AI saved time. Finance couldn’t count it.

At a glance

Organization

Large professional-services organization; affected population of 1,200 professionals in knowledge work.

Situation

AI tools successfully deployed, usage increasing, and productivity reported to leadership primarily as time saved.

Complication

Time saved is not financial value unless the released hour becomes revenue, avoided labor, throughput or reduced external spend. The dashboard made no such distinction — and mixed a 50-week hour count with an annual per-professional value figure.

Entry point

Transformation Oversight into Zero-Based Transformation™. SORT, PROVE and IMPROVE were the most visible stages.

Result

288,000 hours of gross released capacity → 100,800 hours economically converted → $8.0M of underwritten annual economic value, at a 35% modeled conversion rate.

The adoption was real

This case does not involve a failed deployment. The tools worked, they were rolled out competently, and usage was climbing rather than stalling after the launch window. Professionals were completing research, drafting, analysis, summarization and administrative work faster than before, and they said so.

Leadership was right to treat that as a result. Where it went wrong was in the translation: estimated hours saved were multiplied by a blended cost rate, and the product was reported as value on the same dashboard as cost reductions and revenue wins.

That single arithmetic step converted an operational improvement into a financial claim — and committed the organization to a number the P&L had never agreed to produce.

The unresolved question: what happens to the saved hour?

An hour released from a task does not become money by being released. It becomes money only if it turns into something the business can count: incremental revenue, avoided hiring, increased throughput, reduced external spend, or capacity redeployed to work that generates one of those.

Absent that conversion, the released hour is still worth having. It may reduce overtime pressure, improve quality, shorten a cycle, or make a role sustainable. Those are legitimate benefits and they belong in the story. They do not belong in a benefits column that a CFO will later be asked to reconcile.

Is time saved actually economic value?

The distinction matters most at the point of reporting. A dashboard that shows $22.8M of AI value creates an expectation, and expectations set at that level are difficult to withdraw later without the entire program losing credibility — including the parts that genuinely converted.

Why a time-saved dashboard misleads

Three assumptions are usually buried in the multiplication. First, that the saved time is uniformly distributed — when in practice some roles release hours that are immediately sellable and others release hours with no external demand behind them. Second, that the cost rate applied represents money the organization can actually stop spending. Third, that the operating model can absorb the released capacity where it appears.

The third assumption is the one that usually fails. Capacity released in a role with no queue of chargeable work in front of it is not convertible this year, however real the time saving is. Converting it requires a change to how work is routed and allocated — which is an operating-model change, not a tooling change.

This is also why decision rights matter more here than in most technology programs. If it is not clear who may reallocate released capacity, nobody does, and the hour is absorbed silently.

Capability, usage and value are three measurements

The work was to separate the chain that a single dashboard number collapses, and then underwrite only the portion with a credible conversion route in the operating model as it currently ran.

The chain, stated in full

Technical capability, then usage, then time released, then capacity converted, then economic outcome, then evidence. Each link can be healthy while the next one fails, and only the last two are financial. A program reporting at link three and claiming at link five is the pattern this case corrects.

The gross opportunity, stated honestly

Rather than dispute the productivity benchmark, it was applied openly and on one consistent annual basis. The published research reports approximately five hours saved per professional per week for the legal professionals it surveyed, which it annualizes to nearly 240 hours — so 1,200 professionals gives 288,000 gross released hours a year. At the same source’s figure of roughly $19,000 of annual value per professional, the gross opportunity is $22.8M. Both figures now sit on the same annual basis, which the original dashboard did not: it mixed a 50-week hour count with the published annual value figure.

The underwritten share

Of that gross opportunity, 35% was underwritten as economically convertible in the current operating model. That rate is an underwriting judgment, not an observed result: it reflects the share of released hours sitting in roles with chargeable demand in front of them or with an avoidable cost attached. The 20 / 10 / 5 split across the three categories is a modeled allocation on the same basis.

Critically, the three categories are not valued with one blended dollar-per-hour. Each is converted through its own economics: billable conversion at realized rate and contribution margin, avoided capacity cost at actual labor cost per avoided full-time equivalent, and redeployed capacity at attributable output and margin. A single benchmark rate applied to all released hours is the error the original dashboard made.

The remainder, reported as capacity

The unconverted balance was reported separately as released capacity, with named owners and its own decisions about where it should go. It was neither hidden nor counted.

From a gross opportunity to a recognizable number

Classification: the 288,000 hours and $22.8M are gross opportunity derived from a published benchmark. The three conversion categories are underwritten, with recognition evidence defined per category — timesheet-and-invoice evidence for billable conversion, approved requisitions withdrawn for avoided cost, and attributable output for redeployment, which remains the most model-dependent of the three. The remaining 187,200 hours were not lost: they may produce employee, quality, cycle-time or innovation benefit, and they simply should not be reported as P&L value.

Two numbers, reported separately

Leadership stopped reporting a single AI value figure. Economic conversion went to Finance with recognition evidence defined per category. Released capacity was reported as capacity, in hours, against the roles where it had appeared — not converted to money.

That second report turned out to be the more useful of the two. It made visible where capacity was accumulating with no demand in front of it — which is a routing and allocation problem the organization could act on, and the route by which next year’s conversion rate rises above 35%.

Decision rights over released capacity were assigned explicitly, because an hour nobody is empowered to reallocate is an hour that will be absorbed.

The results

Of 288,000 gross released hours annually, 100,800 were economically converted, producing $8.0M of underwritten annual economic value against a $22.8M gross opportunity.

Reported as $22.8M, the program would have carried a $14.8M expectation it could not satisfy. Reported as $8.0M with 187,200 hours of capacity named alongside it, the program carried a number Finance could reconcile and an agenda for the following year.

Why the result held

Because only convertible capacity was ever committed, the reported value did not have to be revised downward as the operating model caught up. Each of the three conversion categories had a recognition definition agreed with Finance before it was claimed — and the redeployment category, being the most model-dependent, was reported as underwritten rather than realized until output could be attributed.

The distance between benefit potential and benefit realization is the whole of this case. Naming both numbers, rather than collapsing them into one, is what made the smaller number defensible.

What this means if you are reporting AI value

If a single AI value figure is on its way to a board pack, the question to ask before it lands is which portion of it Finance has agreed it can see.

Evidence basis

External research establishes both the scale of released time and the reason it does not convert on its own. It is used here as the input to the gross opportunity and as context for the operating-model constraint — not as evidence of the converted result.

• Thomson Reuters Institute, Future of Professionals 2025: Mind the gap (2,275 respondents, fielded February–March 2025). Respondents expect AI to save an average of five hours per week; the report states that legal professionals are expected to free up nearly 240 hours a year, up from 200 in 2024, unlocking an average annual value of approximately $19,000 per professional. Applied here as the input to the gross opportunity, on the understanding that the surveyed population differs from this case’s.
• IBM Institute for Business Value, The AI-human operating model (2026). Reports widespread ambiguity in decision rights around AI-enabled work, and argues that performance depends on the operating model built around the technology rather than the technology alone. Cited as context for the operating-model constraint, not for the conversion rate.