The Automation Arbitrage Mistake

The Gartner numbers say most of those headcount cuts aren't paying for themselves.

The transformation business case has not changed much in fifteen years. A new technology arrives, a vendor frames the upside, and somewhere on the savings page, a headcount reduction line carries most of the value. The faces and the acronyms change. The shape of the slide does not.

This time the line item is AI. The savings are framed as automation efficiencies. The question on every delivery leader's desk is whether to commit to the number, defend it through the board, and steer the programme towards a finish line that has people on the other side of it.

A piece of recent data is worth your attention.

Earlier this month, Gartner published findings from a study of 350 executives at global enterprises with at least one billion dollars in annual revenue. Eight in ten of them had cut workers as part of their AI pilots. None of those cuts correlated with higher return on investment. Workforce reduction rates were almost identical between high-ROI and low-ROI companies. The firms reporting the strongest returns were the ones that used AI for what Gartner called people amplification, meaning the technology was deployed to enlarge what their existing workforce could do, not to take it apart.

If the Gartner numbers hold, most of the headcount reduction lines being defended this quarter are not paying for themselves. The savings are real on paper. The ROI is not real in practice.

Sam Altman, in a February interview, put it plainly. There is, he said, some AI washing happening, where companies are blaming AI for layoffs they would otherwise have made. It is a useful admission from the person with the most to gain from the opposite story. Some of what is being booked as AI savings is restructuring dressed in a more fashionable costume.

There is a longer pattern here, and it is the part delivery leaders should sit with.

Twenty years ago, the cost story was labour arbitrage. Move the work to a lower-cost geography, recover the margin. The Indian and Filipino BPO industries grew on the back of that logic, to six million and two million jobs respectively. What the early business cases consistently underestimated was the cost of judgment, escalation, cultural fluency, and the small acts of discretion that frontline staff perform without being asked. Those costs found their way back into the system through service degradation, churn, and quiet rework. The savings rarely held at the level the original deck promised.

Automation arbitrage is the same logic with a cleaner narrative. Move the work to a system that does not require wages, recover the margin. Gartner is already pointing to the same correction. Their forecast is that by 2027, half the companies that cut customer service staff because of AI will rehire for similar roles, under different titles. The leader who delivers the cut may well be the one who delivers the rehire.

This is a useful exercise in premeditatio malorum. Sit for a moment with the version of the programme where the savings do not arrive, the service degrades, and the rehire begins twelve months after the go-live celebration. What would you wish you had done differently at the business case stage? What conversations would you wish you had insisted on?

The Stoic distinction here is not abstract. The market narrative, the share price, the vendor's roadmap, none of that is in our gift. What is in our gift is the rigour with which we interrogate the business case, the structure of the change we design, and the duty of care we hold towards the people whose roles sit inside the savings line. Marcus Aurelius kept reminding himself in the Meditations that we become what we give our attention to. A programme is no different.

There is a version of this work that is craft. Build AI into how people do their jobs, amplify the parts of the role that are hardest to do well, give the workforce a path through the change rather than a queue out of it. The Gartner data says that version pays better. It is also, not coincidentally, the version that lets a delivery leader sleep at night.

The other version, the one with the optimistic savings line and the quiet redundancies, may not deliver the ROI that justified it. The number on the slide is not the same as the number in the bank.