The product cannot be disrupted, the operations can

Nobody is going to write a model that climbs a ladder and rehangs a warehouse door. That is the good news, and it is most of the reason a business like that is safe to own for thirty years. The opportunity is everything wrapped around the ladder, because in a company of that size it is still being done by hand.

A service business is two businesses stapled together

Look at what a door service company actually sells and you find two very different things sharing one bank account. There is the physical half: a trained person, in a van, with the right part, standing in front of a broken roller shutter at a logistics depot that cannot ship until it works. Then there is the coordination half: the call that came in, the quote that went out, the slot that got booked, the technician who got dispatched, the parts that got ordered, the job that got signed off, the invoice that got raised, and the customer who got asked again eleven months later whether the annual service is due.

The first half is why the business exists and why it cannot be competed away by software. The second half is where the hours go. In most companies at this size it runs on a desk phone, a paper diary, a spreadsheet, a WhatsApp group, and one person who remembers everything. When people talk about AI coming for small businesses, they are almost always imagining it coming for the first half. It is not. It is coming for the second, and the second is the half nobody defends.

That combination is rarer than it sounds, and it is the whole reason this category is interesting. A business whose product is easy to automate is not safe to own for a long time, because the thing you paid for can be reproduced. A business that is already run properly is safe but expensive, because the previous owner has already been paid for the improvement. What you want is a business whose delivery cannot be reproduced and whose administration has never been touched.

The number that makes the case

This is usually argued as a hunch. It does not have to be. The KfW runs a panel survey of the German Mittelstand, and in February 2026 its research arm published the current read on who is actually using AI. Twenty percent of Mittelstand companies used AI between 2022 and 2024, roughly 780,000 companies, about five times the rate recorded in the 2016 to 2018 survey. As a headline that sounds like the wave has already arrived.

The headline is not the interesting part. The breakdown is. Large Mittelstand companies, fifty employees and up, sit at 36 percent. Small ones sit at 19 percent. Knowledge-based services are at 29 percent, ahead of manufacturing and construction. And the sharpest cut in the whole report is by market geography: companies that operate within a 50 kilometre radius of their own front door use AI at 14 percent, against 27 percent for internationally active firms.

So the businesses most likely to be for sale, because their owner is turning 67, are also the businesses least likely to have been touched by any of this. That is not a coincidence and it is not a criticism of the owners. Somebody who has run a fifteen-person operation for thirty years has been busy running it. The gap is structural, and KfW's digitalisation report puts a second number on it: small companies make up 73 percent of the Mittelstand firms running digitalisation projects, but account for only 24 percent of the spending, down from 31 percent in 2016. The gap is not closing. It is widening.

The manual half, itemised

Vague claims about back-office inefficiency are easy to make and impossible to act on. Here is the actual list, in the order the work happens, for a business that sends people to sites. Every one of these is a place where a real company loses either revenue or hours, and most of them are legible from outside before you have seen a single page of accounts.

1. The call nobody answers

   The phone rings while everybody who could answer it is on a job. There is no queue, no callback, and no record that it happened, so the enquiry does not show up as a lost sale. It does not show up at all.

2. The quote that takes four days

   Pricing lives in the head of the owner or one senior technician. The quote waits until that person has an evening free. By the time it lands, the customer has had one from somebody faster.

3. The dispatch that is a phone call

   Who goes where tomorrow gets decided the night before, by one person, from memory, weighing traffic, parts and who gets on with which customer. It usually works. It does not survive that person being ill, and it never gets better at it.

4. The follow-up that never happens

   A quote goes out and nothing chases it. Not because anyone decided not to, but because chasing is nobody's job and the person it would fall to is already dispatching tomorrow.

5. The invoice that goes out late

   Job sheets come back on paper, in a van, at the end of the week. Invoicing waits on the paper. Cash waits on the invoicing, and the business quietly finances its own customers.

6. The service interval nobody tracks

   Plenty of these companies sit on years of installed equipment carrying a maintenance interval, with no system that knows when each one is due. The recurring revenue exists physically. It just does not exist administratively.

Read that list again and notice what is not on it. Not one of those six items is the skilled work. The technician is excellent. The repair is correct, and the customer is happy with the repair. Everything failing is the wrapper, and the wrapper is made of information moving between people, which is the one thing this technology is genuinely good at.

The trade is not the bottleneck. The paperwork around the trade is the bottleneck, and it has been for years, quietly, in a business that looks fine from the outside.

The objection: this is just software, and software already exists

This is the strongest argument against everything above, so it deserves a real answer. Field service software has existed for twenty years. Scheduling tools, mobile job sheets, automated invoicing and maintenance reminders are not new, they are not clever, and they are not expensive. If the gain were as large as the pitch suggests, the market would have closed it already. It has not, and pretending the reason is that nobody thought of it is not credible.

The reason is that adoption in a small company is not a software problem. It is a sequencing and attention problem. Installing a system means somebody has to redesign how the work flows, migrate years of records that only exist on paper, retrain people who have done it the other way for a decade, and hold the line through the six weeks where the new way is slower than the old way. The owner is the only person who could lead that, and the owner is the same person doing tomorrow's dispatch. There is no capacity. That is the actual constraint, and it never appears in a software demo.

What has changed recently is narrower than the marketing suggests, and the detail matters. Two things are genuinely different now. The first is that a system can read unstructured mess, a voicemail, a photo of a handwritten job sheet, a supplier PDF, without somebody first typing it into a clean field. That was the step that killed most previous attempts, because it front-loaded all of the pain. The second is that the integration work which used to need a project now takes a fraction of one. Neither of those makes the technician redundant. Both make the wrapper cheap to rebuild.

Durability is the point, not a consolation

There is a habit in this industry of treating a physical, unglamorous business as the thing you settle for when you cannot get into something better. It is the opposite. The fact that the delivery cannot be automated is the asset, because it is what stops any improvement you make from being competed away the moment you make it.

Think about what happens when you improve the operations of a business whose product is itself software. You have made a more efficient version of something a competitor can rebuild, and the gain flows out to customers within a couple of years. Now do the same to a company whose product is a licensed technician standing in a plant room at four in the morning. The efficiency you added sits behind a moat you did not have to build, made of vans, certifications, parts inventory, twenty-year customer relationships, and the fact that there is a national shortage of the people who can do the work at all.

Durable core

The part of what a business sells that cannot be reproduced by a model or a competitor's software, usually because it is physical, licensed, or both. It is what makes the improvement you make to the rest of the business permanent rather than temporary.

The wrapper

Everything around the delivery: capture, quoting, scheduling, dispatch, follow-up, billing, and recall. Almost always manual in a company of this size, and almost always the cheapest thing in the business to fix.

This is also why the order matters. Improving the wrapper of a business you do not own gets you a grateful client. Improving the wrapper of a business you do own gets you the whole gain, permanently, on an asset that keeps producing after you stop paying attention to it. The four layers apply here the same way they apply anywhere, except that in this category it is almost always the bottom two that are failing.

How to tell whether a business is in this position

Most of this read can be done from outside, before any conversation, in about twenty minutes. You are looking for one specific combination: a real business with an untouched back office. Not one or the other, both. A weak business with obvious problems is not the same opportunity, and it is a much worse purchase.

• The work is physical, on site, and needs a licence or certification to perform
• There is a maintenance or inspection obligation on the equipment, so the revenue repeats whether or not anyone is selling
• Customers are commercial rather than domestic, so the relationships outlast the person who signed them
• The website is a brochure: it describes services and gives a phone number, and nothing on it can book, quote or capture anything
• Nothing catches the call when it is missed, and there is no trace of a callback system
• Reviews are positive but sparse, which means the work is good and nobody has ever asked
• No dispatch software, no CRM, and job sheets that come back on paper
• They are not running ads and never have been

The last four are the ones people misread, so it is worth being explicit. Do not look for the absence of a website. Bitkom and the trades association surveyed 504 German trade businesses in 2025 and found 94 percent of them have one, and 62 percent send invoices digitally. The storefront was sorted years ago. What the same survey found is that 84 percent say AI is simply not a topic in their business, only 29 percent have anybody who could work with it, and 17 percent use anything describable as smart software. So the tell is not a missing website. It is a perfectly reasonable website with nothing behind it: no booking, no capture, no dispatch, no record of the customer between one job and the next. An empty result there is not a gap in your research. It is the finding.

What this changes about what you buy

Held properly, this thesis narrows the buy box rather than widening it, and that is a feature. You are not looking for cheap businesses and you are not looking for turnarounds. You are looking for a genuinely good company, with real customers and real repeat work, that happens to be run the way it was run in 1998.

Being strict about the first half matters because the second half is easy to find. Almost every offline company of this size has a manual back office, so on its own that tells you nothing and barely separates one target from another. What is scarce is a company where the underlying trade is durable, the customers are contracted, a second licensed person already exists, and the only thing missing is the wrapper. That is a much shorter list, and it is the one worth working. The full screen I run before anything else is in what makes a business worth buying, and the practical version of the ops side is in AI-native operations, without the hype.

One warning to close on, because it is the most common way this thesis gets misapplied. The improvement is bounded. Rebuilding the wrapper of a well-run trade business is worth a meaningful share of the reducible administrative cost and the leaked revenue, and none of the skilled labour, the parts, or the vans, which is where most of the cost base actually sits. Anyone quoting a headline percentage against the whole business is selling something. The honest version is that you are going after one specific slice, that the slice is large enough to be worth the work, and that it stays yours because the rest of the business cannot be touched.

Figures from Bitkom Research and the ZDH, Digitalisierung des Handwerks (2025, n=504, representative); KfW Research, Fokus Volkswirtschaft Nr. 533 (11 February 2026, KfW-Mittelstandspanel) and the KfW-Digitalisierungsbericht Mittelstand 2025.