The Two AI Projects I See First at Industrial Companies
Last week I sat with the Head of IT at a European technology distributor. Deep product expertise. A team that has been building specialized knowledge for decades. And no systematic way to preserve any of it.
From that single conversation, two AI projects appeared. Neither of them was customer-facing. Both of them were urgent.
I realized I have this same conversation roughly once a week.
The pattern
Over the past year I have sat in roughly 40 AI discovery sessions with industrial companies across Germany, Austria, and North America. The first AI project almost never starts with the customer.
It starts with a quieter question: how do we stop losing what we know?
Two archetypes dominate the early pipeline. The first is knowledge capture. The second is process automation. They show up in different orders, but both are internal and both are bounded. The companies that get this right tend to land their first AI project. The ones that skip straight to a customer-facing AI feature usually do not.
Use case one: knowledge capture
Every industrial company I work with has a retirement problem. The most experienced people, the ones who know why a parameter is set the way it is, where the workarounds came from, why the customer always asks for that one strange exception, are in their late fifties or older. They are leaving. The knowledge is leaving with them.
Wiki pages do not solve this. FAQ documents do not solve this. The expertise is not structured. It lives in patterns and exceptions and the kind of context that only comes from twenty years of doing the work.
What does help is a knowledge layer that understands how the work is actually done, captures the patterns rather than the documents, and surfaces the right answer when a less experienced person needs it. At H2W Labs we call this KnowKit. The concept is older than the brand: a structured way to preserve operational expertise so the next generation does not have to rediscover it.
Where this lands hardest: companies with deep domain expertise and high specialist dependency. Specialty distributors, custom manufacturers, technical service organizations. The marker is simple. If three or four people retiring would meaningfully damage your operations, you have a knowledge capture project waiting.
What good looks like: a new engineer can find an answer in two minutes that would have taken a thirty-minute call with an expert. The expert is freed up for actual problems. The new engineer ramps faster. The company stops paying the same lesson twice.
Use case two: process automation
The canonical example is accounts payable. High volume. Rule-based. Measurable error rate. A clear before and after.
What AI actually does is unromantic. OCR extracts line items from invoices. The system matches against purchase orders. Routing goes to the right approver. Exceptions get flagged for human review. It is not replacing the AP team. It is removing the predictable work so the team can focus on the exceptions, which are where the value actually is.
Other variants follow the same pattern. Order intake from B2B customers who still send PDFs and email attachments. Delivery confirmations. Customs documents. Anything that arrives in unstructured form and needs to land in structured form inside the ERP.
ROI here is measurable in three to six months. Sixty to eighty percent reduction in manual handling time is typical. The numbers are big enough to fund the next project and conservative enough to survive an audit.
Why the sequence matters
Knowledge capture and process automation come first because the conditions for success are easier to engineer inside the company than outside it.
Internal projects mean the data is available. The stakes are lower. Iteration is faster. Trust builds. The team learns what AI is good at and where it breaks, in a context where they can fix it. By the time they are ready to put AI in front of a customer, they have an honest sense of what the technology can and cannot do.
Customer-facing AI third, because it depends on data quality, internal confidence, and operational discipline that the first two projects build. Companies that skip to customer-facing AI first often fail. Not because the technology is wrong, but because the underlying data is fragmented or the team does not trust the system enough to let it talk to a customer.
The sequencing is not about the technology. It is about organizational readiness.
What I am actually seeing
Most of the conversations I have in Germany and Austria follow this pattern. North America is similar, with more appetite for moving faster and more pressure to show a quick result.
The companies getting this right are the ones that started with the right question. Not "where can we use AI?" but "what do we know that we cannot afford to lose?"
That second question leads to the right first project almost every time.
If you are at the start of this, do the internal inventory first. Then pick your first AI use case from what you find. The shape of the project will tell you which of the two archetypes you are looking at. The roadmap follows from there.
Related: Five Things Nobody Tells You About a CloudSuite Migration on how the same readiness questions show up when modernizing the ERP itself, and From Paper to Digital Shop Floor on how a bounded internal project builds the trust that later AI work needs.