Swedbank cuts about 100 Baltic jobs and asks the question every AI business case skips: what will tokens cost later?

Swedbank will reduce its staff by nearly 500 across the group, ERR reported on 27 October. About 100 of the roles are in Estonia, Latvia and Lithuania, where the bank employs more than 6,000 people. The rest are in Sweden. According to ERR, the group's first-quarter report already projected full-time staff falling from about 17,350 to about 16,800 by the end of 2027.

This round has less to do with AI than the headlines suggest. Olavi Lepp, who heads Swedbank in Estonia, tied it to the end of a period of heavy IT and data investment made when profits were unusually high. With profitability back to normal, the bank needs fewer people for that work. Some can retrain for other roles, he said, but not everyone wants to, so some jobs go. Swedbank Estonia still earned 193 million euros in the first nine months of the year, 2 million more than a year earlier.

Where AI comes in

The AI part is about what comes next. The group is testing AI in more than 90 applications. In Estonia, Lepp said, the first goal is to do more business with the same people, and lower costs are the second benefit. His logic for staffing is plain: if tokens cost less than employing people for a task, layoffs or retraining will probably follow. He also said AI already outperforms people on many tasks that follow formal logic.

Then he added the caveat most AI business cases leave out. AI companies still have to meet their investors' profit expectations, and nobody knows what software licences and tokens will cost once vendors start charging full price.

Why the price question matters to you

A business case that compares today's token price with a salary assumes that price holds. Dismissing a team is hard to reverse. Raising a vendor's price is easy. If you remove the people who know how a process works and the tool then doubles in price, you have neither the savings nor the skills.

Swedbank's sequence is worth copying: test widely, use AI to grow output with the people you have, offer retraining first, and treat cost cutting as the second step.

What to do on Monday

• ✳For each AI use case in your plans, write down what it costs per month today and what it would cost at two and three times that price. If the case only works at today's price, it is not ready to replace anyone
• ✳Track actual AI spend per process, not just licence counts. Token use grows quietly as more people and agents use a tool
• ✳Before you cut a role, document what that person knows about the process and keep at least one person who can run it without the tool
• ✳Offer retraining before reductions and say so openly. Swedbank did, and accepted that not everyone will take it
• ✳Keep your contracts and data portable, so a price rise from one vendor means a switch rather than a crisis

Swedbank can absorb a wrong guess about AI pricing. A 50-person company that has already let its back office go may not.