This year, I worked with lawyers at a large industrial group who built their own AI agents for recurring legal reviews.
The first one used to fill in a review sheet by hand for contracts of more than a hundred pages. Two days of work, by her own estimate. With the agent, two to three hours.
The second one spent more than a week, and not alone, on a report that comes back every six months. The agent now produces the groundwork in a few minutes. Human review time remains, and she does not delegate it.
These are their own estimates, not a measurement. But they share something that matters: a precise task, chosen by the business, whose cost the teams knew before they started.
That is exactly what most AI projects lack.
Pressure is rising, proof is missing
On 19 October, Gartner published its 2027 CIO and Technology Executive Agenda survey of 2,002 CIOs and technology executives. Three figures sum up the situation.
- 58% of CIOs face increased pressure to deliver cost savings from AI.
- 51% expect total cost of ownership to rise.
- 13% say their enterprise has realised significant or transformational value from AI.
So AI is being asked to cut costs, while its own cost is rising and its value is rarely demonstrated.
The Deloitte AI Institute, in a survey of 3,500 executives published on 29 October, shows where attention goes: 71% of companies track their total inference and API spend, 51% track token volume, 44% track per-unit inference cost. Technical cost is tracked. Business gain, much less.
Why value stays invisible
Activity gets measured. Number of people trained, active users, prompts sent. Training teams is useful, and so is tracking usage. None of these figures says whether a task costs less than before.
The task was never chosen. A project launched on « AI for legal » or « AI for finance » has no point of comparison. A project launched on « the review sheet for contracts over a hundred pages » has one.
The cost is incomplete. The licence is visible. Team time to build, test and review is far less so, as are change management and support. A gain compared with a licence alone always looks good.
Weighing full cost against real gain
The method fits on one page. It is filled in with the business before launch, then updated on the decision date.
- The task: which one, in which team, takes too long today or generates too many errors?
- The indicator: processing time, rework rate, cases per person. Only one, chosen by the business.
- The baseline: how long the task takes today, and how many times a month.
- The full cost: licences, team time to build and test, change management, support.
- The real gain: time saved, human review included, multiplied by how often the task occurs.
- The thresholds: the gain that justifies scaling, and the one below which you stop.
- The decision date and the decision-maker: a named business owner who decides on the planned date.
The fifth point makes the difference. An agent that produces a report in a few minutes but needs a day of checking does not deliver the same gain as one whose output can be reviewed in an hour. Review time is part of the cost.
Sorting a portfolio of projects
According to Gartner, a use case has roughly a one in five chance of succeeding. A CIO who has launched ten projects should therefore expect to keep a small share of them. The question is which ones, based on facts.
With the sheet, sorting becomes possible: keep the projects whose real gain exceeds their full cost, adjust those that come close, stop the others. Gartner shows the exercise is still rare: 31% of CIOs have criteria to decide that a pilot will not move into production, and 17% to stop a low-value initiative already deployed.
Stopping a project that does not prove its value is not a failure. It frees the budget for one that creates value.
IT cannot prove it alone
The pressure for savings lands on the CIO. But value shows up in the business: only the teams who do the task know what it costs and what has changed.
That is what the two lawyers at the start show. The agents with the most traction are the ones the business builds and tests on its own work. Gartner also recommends involving business leaders, who are willing to walk away from low-value investments. The message from Gartner analyst Alicia Mullery fits in three words: « Volume is not value. »
Where to start
If you do only one thing this month: list your AI projects in progress and, for each one, look for three pieces of information. The target task, its baseline, its full cost. Those missing one are your first priorities.
I help business leaders and CIOs put this measurement in place and sort their AI projects, with the teams who do the work. Let’s talk about your projects.
Sources
- Gartner Survey Finds Only 13% of CIOs Say Their Enterprise Has Realized Significant or Transformational Value from AI, Gartner, 19 October 2026
- Gartner Survey Finds Only 13% of CIOs Report Significant or Transformational Value from AI, Digital Terminal, 29 October 2026
- Enterprise AI Enters an ROI Reckoning: More Use, Uneven Value, Deloitte AI Institute, 29 October 2026