Error margin has long been considered an internal quality measure: how many corrections, how much rework, how many complaints after delivery. Customers saw little of it, as long as the errors they encountered stayed within an acceptable bandwidth. Error margin was something a supplier judged itself on, not something a customer used to compare suppliers side by side.
That changes as soon as the work behind error margin shifts. Detecting deviations — an incorrectly entered amount, a forgotten step in a process, an inconsistency between two documents — is largely control work: checking something against a fixed set of rules or patterns. That is precisely the type of task where AI can take over a large part, with an employee assessing the outcome and approving or rejecting it with reason. Where this happens, the error margin does not decrease because people pay closer attention, but because there are structurally more control moments than was previously feasible within the same staffing level.
Not every company deploys this control in the same way, and that explains why one provider shows a noticeably lower error margin than another while both operate in the same market.
The difference lies in a few places:
A customer rarely sees error margin as a number. They notice it in how often an invoice has to be sent back, whether a delivery is correct the first time, whether a report is usable without a correction round. At a provider where control has largely shifted to AI with human oversight, that pattern is different: fewer feedback rounds, fewer "we'll just adjust it," fewer surprises after delivery.
That is also why error margin is difficult to claim in a sales conversation. "We work accurately" tells a customer little; it is the reason vague quality promises in sales pitches often sound interchangeable between providers. The comparison only becomes concrete when it is clear which part of the control has actually been taken over, and on what basis.
The shift in error margin affects how a provider scores on other points. Less rework often also means shorter lead times, which changes the comparison on delivery time: see how shorter lead times through automated planning shift the standard in a sector. Fewer errors in execution also saves on aftercare, which plays into how customers assess response speed and resolution rate of after-sales service. And because less rework also means lower costs, error margin indirectly affects how automation changes the room for sharper pricing. Anyone who assesses error margin in isolation misses that connection.
Few companies report their own error rate. Nevertheless, error margin can be read from indirect signals: how often a competitor appears in the news with corrections, how reviews talk about remedial work, how a quotation or delivery process is set up. For competitors who do not publish figures, there is an approach to gather signals about competitors without them publishing their own figures.
How much error margin decreases through AI-supported control depends on the type of error, the quality of the underlying data, and the degree of human oversight an organization maintains. No fixed percentage can be named that applies to every sector or every company, and no one can guarantee that a task is actually taken over. What can be assessed: which part of the control work in a specific company lends itself to being taken over, and which part remains work for humans.
The question of which work in your own organization can genuinely be taken over by AI is answered per task by the work scan of FTE TO AI. For comparison with your peer group, there is the free dimension check: you name where you think you are winning, and see which of those claims can be defended with evidence. The full benchmark, with scores per dimension and an evidence matrix, is under construction.