a real incident.
Moderate confidence
Context
Two years into the generative AI cycle, most large organisations have run dozens of pilots and put only a handful into daily use. This benchmark asks what the organisations with AI in production did differently, drawing on survey responses, interviews with operating teams and our client inquiry data.
What we recommend
- Appoint a business owner for each AI use case, accountable for a measured outcome rather than for delivery of the model.
- Fund data readiness as part of every AI business case, not as a separate programme.
- Run a tabletop exercise on an AI failure before scaling customer-facing use cases.
How we reached these conclusions
This benchmark draws on 214 sources, including structured interviews with practitioners, client inquiry data, vendor briefings checked against reference customers, and public disclosures. Vendor claims are never counted as evidence on their own. Each finding is graded in the Krynthnode confidence ledger, described on our methodology page, and the whole benchmark is due for review by 9 March 2027.