From 2022, Starbucks added equipment and ran its stores with fewer people. Then, in April 2025, it halted the rollout and started hiring baristas again. “We over-rotated on the idea of equipment,” its chief executive, Brian Niccol, told the Financial Times.

Starbucks had concluded that the person behind the counter was part of what customers were buying. Its US sales have grown since, although hiring baristas was only one of the changes it made.

Alex Imas, an economist at the University of Chicago, has an explanation for why Starbucks might be right. In What will be scarce? he argues that when AI makes routine work cheap, spending moves to the work where it matters who is doing it.

If he’s right, the human part of what a service business sells is worth more than most of us charge for it. Nobody has proved that yet, so the sensible move is to pick the moments where a person might be the product and find out whether customers will pay. Most of us have spent years doing the opposite.

Where the money goes when things get cheap

Imas’s starting point is that when something gets cheap, the spending goes somewhere else. In 1900 about 40% of American workers farmed; today it’s under 2%. Cheaper food didn’t leave people idle. They spent what they saved on things they’d never been able to afford, and the jobs followed the money. He expects AI to do the same to cognitive work. Once the advice, the draft and the analysis cost next to nothing, more of our spending will go where it matters who’s doing the work — the physio who remembers last month’s injury, the tutor who knows why a child freezes in exams, the chef cooking in your kitchen.

He calls this the relational sector, and his list runs from nurses, therapists and teachers to fitness instructors, tailors, live performers and a great deal of hospitality and care work. Those jobs already employ nearly 50 million people in the US. That’s the size of it today, and Imas expects it to grow from here.

He’s careful about the risks. If automation outruns people’s ability to find new work, the demand collapse the pessimists fear is still possible. Not everyone agrees with him either: Philip Trammell, responding to the essay, argues that new products could just as easily pull spending away from human-made things. I find Imas more persuasive, but nobody knows, and it’s sensible to plan as though either could happen.

What the evidence does and doesn’t show

The neatest piece of evidence comes from research Imas did with Graelin Mandel. They asked people to bid on art prints that varied in how much AI was involved in making them, and in whether the buyer would own the only copy or one of many. Exclusivity added 44% to the value of human-made art but only 21% to AI-made art. It’s a narrow result about prints, but it suggests that when we pay extra for something rare, part of what we’re paying for is the person who made it.

A Gartner survey of nearly 6,000 customers in 2024 found that 64% would prefer companies didn’t use AI for customer service at all. Years spent in marketing taught me to take preference surveys with a pinch of salt, because what people say they want and what they’ll pay for are often different things. It tells you where to look, not what to charge.

And Starbucks is one company’s bet. By Sherwood News’s count, its average US company-operated store went from 26.8 staff to 19.8 between 2022 and the end of its 2024 financial year. Sales have recovered since the reversal: like-for-like US sales rose 6.5% in the year to September 2026. But the company changed plenty else over the same period, pricing and its store estate included, so the figure can’t tell us how much the extra baristas contributed. All three point the same way, but none proves that customers in general will pay more for a person. Whether yours will is something you’ll have to find out for yourself.

What this means if you sell a service

Pricing is one lever small businesses control, and it matters here. The relational work Britain already has — care, hospitality, tutoring — is where wages have been held down longest, a point I made in May in response to Ezra Klein’s piece on the same essay. A premium only reaches the people delivering the service if a business charges for it and passes some of it on.

Picture a small accountancy practice. AI is making the bookkeeping and the year-end return cheaper to produce, and clients will rightly expect some of that saving in the fee. What may become more valuable is the partner who rings in January to say a tax bill is coming and talks through how to fund it. Today that call is a favour buried in the fixed fee; it could just as well be a service tier with a name and a price that some clients choose and others don’t.

In an earlier piece on difficulty moats I described relationships as one of the advantages that hold up when expertise gets cheap. Imas’s argument takes that a step further. If the relationship is part of the product, it belongs on the price list, where customers can see it.

Five interactions and one question

The exercise I’d suggest takes an afternoon, and it works best with someone who deals with customers every day.

  1. List your five most important customer-facing interactions, such as the first enquiry, the quote, the onboarding call, the complaint or the annual review.
  2. For each, ask whether customers would pay more if a person they trusted handled it. Many won’t care who sends the invoice, and that’s useful to know.
  3. Test the likely ones with a price attached. Offer the human version as a named option to a handful of customers and see who takes it, because a yes backed by money is worth far more than a yes in a survey.
  4. Put your best people where customers pay for them, and automate the rest, with a person still accountable wherever judgement is needed.

Some businesses will find two or three interactions where customers will pay for a person. Some will find none, and that’s worth knowing too. The displacement story treats the economy as a fixed pie that machines take slices from. If Imas is right, the pie changes shape as it grows, and the slice that grows is the one with a person in it. For those of us who sell services, that’s the very slice we’ve spent twenty years trimming.