← Back to Insights

Insight

Saved Hours Spoil

Ariel Agor
Updated
Saved hours spoil unless the roadmap claims them. Companies measure the hour but rarely what it became. Write the capture plan first. Where unclaimed hours go, Name the destination, Price the spoilage, Move the boxes, The one-page capture plan.

Key facts

FactValueSourceAs of
Share of measuring managers who track hours saved as an AI metric (Wharton/GBK)40%beri.net
Share of measuring managers who track net-new revenue from AI-enabled products27%beri.net
CEOs who say AI gains are often absorbed before reaching the bottom line (EY)23%ey.com
U.S. CEOs who say AI agents are transformational for their organization (KPMG)9%kpmg.com
Pre-built AI agents and assets IBM brings to Imagine Enterprisemore than 6,000newsroom.ibm.com

Listen · Read by Ariel

0:00 / n/a

An hour your team saves with AI on Monday will be gone by Friday. Nobody steals it. It gets spent on a longer review, a nicer slide, a meeting that grew to fill the gap, or a second check of something the model already checked. Saved time behaves like fresh produce. If nobody picks it up, it rots where it sits, and by the end of the quarter you can't find any trace that it ever existed.

This is the part of the AI transformation roadmap for executives that nobody draws. The usual roadmap is a sequence of deployments: pilot, scale, agents, autonomy. Each box is a thing you install. None of the boxes says where the freed hours are supposed to go, who collects them, or what happens to the budget line they used to sit on. October 2026 gave us four large CEO and manager surveys in about three weeks, and read together they describe one long, expensive spoilage.

What the October numbers actually say

Start with KPMG. Its 2026 CEO Outlook, released on October 15, 2026, surveyed more than 1,300 CEOs worldwide, 400 of them in the United States. Among the U.S. group, 84% rank AI as a top investment priority over the next three years, and 65% expect to spend at least 10% of their capital budget on AI in the next year. Then comes the number that should stop a board meeting. Only 9% say AI agents are transformational for their organization. Another 39% report significant efficiency or growth gains from agents, and 43% report moderate impacts from targeted use cases.

The EY-Parthenon CEO survey, released in London on October 1, 2026, covered 1,200 CEOs across 21 countries, fielded in August and September. It holds the sentence I keep returning to. 23% of CEOs say AI-driven gains are often absorbed before they reach bottom-line results. Half of them name AI as the largest single contributor to productivity gains over the past 12 months. Only 16% have clear, real-time visibility into AI ROI. Andrea Guerzoni, EY-Parthenon Global Vice Chair, put the whole problem in seven words: "AI is creating real productivity gains, but productivity alone is not a strategy."

On October 13, 2026, Wharton Human-AI Research and GBK Collective published their fourth annual adoption study, "The Adaptive Advantage." They interviewed 1,416 people at U.S. companies with at least 1,000 employees and more than $50 million in revenue, between July 9 and July 29, 2026. Among the 941 managers who measure AI ROI at least occasionally, hours saved per employee per week is the most common metric at 40%. Net-new revenue from AI-enabled products is the least used of the eight measures, at 27%.

Then Gartner, on October 20, 2026, reported that 44% of CEOs struggle to demonstrate AI financial returns to stakeholders, while they keep raising technology spend anyway.

Put these side by side and a pattern appears. Companies measure the hour. They rarely measure what the hour became. The CEOs who say their gains are "absorbed" are describing the exact moment the hour became nothing.

Where an unclaimed hour goes

I have watched this happen enough times to know the routes. There are only a few, and they are boring, and that is why they win.

The first route is review. Brian Smith, who runs the AI, Cloud and Enterprise Technology practice at GBK Collective, described it in the Wharton report: "Most companies are sitting in a supervised middle, paying for autonomy and then reviewing every action." The same study found that two-thirds of departments say they use AI agents, and only 17% let an agent complete a task without a person approving each step. So the agent drafts the claim letter in forty seconds and a senior adjuster reads it for eleven minutes, because nobody changed the rule that says a senior adjuster signs every claim letter. The hour was saved upstream and spent downstream. The dashboard reports the first half.

The second route is polish. When producing a first draft gets cheap, people make more drafts and fuss over them longer. The memo that used to go out at "good enough" now goes through four versions because four versions are free. Quality rises a little, nobody can price the rise, and the time is gone.

The third route is coordination. Freed calendars attract meetings the way empty shelves attract clutter. A manager with six more open hours a week will get invited to six more hours of things, and will say yes, because saying yes feels like being useful.

The fourth route is the quietest. The saved hour simply becomes slack, and slack hides inside the existing headcount. KPMG found that 89% of U.S. CEOs expect headcount to grow over the next three years. I'm not arguing for layoffs. I'm pointing out that hiring plans and productivity claims are being written by different people who never compare notes. When both numbers go up at once and margins stay flat, the hours went somewhere, and it was not to the P&L.

Each of these routes is rational for the person taking it. Nobody is lazy. The adjuster is following the signing rule. The analyst is doing better work. The manager is being a team player. The spoilage is a design failure, and design belongs to the executive team.

The AI transformation roadmap for executives starts at the end

Here is my position, stated plainly so you can disagree with it. The first artifact of an AI transformation roadmap should be a capture plan: a written decision, per workflow, about what the freed capacity becomes, signed by the person who owns the budget it comes out of. Deployment comes second. If you cannot write the capture plan, you are not ready to deploy, because whatever you deploy will feed the four routes above.

This flips the usual order. Most roadmaps go technology first, then process, then a value measurement at the end. By the time the measurement arrives, the hours have already been absorbed, and the measurement politely reports that productivity rose while results did not. That is the EY finding, reproduced at company scale.

Name the destination before the deployment

Every freed hour can only go to a short list of places. You can stop paying for it. You can move it to work that earns revenue. You can sell it as faster service or lower prices to customers who will pay for that. Or you can deliberately bank it as slack for resilience, which is a fine choice if you make it on purpose and know what it costs.

The Wharton data shows why the destination matters more than the tool. Among leaders tracking at least one growth measure, 35% report strongly positive ROI, against 13% of those tracking efficiency alone. Wharton's Sonny Tambe said it with some dryness: "Companies believe AI is going to transform their industries, but many are still asking it to prove itself by saving time." The 35% group did not buy better software. They had already decided the time was supposed to turn into money, so they built a meter that could see whether it did.

Suppose a claims team of 40 people saves an average of five hours each per week with a drafting agent. That is 200 hours a week, about five full-time jobs' worth. The capture plan has to say, before launch, which of those destinations those 200 hours go to. Perhaps 120 go to a backlog of subrogation recoveries that the team never had time to chase, and the plan names the dollar target. Perhaps 80 are banked for the January surge. What the plan cannot say is "the team will be more productive." That sentence is how hours go bad.

Price the spoilage

A capture plan needs a clock. In the illustration above, suppose the subrogation work does not start within six weeks of the agent going live. At that point the 120 hours have already been eaten by review and polish, and you will not get them back by announcing a new priority. Habits set fast. Once the senior adjuster has spent a month reading every agent draft, that reading is now part of her job, and she will defend it in your next planning cycle.

So the roadmap should treat each deployment like a perishable shipment with a delivery date. If the receiving process is not ready, delay the shipment. This sounds slow. It is much faster than the alternative, which is deploying everywhere, getting absorbed everywhere, and spending 2027 trying to explain to your board why the 9% number did not move.

Move the boxes on the org chart

The most striking line in the Wharton report is easy to miss. Among departments reporting strongly positive ROI, 64% changed their management structure because of AI, versus 47% of the others, and they report no more staff reductions or hiring freezes than anyone else. They redrew the chart. They did not shrink it.

That fits the capture argument exactly. The review route survives because the approval chain survives. If every agent output still flows to the same approver it flowed to in 2024, the approver will absorb the gain, every time. Changing the structure means deciding which approvals are retired, which become sampling (say, one in twenty drafts checked rather than all of them), and which become exception handling that only triggers on a flag. Wharton found that 24% of departments that monitor agents automatically have transformed their workflows, against 9% of those that do not. Automatic monitoring is what lets you retire the human read on every item without losing sleep.

The same study found that organizations with a newly designated chief AI officer reported strongly positive ROI at 58%, against 8% with no such role. I'd read that gap carefully, since it is self-reported and correlational. My guess is that the title itself does very little. What matters is having one person whose job includes asking where the hours went and who is senior enough to retire an approval step that a division head is attached to.

Why the vendors will not draw this part for you

On November 10, 2026, IBM and Microsoft announced Imagine Enterprise, a joint platform for building, running and governing AI agents in regulated industries. IBM Consulting brings more than 6,000 pre-built AI agents and assets, Microsoft brings Foundry, Fabric, Entra, Purview and Defender, and forward-deployed engineers from both companies help qualifying customers move from experimentation to production. The first industry frameworks are financial services, insurance and retail. Mohamad Ali, head of IBM Consulting, said: "Clients are no longer looking for AI that simply helps people do the same work a little faster."

I agree with Ali. I also notice that 6,000 agents is a supply of capacity. It says nothing about demand for that capacity inside your company. A forward-deployed engineer can get an agent into production. They cannot decide that your underwriting committee now meets monthly instead of weekly, or that the regional finance team's close process will lose two sign-offs, or that the recovered hours in claims are owed to a revenue target with someone's name on it. Those are decisions about your power structure, and no vendor contract reaches that far. If a vendor tried, you would rightly throw them out of the room.

This is why buying more agents faster tends to make the spoilage worse. Each new deployment frees more hours into an organization that has no plan for receiving them. KPMG found that 88% of U.S. CEOs prioritize realized AI value over strict token budgets, which is the right instinct. But realized value is a property of the receiving organization, and token spend is a property of the sending technology. You can spend without limit on the sending side and realize nothing on the receiving side.

The capture plan, written out

Here is what I ask leadership teams to produce before any new AI workflow goes live. It fits on one page per workflow, and the discipline is in refusing to launch until the page is complete.

The page names the workflow and its current owner. It states the baseline in hours and in money, measured before the tool arrives, because measuring after the tool arrives gives you a baseline that has already been absorbed. It names the destination for the freed capacity, with a number attached: dollars of new revenue, dollars of avoided spend, a service level sold to a named customer segment, or hours explicitly banked as slack. It lists the approval steps that will be retired or converted to sampling, and the date each one changes. It sets the spoilage clock, the date by which the destination work must be running or the deployment pauses. And it names the person who reports on all of this at the quarterly review, someone who has the budget authority to make the retirements happen.

Two things about this page tend to surprise people. First, the technology appears almost nowhere on it. The model, the vendor and the agent framework are footnotes. Second, the hardest line to fill in is almost always the retired approvals. Executives will happily commit to revenue targets they have no plan to hit, but they flinch at removing a sign-off, because every sign-off belongs to someone who will notice.

That flinch is the real roadmap. Wherever your team cannot agree to retire an approval, the hours from that workflow will be absorbed, and you can predict the absorption with some confidence before you spend a dollar.

What this looks like at the portfolio level

Zoom out from the single workflow and the capture plan turns into a capital allocation question. EY found that 48% of leaders are reinvesting extra capacity into innovation and transformation, while only 15% say productivity gains are being reinvested in workforce development. "Reinvesting capacity into innovation" is the phrase I would interrogate hardest in your own company. In my experience it often means the hours went to a dozen small internal projects, each one reasonable, none with a revenue line, all of which will be quietly cancelled in the next budget cycle. Innovation was the story told after the fact about where the absorbed hours went.

The honest version of the portfolio view is a ledger with two columns. The left column lists hours freed by AI across the company, workflow by workflow, as measured. The right column lists where they landed, in money. The gap between the columns is your spoilage rate. Most companies have never seen this ledger because nobody built it, and the four October surveys suggest the gap is large. Gartner's own advice in the October 20 release points the same way: measure unit economics, decision velocity, speed to market and customer outcomes, rather than activity.

Say your ledger shows 30,000 freed hours a quarter and $2 million of traceable value against them. That is a number your board can act on. They can ask which workflows spoil the most, whether to pause new deployments there, and which approvals to retire first. Without the ledger, the board sees AI spend rising, productivity claims rising and margins flat, and draws the conclusion that AI does not work. That conclusion is wrong, but it is the conclusion your own data invites.

The roadmap is an argument about power

Every approval step in your company exists because someone, at some point, did not trust the person before them. AI shifts the economics of trust. When drafting is free, checking becomes the cost, and the checkers are usually senior, expensive and politically strong. A roadmap that sequences tools and ignores checkers will be absorbed by the checkers, through no malice of theirs, because their job description still says check everything.

This is why I tell clients that an AI transformation is an organizational redesign that happens to use software. The 9% of U.S. CEOs who told KPMG their agents are transformational did not get there by buying more agents than the other 91%. If the Wharton data is any guide, they changed who approves what, they tracked growth as well as hours, and they gave someone the authority to say where the time goes.

You can buy the agents from IBM and Microsoft, or from OpenAI, Anthropic or Google, and they will work. What you cannot buy is the decision about where the hours go, because that decision means taking authority away from people inside your own building, and only you can do that. If you hand the roadmap to a vendor, you will get a very good sending system wired into a receiving system that was built for 2019. The hours will arrive, sit on the dock and spoil.

Architecting the capture is the work. It means writing the one-page plans, building the ledger, retiring the approvals, and holding the spoilage clock when a division head asks for one more quarter of full review. It is uncomfortable, it is political, and it will decide whether your AI spend shows up in 2027's margins or in 2027's excuses. Agor AI Advisory does this work with leadership teams: we draw the receiving side first, we name the retirements, and we stay until the ledger closes. The hours your teams freed this quarter are already going bad.

Sources

The Hour Gets Measured. What It Became Does Not.

This verifies the post's central inference, that saved hours are tracked far more often than the value they turn into, by laying out the survey evidence and its limits in order. After 15 seconds the reader should see that the spoilage claim is an inference from several surveys, and that the one resolution with data behind it is naming the destination for the hours before deployment.

  • Companies measure the hour. They rarely measure what the hour became.
  • The spoilage is inferred from the pattern across surveys, not read off any single number.
  • 35% strongly positive ROI with a growth measure versus 13% with efficiency alone: the destination matters more than the tool.
Artifact type trace is staged but not yet rendered. The data is captured in the sidecar JSON.

Source: Figures taken from the post's cited October-November 2026 sources: Wharton/GBK 'The Adaptive Advantage' (via BERI and Knowledge at Wharton), the EY-Parthenon CEO survey, and the KPMG US 2026 CEO Outlook. · verified · as of 2026-11-15

Want this kind of automation working for your business?

Agor AI designs and ships the systems these posts describe, scoped in weeks, not quarters.

Book a Free Strategy Call