What happened

Reuters reported on 21 August 2026 that Swiss National Bank governing board member Petra Tschudin had discussed AI’s uncertain effect on inflation in an interview with Finanz und Wirtschaft.

Her argument allowed for effects in both directions. Productivity improvements could lower some prices, while investment shifts and shortages could create upward pressure in the short or medium term. The SNB subsequently published the interview on its website.

This was a conditional assessment, not a prediction that AI would necessarily cause higher inflation. For business planning, it is a useful reminder that a promising long-term capability can arrive with substantial immediate costs.

Why it matters

An AI proposal often compares the cost of a subscription with the value of staff time it might save. That comparison can omit much of the work required to make the system useful.

A company may need to prepare information, connect applications, train reviewers and maintain a process for correcting errors. A department may also run the old and new methods together while it establishes reliability. These costs can arrive before the expected benefit.

I would ask project sponsors to show the timing, not only the total. Which costs occur before launch? Which continue with every transaction? When should a business outcome begin to improve? Who carries the extra work during the transition?

This makes a proposal easier to manage without pretending that every minute saved immediately becomes cash in the bank.

The bigger shift

The distinction between technical efficiency and business economics deserves a place in every AI review. A task can become cheaper to perform while the organisation chooses to perform many more of those tasks. It can also become faster while creating additional checking elsewhere.

Consider a hypothetical sales team that generates more tailored proposals. The commercial benefit depends on whether those proposals reach appropriate buyers, remain accurate and improve conversion. Producing more documents could simply increase review time if the targeting is weak.

Set a useful unit of value. For that team, it might be the cost of an approved proposal for a qualified opportunity, followed by its commercial outcome. Track total spending as well as the unit cost so that rising volume remains visible.

Build scenarios around uncertain supplier costs and adoption levels. These scenarios should help set a budget and review point, rather than masquerade as forecasts.

The discipline fits naturally within an outcome-led AI strategy, where the operating change matters as much as the tool.

My take

I would welcome productivity gains and still challenge an AI budget that assumes costs will fall automatically.

Ask for a before-and-after view of the whole process. Include supervision, rework and the cost of maintaining the information the system uses. Separate capacity released from costs actually removed, and explain how that capacity will be used.

Give the project a realistic transition allowance, then test its assumptions at a defined checkpoint. If the economics disappoint, investigate whether the problem is model choice, process design, usage volume or an unrealistic original promise.

The useful executive stance is neither automatic optimism nor automatic scepticism. It is a willingness to examine where value appears, when it appears and what the organisation must spend to obtain it.