How do you track what your customers are starting to do themselves with AI

A customer who used to have your account team draw up a quote now has the first version produced by an AI tool and only calls about the exceptions. A customer who hired an implementation partner first tries an AI agent on their own data themselves. This doesn't happen with all customers, and not everywhere at the same pace, but it is happening today, in parts, among customers who have the knowledge and the tools for it. The assumption shifting underneath this is usually not spelled out: "customers need us for this part of the work." That assumption may never have appeared in so many words in the strategy on paper, but the business model rests on it. If it no longer holds, the strategy remains standing while the ground underneath it changes.

Where this signal comes from

What customers do themselves with AI cannot be read from one source. It sits in three kinds of traces. First, in what customers ask about: questions that used to be about execution are now about validation -- "is this correct" instead of "can you make this". Second, in the volume of certain assignments: tasks that are self-contained and repeatable disappear from demand first, while tasks with exceptions and dependencies remain. Third, in what customers announce or procure internally: a vacancy for an AI role at the customer, or an announcement that they are rolling out a tool internally, is an external signal that makes visible the same pattern as how you read vacancies that disappear or change. Anyone who wants to know whether their own offering faces the same fate should also look at what competitors are already announcing: how you track what competitors announce about AI describes that signal separately, because it comes from a different angle but points to the same shift.

How often you should look

Tracking day by day makes no sense: the volume of individual customer questions varies too much to draw conclusions from it. Once a year is too slow: within twelve months a category of work can have tipped from "the customer always asks us for this" to "the customer does part of this themselves" without anyone noticing until revenue had already shown it. A quarterly rhythm works for most sectors: enough volume to distinguish a pattern from an exception, short enough to correct course before an assumption quietly withdraws itself. In sectors where suppliers themselves are building AI into their product at a rapid pace, a shorter rhythm makes more sense -- see how you track suppliers building AI into their product for why these two movements often coincide.

What a change does and does not mean

One customer using an AI tool themselves for a task you normally performed is not a trend. It could be an exception: a customer with a technical team, a pilot that stalls on quality, a choice that gets reversed after a quarter because the oversight required turns out more expensive than expected. A change only counts as a signal once the same pattern appears among multiple customers in the same category, and when it coincides with what you have already seen elsewhere: declining demand for a specific service, a supplier now offering the same task as standard in its product, a price drop in the underlying AI capacity that lowers the threshold for customers. That last point can itself be tracked: how you track the price of AI capacity shows why a falling price per task often changes sooner than customer demand itself.

A change also does not automatically mean the task disappears entirely from your offering. Often it shifts from "we do this in full" to "we do the part that requires oversight and accountability", while the customer takes over the routine part themselves. That is a different offering, not necessarily a smaller one.

Which decision this is tied to

The decision tied to this signal is not whether you adjust staffing -- that touches on the employer's own statutory requirements and falls outside what is answered here. The decision concerns the strategy: does the assumption "customers order this work from us" still hold for this category, or must the offering, the price, or the positioning be revised before revenue has already shown it. That decision should, as standard practice, be recorded -- not only as an outcome but with the reasoning attached: how you record why you made a strategic choice describes why that reasoning matters just as much as the choice itself, because a future leadership team must be able to see whether the assumption from back then still holds. How often the strategy as a whole is checked against signals like this is a separate rhythm question, worked out in how often you should recalibrate a strategy.

The underlying question -- which part of your own work, not the customer's, can genuinely be taken over by AI -- is answered task by task with the work scan from FTE TO AI.

What you can do now