Hours Saved or EBIT Impact? Choosing the AI ROI Metric That Survives Scrutiny

By the Wynstra team · · 7 min read

If you have read two AI ROI surveys this year, you have probably seen two different realities. One says most leaders are happy with their returns. Another says most CEOs got nothing. Both are true, because they are measuring different things. For anyone trying to justify, defend or scale an automation programme, the choice of metric matters as much as the result itself.

Why the numbers disagree

A widely cited MIT study found that 95% of generative AI pilots fail to deliver measurable ROI for companies, a failure rate rooted not in flawed models but in poor integration and misaligned priorities. Around the same time, a Wharton and GBK Collective study of over a thousand enterprise leaders found 80% report positive returns on their AI investments, though only 28% describe those returns as strongly positive. PwC's 2026 CEO survey sits closer to the pessimistic end: 56% of CEOs report neither increased revenue nor decreased costs from AI in the last 12 months, and only 12% report achieving both.

These are not contradictions so much as different yardsticks. Some surveys ask whether people feel AI is helping. Others ask whether a chief financial officer can point to a line in the P&L. McKinsey's 2026 research lands closer to the hard end of that spectrum: only 37% of companies see any EBIT impact from AI. If your business case uses the wrong layer of metric, it will either look better than it is to your board, or fail to show real value that exists.

The three layers of AI ROI

It helps to separate AI performance claims into three tiers, from easiest to measure and least convincing, through to hardest to measure and most convincing.

1. Activity metrics — logins, queries processed, documents handled. These show adoption, not value. A tool can be used constantly and still save nobody any money.

2. Efficiency metrics — hours saved, cycle time reduced, first-contact resolution. These are the default choice: hours saved is the most common ROI measure, cited by 40% of departments, while net-new revenue is the least used of the major measures, at 27%. They are useful, but a board will rightly ask what happened to those hours.

3. Financial metrics — cost removed, revenue influenced, EBIT contribution, payback period. This is where scrutiny lands, and where the shift in reporting is heading: a 2026 survey of 830 IT decision-makers found that direct financial impact, combining revenue growth and profitability, nearly doubled to 21.7% as the primary ROI metric for enterprise AI, while productivity gains collapsed 5.8 percentage points as the leading success metric over the same period.

Why hours saved alone will not survive an audit

Time saved is a real and useful proxy, but it only becomes ROI when it is converted into one of three outcomes: headcount that was not hired, overtime that was not paid, or capacity that was redeployed into higher-value work and produced more output. If none of these happens, the hours simply absorb back into the working day and the saving disappears from the numbers at year-end. When building a case around efficiency metrics, pair every hours-saved figure with a sentence answering: where did that time go, and can someone in finance confirm it.

Set payback expectations before you start

One reason AI programmes draw scrutiny is that sponsors often expect returns on the timeline of ordinary software. That is usually unrealistic. Research compiled from Deloitte's enterprise surveys found that a typical AI use case takes two to four years to pay back, against 7 to 12 months for other technology. Build this into the business case from day one rather than letting a slow first year look like failure. Where organisations do put a number on returns, the results can be strong: among early adopters tracked by Snowflake, those that quantified returns reported a 49% return in 2026, meaning about $1.49 back for every $1 spent. That figure only exists because those organisations measured in the first place.

A practical scorecard by stage

Use different metrics depending on how far a use case has progressed, rather than demanding financial proof from a two-week pilot.

  • Pilot stage: task success rate, time per task before and after, user adoption. Purpose is to confirm the workflow works, not to prove ROI yet.
  • Early production: hours saved translated into FTE-equivalent capacity, error rate, cost per transaction, cycle time against a baseline measured before go-live.
  • Scaled deployment: cost removed from the budget (not just time freed), revenue influenced where applicable, and EBIT or margin contribution reported alongside other capital projects, using the same discipline finance applies elsewhere.

A large 2026 executive survey found that organisations treating agentic AI this way see returns across the board: 94% of respondents report AI agents contribute to both cost savings and revenue, with 6% reporting no contribution, and 84% of executives said they are seeing increasing financial returns from AI initiatives, with 26% reporting that returns are accelerating year over year.

Governance decisions that correlate with real returns

Measurement discipline tends to travel with other organisational choices. The Wharton and GBK study found that at organisations with a newly designated chief AI officer, 58% of leaders report strongly positive ROI, dropping to 24% where oversight was folded into an existing executive's role, and to 8% where there is no such role and none planned. None of this guarantees results, but it is a reasonable signal that someone owning the metric, not just the technology, is what turns activity into value.

Conclusion

There is no single correct AI ROI number, but there is a correct discipline: know which layer of metric you are quoting, convert time saved into money saved before anyone else asks you to, and set payback expectations that match how long AI programmes actually take to earn their keep. Organisations that get this right are not necessarily using better models; they are simply measuring more honestly. If you are building or defending a business case for agents, workflow automation or document AI and want a second opinion on which metrics will hold up, Wynstra works with operations and healthcare leaders on exactly this kind of measurement design.

Sources

  • Ai Roi
  • Business Case
  • Automation Strategy
  • Ai Governance

Have a project in mind?

Tell us what you need. We'll reply within one business day with ideas, a rough timeline and an estimate. No obligation.