Will AI replace the IT service desk?

Nobody has ever hung up the phone and said, "I really enjoyed holding for Tier 2."

The traditional IT service desk, with its queues, ticket numbers, and "have you tried restarting?", is headed for retirement. The people who staff it are not. If that sounds like wishful thinking, consider the bank teller.

The ATM was supposed to end the teller

When ATMs spread in the 1970s and 80s, the prediction was obvious: machines hand out cash, so tellers are finished. What happened was stranger. For roughly thirty years, the number of bank tellers in the United States went up.

ATMs made branches cheaper to run, so banks opened more of them. And once the machine was counting out twenties, tellers spent their time on the work a machine couldn't do: solving problems, giving advice, and building relationships with customers.

The machine took the transaction. The humans got the interesting part.

AI is the ATM of IT support

Most service desk contacts are some version of "How do I...?" How do I reset my password, get into the shared drive, or connect to the VPN from the Chicago office?

An AI assistant on your phone or laptop can now answer those in plain language. The good ones skip the explanation and just do it for you. They're available at 2 a.m., they give the same answer every time, and they have never once lost a sticky note.

The payoff is practical: fewer tickets, faster fixes, and support staff who get to spend their day on problems that need a brain.

Now the bad news

An assistant is only as good as the knowledge behind it. Point one at a knowledge base that was last updated when BlackBerrys were cool, and it will give your users the wrong answer faster, and with total confidence.

I learned this the unglamorous way. Years ago I took over a reactive help desk on a federal program, and we raised first call resolution to 92 percent. There was no AI involved. We fixed the knowledge and the process so the first person to answer could solve the problem.

AI multiplies that kind of foundation. It also multiplies a mess.

And it doesn't own outcomes. On that same program we cut incidents by 70 percent by hunting down root causes and fixing them. An assistant can spot the pattern. A person still has to decide to do something about it.

Where to start

1. Clean up your top 20 requests. Find what drives the most contacts and give each one a single correct, current answer.
2. Go where people already are. Put the assistant in Teams, Slack, or whatever your staff live in. Nobody wants another portal.
3. Set the boundaries. Decide what the assistant may do on its own, what it may only explain, and what data it may see.
4. Measure from day one. Track how many issues it resolves alone, how users rate it, and how long fixes take.
5. Keep the path to a human short. When the assistant is stuck, the handoff should be instant and should bring the conversation with it.

The teller is still there

The service desk isn't dying. It's getting the ATM treatment: the routine goes to the machine, and the people move on to work that's worth their time. Your users get answers in seconds, and your staff stop answering the same question forty times a week.

A shorter version of this article first appeared on LinkedIn. Join the discussion there.