Attrition: AI in the contact center without laying anyone off

I called about something trivial

A few days ago I called a company I am a customer of about something trivial. I had to wait, and I could not understand why it could not be done online. While I was on hold I thought a voice agent would have handled it much better.

When we finished, the agent on the line said: "Give me a good review if you want us to keep having people answer your calls." He had reached the same conclusion I had.

For the normal case, the AI agent is better

For a routine request with no incident and nothing unusual about it, a voice agent is better than a person, no doubt about it. It is faster, there is no hold time, and it will usually know more and solve the problems the company creates for its own customers more consistently.

In other cases affected by these cuts it is less clear. When there is a real incident, or a customer who does not fit any process, the person still does something the agent does not do well.

Attrition does the work

When a steering committee asks me how to bring AI agents into customer service, I have given the same advice since 2025: do not lay anyone off, and cover every natural departure with the agent instead of a new hire. A contact center loses people every month, and every new hire needs three or four weeks on the learning curve before they are productive.

The agent is trained once and works from the next day, and when someone leaves there is no need to go through those weeks again. If every departure is covered this way, headcount comes down on its own, with no conflict with the people who stay and no severance. The line I use is always the same: don't fire anyone, use attrition and get more done with fewer people.

You never get to 100%

In the projects I have worked on, this approach replaces between half and two thirds of the seats in about six months. The exact pace depends on each contact center's attrition rate, and that is the first number I ask for.

You never get to 100%. The agent is very good at processes and quite bad at people, and every contact center keeps a share of work that belongs to people.

When the plan fails, you do it in a day

Attrition is a method for a company that works and wants to be more productive. When the plan fails and the company has no future, the method is the opposite. In February 2023 we let go of 100% of the staff of a portfolio company on the same day, after getting a shareholder's approval that morning.

Ben Horowitz draws the line well in The Hard Thing About Hard Things (2014), a book I have been recommending for years: there is a peacetime CEO and a wartime CEO. Attrition is a peacetime method, and a committee needs to know which of the two it is in before choosing it.

Higher-value work is possible and unlikely

What I don't buy is the line that comes with almost every one of these projects: that AI will free us to focus on the things that add more value. In a given company it can happen, but it is not what will happen in most of them.

The savings are the reason the committee approves the project, and in a contact center there is no higher-value job waiting for the person who stops taking calls. Attrition is the least traumatic way to make the adjustment, although it does not redeploy anyone either: the seat that goes empty is simply never filled again.

What I ask for in a steering committee

Two things before signing off on the project, and both fit in one meeting:

• the contact center's monthly attrition rate, because it sets the timeline for the adjustment
• which processes move to the agent and which cases stay with people, written down before starting

And a third one that does not depend on the vendor: telling the staff from day one that there will be no layoffs, because without that the project runs into people working against it.

A good review from me is not going to save that agent's job. Very strange times are coming and, without a doubt, there will be problems to solve.