In manufacturing, the work largely sits in three streams: the upfront process of costing and quoting, the planning and production control on the floor, and the quality control that determines whether an order is good enough to ship. On top of that, there is the work tied to machines and materials themselves: maintenance, procurement of raw materials, and the administration around certification and traceability. Which of these streams eats up the most hours differs per company and per type of product, but the outcome of a deal is rarely determined by the stream with the fewest hours.
The conditions that steer the outcome are partly technical and partly human. A quote has to match the actual lead time, not a fixed margin that happens to work out most of the time. A delivery time is only as good as the planning that produces it, and that planning is only as good as the data it runs on. And in manufacturing, quality is not judged on an average, but on the outliers: the one batch with a defect that a customer remembers.
The question is not whether AI is taking over work in manufacturing, that is already happening in parts. The question is what that does to the dimension on which a customer chooses between suppliers. As long as delivery time depends on a planner's workload, the party with the most experienced people on the floor wins. Once planning is largely done by software that runs through scenarios and only has a human approve exceptions, delivery time shifts from a capacity issue to a data issue: whoever has the best input wins, not whoever has the most planners.
The same applies to costing. A quote that used to take days because someone had to work through drawings can now partly be generated automatically, with an engineer reviewing the outcome for illogical deviations. Does the party that quotes fastest still win, or the party that quotes most accurately? That answer is not the same everywhere, and that is exactly why the comparison between suppliers changes: the dimension on which parties used to win still exists, but what fills that dimension is no longer what it was.
Quality control is the third example. Visual inspection that AI takes over does not change whether a customer asks for quality, but it does change what "scoring better on quality" means. A company that still does inspection entirely with people is then no longer competing on the consistency of the human eye, but on something else: speed of correction, transparency about deviations, or warranty terms. Whoever fails to see this keeps defending a dimension that no longer counts the way it used to.
This difference between companies is not about ambition but about what there was to automate given their product mix and their data. A manufacturing company with a lot of series production and standardized drawings has an earlier basis for letting costing be taken over than a company that mainly delivers custom work where every order is a new exception. A company with years of sensor data from its machines can make maintenance predictive; a company without that data first has to invest before AI can take anything over there. The shift therefore does not hit every company in the same order, and that means that the peer group at any given point in time contains both companies that already score on the new dimension and companies still defending the old one.
Whether AI-driven capacity gains also lead to fewer FTEs is a decision for the employer, subject to its own legal requirements; here the only question is which dimension decides a deal and who is scoring on it at this moment.
This shift is not limited to manufacturing. What delivery time means once planning takes over the work of transport planners shows a comparable pattern, as does the question of what consultancies distinguish themselves on once research and reporting are largely automated. If you recognize a dimension on which you believe you are winning, the next question is what you do when it turns out a competitor is now ahead of you on it: what to do about a lag on a dimension that decides the deal offers an approach for that.
The underlying question of which work in your company can genuinely be taken over by AI is answered task by task with the work scan from FTE TO AI.
Before defending yourself on a dimension, it is worth knowing whether that dimension is still what it used to be, and whether your claim on it holds up. A claim such as "we are faster" or "we are more accurate" is only worth something if it can be substantiated; how that works is described in how to substantiate a claim about your own quality. With the free dimension check, you name what you believe you are winning on and see which of those claims can be defended with evidence. The full benchmark, with your position and that of your peer group on every dimension, is under construction.