BusinessGPSOwn the capacity to change.

Gartner estimates that cutting 2% of a 60,000-person workforce, then rehiring 30% of them, could cost nearly $100 million. Ask what an AI saving costs to keep

A saving booked by replacing people with AI stays open after go-live. In an organisation of 60,000 employees, Gartner estimates that cutting 2% of staff and rehiring 30% of them could cost nearly $100 million. So before such a saving counts as permanent, ask one question. What would it cost if some of that work came back?

Gartner published the estimate on 16 November 2026, in a release on the risks facing workforces in 2027. Gartner sells research and advice to the leaders it writes for. The $100 million is a model estimate on Gartner’s own worked example, not a measurement. It is about cutting staff and hiring them back, not about any one go-live.

Start with the savings that depend on people leaving. One list, worked through properly, shows how much of the case rests on work the AI must now do alone.

Gartner’s example prices the round trip of cutting and rehiring

Gartner’s release of 16 November sets out one worked example. It takes an organisation with 60,000 employees. Gartner estimates that reducing the workforce by 2%, then rehiring 30% of those people, could cost nearly $100 million.

Here is the arithmetic behind that example, which is ours. Cutting 2% of 60,000 removes 1,200 roles. Rehiring 30% of those 1,200 brings back 360 people. Gartner estimates that round trip could cost nearly $100 million.

The same release carries a prediction. Gartner predicts that 30% of employees who were terminated and replaced by AI will be rehired by 2029. It is a forecast, not a result.

The estimate prices cutting and rehiring. The prediction says some people replaced by AI will be hired back. For a sponsor, both point one way: some of the work removed with AI may return, and the return has a price.

A business case can book the saving before the work has gone

We think the risk sits in the timing of the business case.

A business case books a saving from roles removed on the day the roles go. That day is often close to go-live. The plan shows lower headcount, and the saving starts.

Whether the AI really does all of that work is learned later. It shows in the months after go-live, as volumes rise and unusual cases arrive. Some work the AI settles well. Some it passes back to a person. Some it gets wrong, and a person puts it right.

Volumes matter too. A tool tested in a quiet month may meet its first year-end, or its first audit, long after the roles have gone.

If enough work comes back, it returns in one of 3 forms:

Each of these lands in a different budget, months after the saving was counted. No single line shows it as the cost of the saving. So a sponsor can see the saving on the plan and miss the cost that undoes part of it.

An invented example shows the shape of the question

Every figure here is invented, to show the shape. Use your own.

A business case counts the saving from 40 roles in accounts payable and customer billing. The roles are removed once an AI tool is live. The saving is booked for the full year.

By month 6, exceptions the tool cannot settle are building up. The team that stayed works late. A backlog of supplier queries grows.

The finance lead agrees to bring back 10 roles’ worth of work. Some comes back as new hires. Some comes back as contractors while the hiring runs.

So the case counted a saving for 40 roles. The saving kept is for 30, less the cost of hiring and training 10 people. It also carries the months of contractors in between.

The tool may still save a great deal. What the example shows is that the saving on the plan and the saving kept can differ. And the difference arrives after go-live, when the programme has moved on.

The sponsor is the one person who sees both halves

The programme sees the go-live and the roles removed. It often disbands soon after. Operations sees the backlog and the overtime. Finance sees the hiring budget and the contractor invoices, in different months and on different lines.

Only the sponsor answers for the business case as a whole. That makes the sponsor the one person placed to ask whether the saving held. It also makes the sponsor the person the board will ask, if headcount climbs back.

Asking early is cheap. A list of the savings that rest on roles removed takes a programme lead an afternoon. Finding out a year later, from the hiring budget, takes far longer and leaves less room to act.

Ask which savings assume roles removed once AI is live

Ask your programme lead for one list, in writing.

  1. Each saving in the business case that assumes roles removed once AI is live.
  2. For each one, the work those roles did, and what does that work now.
  3. What it would cost if some of that work came back, as hires, contractors or overtime.

Ask for the source of each line. The saving comes from the business case. The cost of work coming back will be an estimate, and the sheet should say whose.

Rough figures will do. Where the programme lead cannot say what now does the work, the saving rests on an assumption nobody has checked since go-live.

If you signed the business case, start with the savings you signed.

If an integrator or a supplier built the AI tool, ask the programme lead to get their view of what it handles today. The cost of work coming back is still your own.

If the list shows no saving that depends on roles removed, the case is clear of this risk on this point. If it shows several, you know where to look before the next budget round.

Done when: one sheet lists each saving that rests on roles removed once AI is live. Beside each are the work those roles did and a cost if some came back.

One question tests whether the saving will hold

Put it to the programme lead in these words:

Which savings in our business case assume roles removed once AI is live? And what would it cost if some of that work came back?

Ask it of the live programme now. Then ask it of each new business case before a saving from AI is counted as permanent.

If the answer comes back as a list with sources, you know where the saving rests on roles removed. If it comes back as a pause, the checking starts there.

What comes next

This issue’s long read turns to the people who stay after go-live. It asks who owns the load of change on them.

The BusinessGPS Baseline report grades each step of how an organisation runs change as red, amber or green. To see a finished one, open the BusinessGPS Baseline report, run on BusinessGPS itself.

Sources

Every edition

Get After Go-Live Notes by email, every week

The weekly email arrives on the day After Go-Live publishes. It brings the week’s page, a link to the long read, and a link to our own Baseline report. We may also send BusinessGPS news and updates as separate emails, when we have them. Nothing else.

TechShifts Media Limited, trading as BusinessGPS, will email you to confirm your address. Nothing is sent until you do. Our email tool sends the email for us. Every email has a link to stop it. Your address is used for After Go-Live Notes and BusinessGPS news and updates only. What is kept, and for how long, is in the privacy notice.