The manufacturing industry consists of work that is difficult to summarize in one type of task. There is planning work: scheduling orders, coordinating material flows, dividing capacity across machines and shifts. There is quality work: inspecting, measuring, assessing deviations and documenting. There is maintenance work: diagnosing faults, ordering parts, planning preventive work. And there is administrative work surrounding all of it: certificates, cost accounting, reporting to customers and regulators.
These types of work differ greatly in how predictable they are. A planning task with fixed rules and historical data lends itself differently to automation than assessing a weld seam that falls just outside the norm but is probably fine anyway. That latter assessment depends on experience, on context that is not recorded in any system, and on responsibility that someone must be able to bear. That difference largely determines which ambition is realistic for which part of the work.
The same threefold division runs through work everywhere in the manufacturing industry. Part of it can be taken over by AI: repetitive checks based on sensor data, compiling standard reports, recognizing known fault patterns. Part of it goes partway, with a human approving or rejecting: quality assessments where an algorithm makes a proposal and an operator or quality employee confirms or corrects the decision, with reason. And part of it remains human work: resolving a fault no one has seen before in that form, negotiating with a supplier about a deviating delivery, taking responsibility in a safety incident.
What already falls into the first category in one factory today may still fall entirely into the third category in another. That difference rarely comes down to the technology itself. It comes down to whether the data is in place, whether the sensors measure the right things, whether employees trust the system enough to leave something to it, and whether there is someone who may approve the outcome without the process grinding to a halt while awaiting a signature.
A management team that says "we want AI-driven quality control" is also saying something about the organization that must be able to support that. There must be someone who assesses the system's error messages, not just in theory but during the daily shift. There must be clarity about who may override a rejection by the system and on what grounds. And there must be spare capacity to set up and maintain the system, capacity that is often already fully allocated to ongoing work.
This is where many ambitions in manufacturing get stuck: not on the question of whether the technology can do it, but on the question of whether the organization has already filled in the role around it. Who decides when the system is uncertain? Who is accountable when a customer complains about a product that passed an AI-driven inspection? These questions touch on responsibilities within the workforce, and decisions that follow from them are subject to their own legal requirements; that assessment does not belong here.
Two factories with comparable machines and comparable products can differ greatly in what AI actually takes over from their work. One has collected and structured sensor data for years, the other has the same sensors but the data disappears into logs that no one has ever cleaned up. One has a quality team that has weekly time to assess deviations and adjust the system, the other has that team fully occupied with ongoing production. Neither is wrong, but it determines how much of the ambition is achievable this year and how much next year.
These differences are not unique to manufacturing. The same tension between ambition and readiness plays out in the transport sector, where planning and route selection raise similar questions, and in professional services, where oversight of advisory outcomes plays a role similar to quality control on the floor. The question of how to split an ambition into pieces that an organization can actually handle in a year also recurs in every sector, and is addressed separately in an explanation of phasing an ambition that is too large for one year.
The question underlying all these examples is not whether AI can take over this kind of work in general. It is which specific part of the work in this specific company can actually be taken over, given the data, systems, and people currently in place. That question is answered per task with the work scan from FTE TO AI, regardless of the image management itself may have.
An ambition only becomes usable once it is broken down into what the organization must be able to do, and tested against what it can already do. How quickly an organization can truly change on such points is a question that stands apart from the ambition itself and is worked out separately in a discussion of the actual pace of change of organizations.
As a first step, there is a free readiness check: eight short questions, one per dimension, giving a picture of where your organization is furthest along and where it lags most. The full ambition assessment, with the four layers from vision to target state and the translation into roles and decision rights, is under construction.
Vertel wat u wilt bereiken, dan kijken we samen wat daarvoor moet staan.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.