Where Freed Capacity Goes
The AI saved the time. The report says so, in hours. Six months later, nobody can point to what those hours bought. They went somewhere. Something decided where.
What freed capacity is
Freed capacity is the time, attention or money that a change has released but not yet committed to an outcome.
Take a constructed example. An insurer's claims handlers now have AI that summarizes the file and drafts the decision letter. Each claim takes a handler less time. But claims still wait: for documents from the customer, and for a small group of senior adjusters who review complex claims for every line of business. Every storm season, a backlog builds.
The handlers' saved time is freed capacity. It is not a result yet. It is an input to a decision nobody has made.
Freed capacity rarely stays free for long. It turns into more claims, more polish on each letter, an earlier finish or a smaller budget.
Treat a reported saving as the most that was freed. People's estimates of the time they save can run ahead of what the clock shows, so start from what was measured.
What freed capacity can buy
The claims operation could settle claims sooner. It could clear the backlog with the same team. It could carry fewer handlers next year. It could put the time into the complex files that keep getting reopened. It could stop the Saturday overtime. It could fix the document request that makes customers send things twice. Or it could keep a buffer for the next storm.
Those seven outcomes are worth checking wherever capacity is freed.
| Outcome | It needs | You know it happened when |
|---|---|---|
| Faster delivery | The faster step is where the customer's elapsed time is spent | The customer's elapsed time falls, not just the task time |
| More output | Work waiting for the people who got faster | More work is finished with the same paid hours |
| Lower cost | A decision to cut hiring, overtime or contractor hours | Paid hours or spending fall, while customers get the same service and quality |
| Better quality | Time allocated to prevention and hard cases | Rework, reopened cases or defects fall |
| Relief | Protection that keeps new work from refilling the time | Overtime and after-hours work fall, and stay down |
| Capacity to improve | Time set aside, and credit for problems prevented | The next improvement goes into daily use |
| A buffer | Spare capacity kept at the step where peaks hit | Peaks pass with no backlog carried into the next week |
Outcomes that were not possible before are a different question: new capability.
What decides when nobody does
The claims operation never chose among the seven. Next year's budget already assumes fewer handlers, because the vendor's estimate of hours saved went straight into it. Nobody decided that this year. A budget rule did.
That rule is a default: whatever spends freed capacity when nobody chooses. Look for these:
- Budget and accounting rules turn the gain into a cost cut. Australia has long taken a fixed annual share of public service agency budgets as an "efficiency dividend." Many small agencies met it by cutting work, which a parliamentary committee called a false economy (Joint Committee of Public Accounts and Audit, 2008).
- Work refills the time. In Denmark, users of AI chatbots reported saving 2.8% of their work hours. About 80% of them put the time into other tasks, and their recorded hours and earnings did not change (Humlum and Vestergaard, 2025).
- Production pulls people back from improvement the moment problems ease. A manager on Du Pont's maintenance improvement program put it this way: "As soon as you get the problems down, people will be taken away from the effort and the problems will go back up" (Repenning and Sterman, 2001).
- Demand rises to fill capacity. A faster service attracts more work.
- Standards creep. The decision letter gets one more polish, because now there is time for it.
- Staffing tools turn shorter tasks into fewer scheduled hours. That is what they are built to do.
Most defaults are old decisions, made before this capacity existed. When nobody tracks the time, nobody sees them at work.
The outcomes compete, and some feed each other
Analog Devices ran a quality program that worked: yield doubled, cycle time halved and defects fell tenfold. The improvement freed capacity. The freed capacity led to layoffs of nearly 12%. After that, the program stalled (Sterman, Repenning and Kofman, 1997).
Analog took the freed capacity as cost savings and lost the improvement that had produced it. The same hour cannot be spent twice. An hour taken as savings cannot also fund improvement or sit in a buffer. At a busy step, it cannot shorten the wait either: if staffing is cut to match the speed-up, the step stays as loaded as before. Different hours can go to different outcomes.
Some outcomes feed each other. Better quality removes rework, and less rework frees more capacity.
Taking load off the senior adjusters helps the whole operation, because they serve every line of business. Suppose the handlers spend some of their freed time sending complete files to the adjusters. A complete file spares the adjusters the chasing of missing documents and a second review. Their load drops, and the wait falls in every line of business at once. That is how busy queues behave: when a step runs near full, a small cut in its load shortens the wait sharply.
Order matters too. Wiremold reinvested first. It pledged that no one would lose a job to improvement. When improvement freed people, Wiremold moved them to improvement teams and let sales growth absorb them. By the company's own account, work flowed faster, inventory fell, and sales and profit rose (Byrne, 1995). Protecting jobs alone is not enough: in the Analog study's simulations, it deepened the company's financial stress. Demand and pricing set limits too.
Make the decision
- Find out what was freed, and in what form. Minutes scattered across a team cannot buy what a block of usable hours can. The handlers' saving arrives as minutes on every claim, spread across the team.
- Find where it sits. Freed time at the step that limits the work can speed the work up. Freed time upstream of a queue only gets work into the queue sooner. The handlers sit upstream of the senior adjusters, so a complex claim the handlers finish sooner just waits longer for review.
- Name the default deciding today. Check the budget, the staffing model and where the time is going now.
- Choose the outcomes, their order and an owner for each. A choice nobody owns slides back to the default. One choice is to fund the next improvement first, such as fixing the document request, and spend what it frees on the one after. Mark which outcomes are cash-releasing, meaning a budget or paid hours fall; the rest, such as faster settlement, release no cash but are still benefits. If you cannot change the budget or staffing model, bring your findings from steps 1 to 3 to whoever can.
- Before you redirect the time, decide what evidence will count. What you are testing is the choice, not the AI. Name the measure that should move and a date to look again. For a baseline, measure the work after the AI arrived and before you redirect the time, or use the same season last year. For comparison, use teams that have the AI but have not yet redirected their freed time. If the hours go to the complex files, the measure is the share of claims reopened.
Sources and lineage
Where Freed Capacity Goes is part of Applied Enterprise AI, a body of work developed by Curtis Hibbs and Joshua Barnes. The outcomes, defaults and sequencing here are a synthesis to test, not a validated model. The reinvestment loop, in which each improvement funds the next, comes from Applied End-to-End Flow. The page builds on Turing Post's capacity-to-outcome chain and BCG's call for deliberate choices. It also draws on:
- Humlum and Vestergaard, Large Language Models, Small Labor Market Effects, BFI working paper 2025-56.
- Joint Committee of Public Accounts and Audit, Report 413: The Efficiency Dividend and Size in Australian Public Service Agencies (2008).
- Repenning and Sterman, Nobody Ever Gets Credit for Fixing Problems that Never Happened, California Management Review (2001).
- Sterman, Repenning and Kofman, Unanticipated Side Effects of Successful Quality Programs, Management Science (1997); working paper.
- Byrne (1995), How Wiremold Reinvented Itself Through Kaizen (Lean Enterprise Institute).
- Sarre et al., NIHR (2019), on the Productive Ward.
- Se and Schenk, AI Saved Time. Where Did the Value Go?, Turing Post #7 (2026).
- BCG, Making AI Productivity Pay Off (2026).