Five Percent for AI: Why the Order of the Cuts Matters

Note: This story was written with AI assistance and edited by me.

On October 22, Barclays announced it will cut about 5% of its workforce, close to 5,000 jobs, mostly in support functions, as it moves more operations to AI systems. The next morning, American Express said it will eliminate 5% of its global workforce. Based on roughly 76,800 employees at the end of 2025, that is about 3,840 roles. CEO Stephen Squeri said the company needs "to become a more agile and efficient organization." Reuters counts Barclays among a long list of companies linking layoffs to AI this year, from HSBC to Kroger.

Those are the facts as reported. Everything below is my opinion.

My view: cutting headcount in the name of AI before the AI is doing the work is backwards. Redesign the work, prove the tools carry it, redeploy the people who know the process, and only then decide what roles you no longer need.

A round number is a budget, not a plan

Five percent is a clean figure for an earnings call. It is rarely what the work actually requires. When a cut is sized from the top down, the people who leave are often the ones who understand the exceptions, the workarounds and the customers who do not fit the standard flow. Those are exactly the cases AI handles worst.

If the automation is not ready, the remaining team absorbs the gap, service quality slips and the savings show up on paper long before they show up in operations.

Where AI really changes the workforce

In my experience, AI rarely removes a whole job on day one. It removes tasks: drafting, summarizing, first-pass review, routine tickets. That frees time, and the smart move is to decide in advance what that time is for.

Some of it should go to work that was always understaffed, like quality, security and customer follow-up. Some of it should become real savings. The order matters.

What leaders can take from this

Map the work before you map the cuts. List the tasks AI will take over and test it on live volume.

Keep the process experts. Move them into the roles that supervise, correct and improve the AI. They are your quality control.

Measure the outcome, not the headcount. Track error rates, cycle times and customer satisfaction before and after. If those slip, the cut was premature.

A fair counterpoint

To be fair, large banks and card issuers carry heavy support costs, and both companies may have tested these tools for months before announcing anything. Some roles really will disappear, and saying so early can be more honest than a slow drip of cuts. My view is only that the announcement should follow the proof, not lead it.

What I have seen when cuts come before the change

Over the years I have watched teams shrink on a schedule that was set before anyone knew how the new process would work. The pattern is familiar. The people who leave first are often the experienced ones who can find another job quickly. The people who stay inherit their work, plus the job of teaching the new system everything the experts used to know.

The projects that went well did it the other way around. We ran the new process alongside the old one, let the experienced staff break it and fix it, and only reduced the team once the numbers held for a few months. It was slower to announce, but it rarely had to be undone.

Takeaway

AI is changing tasks faster than it is changing whole jobs. Leaders who redesign the work first keep their customers happy and their best people engaged. I took a firm from $6M to $60M in under three years, and every major change we made held up better when we proved it before we scaled it.

If you are planning an AI rollout, write down which tasks it will take over and how you will measure quality before and after. Make that list before you make any staffing decision.