Productivity Fell From 23.8% to 18.0% as the Top AI ROI Metric. What Replaced It

Productivity has lost its place as the default way to justify AI spend. In Futurum's survey of 830 IT decision-makers, the share naming productivity as their primary AI ROI metric fell from 23.8% to 18.0% in six months. Revenue and profitability measures, counted together, now lead at 21.7%. The awkward part is that company earnings have not moved to match the new scoreboard.

The short answer on AI ROI metrics in 2026

Buyers have stopped accepting hours saved as the headline justification for AI. Financial metrics, meaning revenue growth and margin improvement, replaced productivity at the top of Futurum's ranking. Efficiency still sits close behind at 19.2%. The shift is real in what buyers say, small in what they fund, and invisible so far in reported earnings.

That gap is the subject of this post. A ranking of ROI metrics is a ranking of what people are willing to defend in a budget meeting. It is not evidence that AI started producing revenue. If you treat it as evidence, you will write a business case that sounds current and fails its first review.

My position is narrow and practical. Productivity was never a bad metric. It was an unfinished one, because almost nobody wrote down where the saved hours were supposed to go. The financial framing forces that missing step, which is why it is worth adopting even though the survey data behind the trend is softer than the headlines suggest.

AI ROI metrics before and after: the Futurum numbers

The source is the 1H 2026 Enterprise Software Decision Maker Survey, released by The Futurum Group on 17 February 2026. It covered 830 global IT decision-makers and compared results against the firm's 2H 2025 wave. The press release does not state fieldwork dates, and the full report sits behind a download form.

Here is every ROI metric the release gives a number for, with the prior figure where one exists.

Figures are from the Futurum press release only. The listed shares sum to 67.1% for the 1H 2026 wave, so roughly a third of primary answers went to options the release does not name.

What the analyst behind the survey argues

Keith Kirkpatrick, the Futurum research director quoted in the release, frames it as a maturity story. "The productivity argument was the right metric for the GenAI pilot phase, but the market has matured," he said. He added that sales teams leading with "save 4 hours per week" are "entering a losing conversation."

That second line is aimed at vendors, and it is the useful one for buyers too. If a pitch to your own finance team leads with hours saved per seat, it is now the weakest version of the argument you could make. The same release reports that 42.9% of respondents now prefer consumption-based pricing for GenAI features, which matters here because metered pricing makes cost per outcome visible in a way a flat seat fee never did.

Why "nearly doubled" needs care before you quote it

Start with the headline claim. The release says direct financial impact "nearly doubled to 21.7% of primary responses." It also says Futurum split the prior "overall financial performance" option into two separate measures, revenue growth and profitability. Both facts appear in the same release, and most secondary coverage repeats only the first.

Splitting one answer option into two changes how people answer. A respondent who would have picked productivity because "financial performance" felt too vague now sees "bottom-line profitability" written out plainly. Survey researchers treat answer wording as a known source of shift. Some of the 21.7% is therefore a measurement effect, and the release gives no way to size it.

The productivity decline does not have that problem to the same degree. The 23.8% and 18.0% refer to the same named option. Yet even that drop could partly reflect respondents moving to the two newly visible financial choices. The defensible reading is that productivity fell as a stated primary metric, and that financial framing rose by an amount that cannot be cleanly measured from the public release.

I would quote the 5.8-point productivity decline and avoid the word "doubled." It is the stronger fact, because it compares like with like. Some later write-ups also describe the change as a 5.8% decline, which is wrong. It is 5.8 percentage points, roughly a quarter of the starting share.

Which functions moved first on AI ROI metrics

The enterprise software survey has no department breakdown. A second Futurum study does, and it points in the same direction from a different angle. The 1H 2026 Digital Leadership and CIO survey, released on 6 March 2026, tracked 203 to 248 CIOs per quarter across three waves of 2025.

Among those CIOs, productivity as a desired AI outcome fell from 67.5% to 41.8% between the Q2 and Q4 waves. Innovation rose from 17.2% to 32.4%, and modernisation from 18.2% to 32.4%. "The generic efficiency argument for AI is dead," said Dion Hinchcliffe, the Futurum analyst quoted.

Sales and finance: the productivity-heavy functions

The same CIO survey shows sales department AI use falling from 68.0% to 49.6%. Sales was the showcase for hours-saved pitches: drafting emails, summarising calls, updating records. When a function's whole case rests on time saved, it is the first to lose budget once time saved stops counting.

Finance shows the pattern from the inside. A Gartner survey of 204 finance leaders in March 2026 found 45% of finance AI investment leans towards productivity, while just 20% of finance AI projects lean towards decision quality. Gartner's warning is specific: once a task becomes faster, "the benefit often plateaus" unless the efficiency changes a broader business decision.

Research, IT and the functions with a priced output

R&D use rose from 9.9% to 27.9% in the CIO data. That fits the innovation framing, because a new product has a revenue line and a saved hour does not. IT operations offers a second clue. In Gartner's September survey, intelligent IT asset and cost optimisation had the highest share of positive returns among IT use cases, at 40%.

The common thread is not the function. It is whether the work produces something with a price attached: a contract, a cancelled licence, a shipped feature. Functions with priced outputs moved to financial metrics first because they could. A broader version of this argument, applied to agent deployments, sits in the comparison of agent payback claims by function.

How the AI ROI metric shift squares with McKinsey's flat EBIT number

If buyers now demand financial impact, you would expect financial impact to appear somewhere. It has not, at least not in the largest recurring survey of the question. McKinsey's 10th State of AI report, The State of AI in 2026: On the Road to ROI, found 37% of respondents attribute at least some EBIT impact to AI. FM Magazine's summary describes that as "around the same as last year."

The high performer group did not grow either. Organisations attributing at least 5% of EBIT to AI and describing the impact as significant stayed at about 6% of respondents. Meanwhile 80% said AI has improved their individual productivity, and 50% said it helps them make better decisions.

Three readings of the same gap

The first reading is lag. Buyers changed their metric in late 2025 and early 2026, and earnings take quarters to reflect new priorities. On this view the EBIT figure should rise in the 2027 edition, and the metric shift is simply early.

The second reading is leakage. Individual time savings are real but get absorbed: longer lunches, more meetings, slightly better emails. Nothing reaches the income statement because nobody cut a cost or sold more. This is the mechanism Gartner describes as the plateau, and it is the one the evidence supports best. The macro version of the same puzzle is covered in the analysis of why AI productivity has not shown up in GDP.

The third reading is that the metric change is mostly rhetorical. Gartner's survey of 1,303 respondents, fielded January to April 2026, found productivity was still a key target outcome for 75% of functional leaders and took roughly 30% of functional AI spend. That is nearly twice the next objective. People now describe success in financial terms while still buying productivity tools.

I think the second and third readings are both true, and they reinforce each other. A team that justifies a copilot with "revenue impact" but deploys it exactly as before will get exactly the same leakage. Renaming the metric does not move the money.

Where this argument is weakest

Several things in this post rest on thinner evidence than the confident headlines around the Futurum survey imply. Here they are in order of seriousness.

The data is press releases, not datasets

Every Futurum figure here comes from a press release. The full reports sit behind registration, so the question wording, the response distribution and the fieldwork dates are not public. The two Futurum surveys also cover different people: 830 IT decision-makers in one, about 200 to 250 CIOs per wave in the other. Combining them as a single trend is a judgement, not a measurement. The McKinsey figures were read through FM Magazine's summary because the report PDF did not load during research, so the exact wording should be checked against the original.

There is a quieter conflict of interest too. Futurum is an analyst firm whose clients include the vendors being judged on these metrics. A finding that buyers now want P&L proof is useful to vendors selling P&L dashboards. That does not make it wrong. It does mean it deserves the same scrutiny you would apply to a vendor's own ROI case study.

The case that productivity is still the right metric

For some work, hours really are the output. A software team constrained by engineering capacity converts saved hours into shipped features fairly directly. A support team with a hiring freeze converts handling time into backlog that does not grow. In those settings, forcing a revenue attribution adds guesswork and invites gaming.

Financial metrics have their own failure mode. Revenue attribution for a tool used across a sales team is easy to inflate and hard to disprove. A business case that claims "AI-influenced pipeline" can look stronger than an honest productivity case while measuring far less. If the shift to financial metrics produces more of those, it will have made AI evaluation worse. The honest version of a productivity case beats a dishonest version of a financial one.

An AI business case template built on financial metrics

The fix is not to swap one word for another. It is to finish the chain that productivity cases leave open: hours saved, then what happens to those hours, then which line on the P&L moves, then when. The table below sets the weak version against the one that survives a finance review.

The one line most AI business cases leave out

The conversion step is where most cases quietly fail. "Saves 4 hours a week" becomes money only through one of three routes. You avoid a hire you would otherwise have made. You cut spend on contractors or another tool. Or you put the time into work that produces revenue, and you can name that work. If none of the three applies, the honest financial value of the productivity gain is close to zero, whatever the time saved.

Write the route into the case, with the person who owns it. A business case that says "the support lead will not backfill two open roles in Q1" can be checked in Q2. One that says "frees capacity for higher-value work" cannot. Gartner recommends finance teams measure "enterprise impact rather than just the number of pilots or hours saved." That is the same point in its own language.

Set the stopping rule before the start

Add a kill criterion on the first page: a number, a date, and what happens if the number is missed. Gartner's high performers, defined partly by tracking ROI constantly and discontinuing weak initiatives, reported positive returns on 81% of their AI initiatives. Low performers did not know the return on 29% of theirs. The practice of setting that threshold in advance is covered in the guide to kill criteria for AI projects, and the reasons teams avoid it in the piece on sunk-cost pilot funding.

One more check belongs in every case. Ask what the result would look like if the project failed. If the failure case and the success case would produce the same report, the metric is not measuring anything. That test catches most "AI-influenced revenue" claims before they reach a board.

Frequently asked questions

Where to start this week

Take the AI business case your team will renew next and find the sentence that claims time saved. Underneath it, write which of the three routes turns that time into money, and who owns it. If you cannot pick one, you have found the reason the renewal will be hard, and you have found it before finance does.

Then pull the baseline. If the case was approved on a vendor's benchmark, measure your own figure for the same unit of work now, and date it. A current number you own is worth more in a budget meeting than any ranking of AI ROI metrics, including the one this post was built on.

Related reading