AI governance is becoming a real cost of putting AI into production, not just paperwork for the compliance folder. ITPro reports that controls added late can add months to a project, while human review and manual checks still do much of the work around AI-enabled processes.
Watch Desk analysis
What happened
In an article published on 29 October, ITPro spoke to analysts Enza Iannopollo of Forrester and David Menninger of ISG about the often-hidden costs of enterprise AI governance. They describe the workload: approvals, testing, audit trails, documentation and human review, spread across IT, security, legal, risk and data teams.
Menninger says ISG research found about one-third of 1,200 AI use cases had reached production, double the share from a year earlier. For use cases still to be completed, he says the average cost would be twice what had already been spent and the work would take more than a year. He also says 90% of AI-enabled work still involves people, either assisting with AI or reviewing and approving its output.
The article gives one measure of the tools gap: fewer than one-third of more than two dozen vendors evaluated met ISG’s criteria for agent evaluation tools, and fewer than one-fifth met its criteria for A/B testing. Gartner forecasts spending on AI governance platforms will rise from $275 million in 2026 to $462 million in 2027.
Why it matters
The business case for AI is often presented as a choice between paying for models and saving on labour. These figures point to a more complicated ledger: oversight takes time, projects can stall, and a rushed deployment may need costly rework. The governance bill is not necessarily the biggest one, but it can affect whether a pilot gets beyond the starting gate.
The practical advice from the analysts is to build controls into pilots, then scale them proportionately. Automate routine checks where possible and focus human review on higher-risk uses, rather than making every project queue for the same manual committee.
Our read
“Move fast and add governance later” is a fine way to make both the move and the paperwork slower. The useful point here is not that every AI project needs a new layer of bureaucracy. It is that testing controls early, and matching them to the risk, may save teams from retrofitting a system after it is already tangled into the business.
The numbers come from research cited by the analysts, not a universal price list for AI governance. But the direction is hard to ignore: production requires more than a promising pilot and a model bill.
What to watch
- Whether more AI pilots reach production as governance is built into projects earlier.
- Whether organisations automate routine checks without weakening review of higher-risk uses.
- Whether vendor tools close the gaps identified in ISG’s evaluation.
Discussion spark: Should companies build governance checks into every AI pilot from day one, or reserve the heavier controls for systems that reach production?
Sources and evidence
- Source update (29 October 2026, 08:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.