Production is up, ROI is flat: enterprise AI's discipline gap
Domino Data Lab's fifth annual enterprise survey finds 93 percent of organisations report improved AI production capability, yet for 57 percent the returns still fail to outpace spend, exactly as in 2025.
TL;DR
- Domino Data Lab's fifth annual survey of 639 senior enterprise AI leaders finds 93 percent reporting improved AI production capability in 2026, up from 88 percent.
- Yet 57 percent say AI ROI still fails to outpace spend, a share unchanged since 2025.
- Getting models into production, the constraint everyone spent three years fixing, turns out not to have been the binding one.
For years, the standard explanation for disappointing enterprise AI returns was operational: models were stuck in notebooks, pilots never reached production, infrastructure was immature. The fifth annual Domino Enterprise AI Report, released this week, quietly retires that explanation. Production capability is now nearly universal, 93 percent and rising. The share of enterprises whose returns fail to outpace their spend has not moved at all: 57 percent, exactly where it stood in 2025.
When the bottleneck moves and the number doesn't
| Domino 2026 survey (n = 639 AI leaders) | 2025 | 2026 |
|---|---|---|
| Report improved AI production capability | 88 percent | 93 percent |
| AI ROI fails to outpace spend | 57 percent | 57 percent |
Two curves, one climbing and one flat, are the cleanest evidence yet that deployment and value are different problems. This is the gen AI paradox in its 2026 form. The 2023 constraint was capability, the 2024 constraint was productionisation, and both have been substantially solved. What remains is the part that was never technical: choosing workflows where AI changes a business outcome, measuring against a baseline, and stopping the projects that do not clear the bar.
The flat 57 percent also hints at a portfolio problem. Spend has grown with capability, so holding the same failure share means the absolute gap between cost and return has widened. More production capacity applied to the same undisciplined portfolio produces more polished deployments of the wrong things.
The discipline gap
What separates the 43 percent whose returns do outpace spend is, in our experience, rarely better models. It is three unglamorous habits. They define value before deployment, in operating terms a CFO would sign. They instrument the baseline first, because a saving you cannot compare is a story, not a result. And they run AI as a portfolio with kill criteria, cutting losers early and reallocating to the workflows that compound.
Closing your own gap
- Audit the portfolio against one question: which deployments have a measured baseline and a P&L owner? Anything with neither is a pilot, whatever it is called.
- Reframe the executive dashboard from activity metrics, models shipped and seats licensed, to outcome metrics per workflow.
- Make stopping normal. A fixed review that retires the bottom slice of projects each quarter converts capability growth into return growth.
Production was the price of entry. The 57 percent shows that entry was never the game.
Related reading
Source: Fifth Annual Domino Enterprise AI Report, Domino Data Lab.