This morning the voice at the drive-through donut shop followed a simple script. It asked for our donut order, how many, whether we wanted anything else, and then prompted the price. It was a female voice. Instantly I knew it was AI. At the window, it was a human. Actually two humans. One was a supervisor.

That small interaction stuck with me because it may be a glimpse into where customer support and business operations are heading. The AI wasn’t trying to replace the entire experience. It was handling a repetitive part of the transaction: take the order, confirm the quantity, ask the next question, move the customer forward. The human was still there to confirm the order, take payment, and hand over the glazed donuts.

That’s a much more interesting use of AI than simply asking whether artificial intelligence will replace people.

The better question is: which parts of the process actually require a person?

Every organization has repetitive work. Someone answers the same questions. Someone transfers information from one system to another. Someone checks whether a form is complete. Someone schedules the appointment, sends the reminder, routes the request, prepares the report, or asks the same five questions every day.

For years, we’ve accepted much of this as simply part of doing business. It doesn’t have to be. A well-designed system can automate the predictable parts of the process while keeping people where judgment, relationships, creativity, accountability, or simply a human interaction matter. That’s different from automating everything.

I’ve spent a lot of time thinking about back-office systems because the objective isn’t technology for technology’s sake. The objective is to remove unnecessary friction. If a machine can reliably perform a repetitive task, let the machine perform it. Then let people spend their time doing work that actually requires people.

There is an economic consequence to this as well.

Every repetitive administrative task consumes labor. Labor consumes time. Time has a cost. As organizations grow, small inefficiencies compound into payroll, overhead, slower response times, mistakes, and additional layers of administration.

Automation changes that equation. But good automation shouldn’t make the customer feel like they’ve entered a maze designed to prevent them from reaching a person. We’ve all experienced those systems. If you’re a veteran and ever had to call the VA or state licensing department, you know exactly what I’m referring to. Press one. Press two. Repeat yourself. Get transferred. Start over. That’s not better operations. That’s simply transferring the company’s inefficiency to the customer. Government demands more taxes, services suffer.

The donut shop experience was different.

The automated system handled a narrow, predictable transaction. Then there was a person at the window. Maybe that’s the better model. Automate the repetition. Keep the judgment, the relationship, the human.

Oh, and if you’re wondering which donut shop. It was Krispy Kreme. More about the AI launch and issues it faced here.