How businesses can turn AI productivity gains into capacity for growth and stronger judgement.
When AI reduces the effort needed to get work done, I want to know what the business intends to do with the capacity it gains. More demand may be waiting. There may be work worth doing that nobody has had time for. The team may simply need a less exhausting week.
Each is a reasonable destination. Leaving the choice to whoever fills the calendar first seems a poor investment policy.
In Building Judgement as AI Changes the Work, I asked how people develop experience when AI changes the tasks they used to learn through. This follows that question into the business: how do we use recovered capacity to improve delivery while helping more people exercise sound judgement?
Measure the Capacity AI Frees Up
AI can reduce routine preparation and help people compare information. Those gains matter. I would assess them across a function or connected set of workflows, where we can see what happens to the work after one activity gets faster.
Separate the effort needed to complete a task from the time it spends waiting. Include checking and corrections. Compare work of similar difficulty, so a quieter week doesn't become an AI success story.
Anthropic's January 2026 Economic Index examines AI use at task level, including complexity and the autonomy given to AI. That is a useful way to examine work within a role. Its measures of success are inferred from conversation records, however. They do not independently establish what happened in the business afterwards.
Small savings scattered across a week may also need work to be reorganised before someone can use them. Twenty minutes between meetings is useful. It isn't automatically twenty minutes available for supervised practice.
Decide Where the Gain Should Go
I'd make the allocation explicit, based on what the function needs:
- Meet demand: use capacity to complete more worthwhile work at the required standard.
- Improve delivery: address recurring errors or delays that consume effort elsewhere.
- Develop capability: give colleagues practice in decisions that currently depend on a few experienced people.
- Reduce overload: create breathing room where sustained pressure is damaging the work.
These choices can coexist. The mix should reflect the business, rather than a fixed percentage assigned to learning or a default increase in everyone's output target.
A business may have demand it cannot meet because its experienced people are already stretched. Making their experience easier to draw on could help colleagues take on more demanding work, with support available for decisions that need it. I'd assess whether that creates capacity for growth while maintaining quality.
Put Experience Into the Work
Consider a hypothetical operations team preparing for a seasonal rise in demand. It needs to decide how to handle the additional work within its budget and service commitments.
AI could help assemble the workload picture and compare possible responses. An experienced manager might know that certain work requires specialist skills, or that a previous staffing arrangement struggled under similar conditions. Those details could change an otherwise reasonable recommendation.
I would break the work down like this:
- Define the need. Agree which service commitments must be protected and which constraints are fixed. Don't let an assumed objective, such as minimising staffing cost, quietly become the only objective.
- Build the evidence. Assemble demand and capacity information. Check its currency and identify important gaps. Ask experienced colleagues for context that could change the decision.
- Compare the options. Have AI assist with the comparison, making assumptions visible. Test what happens if demand differs from the forecast or required skills are unavailable.
- Make the judgement. Ask a colleague to recommend a response and explain the trade-off. The experienced manager challenges the reasoning where needed. Previously agreed decisions can remain with the team; unfamiliar or consequential exceptions need escalation.
- Retain the useful lesson. Record what changed the recommendation and why. Have the people involved check that account before it becomes guidance for others.
- Check the result. Compare what happened with the intended outcome. Use the result to revise the guidance, rather than preserve a recommendation merely because it was approved.
The opportunity extends beyond completing that planning exercise sooner. On a later case, AI could retrieve the relevant precedent and prompt a colleague to check whether the same conditions apply. Someone who previously needed help framing the problem may be able to bring a better-supported recommendation to review.
That is where I would invest some of the recovered time: working through consequential cases with experienced people, including shadowing live work. AI could help retain and reuse the lessons. People still need opportunities to practise applying them.
Make Experience Go Further
Experienced colleagues should be able to focus on the decisions that need their judgement. AI could organise the supporting evidence and highlight unresolved assumptions, reducing the work needed to understand a case. A conversational interface could let them question an assumption and inspect the source behind it.
The test is whether this helps them catch what matters. A smooth summary that hides a missing constraint has saved nobody much trouble.
Reusable guidance needs checking too. An apparently sensible change to instructions can affect many later outputs. Anthropic's evaluation guidance distinguishes testing new capability from checking that changes preserve behaviour that already worked. It also recommends calibrating AI assessments against expert human judgement. I'd apply that discipline to any system intended to support business decisions. Whether a particular approach improves judgement remains something to test in practice.
Give the Investment a Result to Answer For
For the operations example, I'd assess whether the team can handle additional demand at the required standard, alongside total effort and rework. I'd also examine how much help from experienced colleagues comparable decisions require, and whether people recognise exceptions that need that help. Fewer questions alone could mean misplaced confidence.
Choose a function or connected set of workflows. Establish the starting position, then make an explicit decision about where recovered capacity will go. Assign responsibility for that decision and agree when to review the result. Include the effort spent maintaining the AI and its guidance.
For an initial estimate, the AI Business Case Builder can help explore potential capacity and investment costs. Those assumptions still need testing against the work.
My earlier article argued that learning needs a place in the design of work. Recovered capacity gives us an opportunity to make that place, while improving what the business can deliver.
AI efficiency gains need an investment plan. I'd expect that plan to say what we will do with people's time, and how we will know it was time well spent.
Tushar Singh is Head of AI and Data at ELab AI, which advises organisations on AI enablement and governance across New Zealand, Australia and Hong Kong.