Delivering quality and proving quality are not the same work. The first happens on the floor, on the phone, in the code. The second happens afterward: checking, flagging deviations, correcting, recording that it's correct. That second part is where the difference now arises. Not because AI would be better at the craft itself, but because the checking work — the part that determines whether an error is noticed before the customer notices it — can partly be automated.
A year ago, quality assurance was a matter of sampling and human attention. Whoever had more capacity for checking caught more errors. That is still true at companies where that work still lies entirely with people. But where systems continuously test every transaction, every delivery, or every line of code instead of a sample, the standard shifts. Not because the people there are better, but because the coverage is greater.
Behind every quality claim sits a chain of tasks. Inspection: is this product, this delivery, this line finished as agreed. Deviation detection: does this fall outside the norm, and would someone have seen it without the system. Feedback: does the information reach the person who can correct course in time. Recording: is there a trail that shows testing took place, not just that the result turned out well.
These tasks fall into the three categories that recur everywhere. Deviation detection based on patterns in large datasets is often something a system can take over, provided the data is clean and complete. The judgment of whether a deviation is acceptable within the context of a specific customer usually remains a task with human oversight: the system flags it, someone approves or rejects it, with a reason. And the real professional judgment — is this work good, not just according to the standard but according to the client's sense of it — remains human work. None of these three is set up the same way everywhere, and that is exactly why the comparison between companies now measures something different than it did five years ago.
If the checking work is largely automated at one company and not at another, you are no longer comparing two companies with the same quality effort and a different outcome. You are comparing two different kinds of coverage. The company where every line is tested instead of a five-percent sample finds errors the other structurally misses — not through better work, but through more observation.
That changes what a quality claim is worth. "We apply strict quality control" used to mean: we assign people to it. Now it can also mean: with us, almost nothing is missed because the checking does not depend on how many inspectors were available that week. That is a different promise, and the customer who hears both does not know which one he is hearing without asking how the work is actually organized.
The shift is also visible in what remains for people to do. Where detection has been taken over, the quality function shifts toward judgment: is this deviation a problem, and what do we do about it. That is different work than tallying and counting, and it requires different skills. Companies where that shift has not yet been made still assign that professional judgment to the same people who also did the tallying — with less time for the judgment because the tallying has not yet fallen away.
The speed at which this work shifts does not depend on the willingness to innovate. It depends on whether the underlying data is structured enough to test automatically, whether the customer has a process that is easy to track digitally, and whether it is already established what "good" precisely means for that specific product or service. A production process with fixed specifications lends itself more easily to automated testing than an advisory process in which quality lies largely in the conversation.
That is also why quality rarely stands on its own. The lead AI delivers on quality is connected to how a company has organized its after-sales service, whether disruptions are noticed as quickly as they are resolved, and to the technical knowledge needed to assess the value of a flagged deviation. A system that reports a thousand deviations a day is not a lead without people who know which of them matter.
Whether a competitor has already partly transferred their quality work to a system is not visible from the outside in the end result. It is visible, however, in how they talk about it: in coverage and turnaround time, or in effort and staffing. When you yourself make a quality claim that you cannot back up with a figure, it is worth looking at how you substantiate a claim about your own quality before a customer or competitor does it for you. And if it turns out you are behind on this dimension, there is an approach to the question of what you do with a lag on a dimension that does not start with panic but with evidence.
The underlying question — which part of the quality work in your company can truly be taken over by AI, and which part remains human work — is answered by the FTE TO AI work scan per task, not at the level of the whole department.
As a first step, you can take the free dimension check: you name where you believe you win, and see which of those claims can be defended with evidence. The full benchmark, with the evidence matrix for your sector, is under construction.