The enterprise AI ROI gap: why felt productivity is not proof

The headline research is blunt. A 2026 survey of 100 senior enterprise AI leaders found that more than two thirds of enterprises still rely on estimates rather than measured financial results to assess return on investment (Source: ModelOp, 2026 AI Governance Benchmark). Gartner forecast worldwide spending on generative AI at US$644 billion for 2025, most of it on servers, phones and PCs with AI built in (Source: Gartner, 31 March 2025).

Felt productivity is not a financial result#Link to this section

Here is why the distinction matters. Individual wins are real and show up quickly, but they rarely make it to the P&L. PwC's 29th Global CEO Survey of 4,454 CEOs across 95 countries found that 56% report seeing neither revenue gains nor cost reductions from their AI initiatives, and only 12% have achieved both increased revenue and reduced costs (Source: PwC, 29th Global CEO Survey, 2026). A separate 2026 survey from Plug and Play found that 74% of the world's largest enterprises now run AI in production, yet half of those production-stage companies cannot tell you whether any of it worked (Source: Plug and Play, 2026, reported by Forbes).

The market has already noticed the difference between activity and outcome. In the Futurum Group's 2026 Enterprise Software Survey of 830 IT decision-makers, direct financial impact, combining revenue growth and profitability, nearly doubled to 21.7% as the primary ROI metric for enterprise AI, while productivity gains fell away as the leading success metric (Source: Futurum Group, 2026 Enterprise Software Survey). Boards and CFOs are asking AI to connect to the P&L, not just reclaim hours.

Why estimates keep standing in for evidence#Link to this section

Estimates persist because the easy metrics look like proof. A dashboard that shows seats logging in and prompts climbing tells you people are busy, and that is worth knowing. Whether a dollar of margin moved is the question around it.

There is also a structural blind spot in how impact gets measured. Comparing patterns across teams can show where AI appears to go with better results, but correlation is not attribution. Felt gains and correlated gains are comfortable to report, and neither is a return you can defend in a budget review.

The consequences show up when someone has to own the number. In Open Future Forum's August 2026 finance data, 72% of CEO and founder respondents expect a measurable return on an AI investment inside six months, while among finance respondents the figure is 45%, a 27-point gap between the seat that signs for AI and the seat that has to prove it (Source: Open Future Forum, August 2026).

What measurement discipline looks like#Link to this section

Three habits help most. Set a baseline before deployment, model the fully loaded cost rather than just the licence fee, and match the right payback model to each use case rather than applying one blanket metric everywhere.

So what should a leader actually track? Start with a clean baseline for the specific outcome you expect the tool to move, captured before rollout so you have a real before and after. Track adoption as it is, meaning who is actually using the tool and who quietly stopped, because paid seats that go dark are pure waste. Track perceived time saved and, crucially, where that reclaimed time goes, since hours freed and then lost to other tasks never reach the P&L. Then tie each of those signals back to the fully loaded cost of the tool, not the sticker price.

What does good look like? Good is a number a CFO can follow line by line, with its limits stated, where the benefit outweighs the total cost, adoption is broad and sustained rather than concentrated in a handful of power users, and the outcome you named up front is the outcome that moved. That is what The GAiGE is built to show. Short pulses, anonymous by default, show who uses each tool, the time people say it saves and the licences nobody uses, beside what each tool costs, so you can see which tools are worth keeping.

Proof is likely to decide which AI budgets survive the next planning cycle, and measured evidence makes a stronger case there than estimates do.