Article

How does a service business become AI-native?

A service business becomes AI-native by capturing its expertise and letting AI prepare the recurring work, while your people give the approval.

BombosBombos team··9 min read

By Piet Baudoin · September 2026

A service business becomes AI-native by capturing its expertise and redesigning the work around it: AI does the recurring legwork and prepares it, your people review it and give the approval.

That starts with the service call that stalls, or the annual accounts waiting on answers. The bigger picture is in what the agentic economy means for your business. This article is about the step you can take with the business you already have.

Why look at your business again for this?

Because you want to handle more work with the same people, and speeding up one isolated task is not enough for that. According to ABN AMRO, 42 percent of owners in business services name staff shortages as their biggest bottleneck. The bank has its own interest in financing that sector. The shortage is not the same everywhere: the Dutch employment agency UWV sees a severe shortage of plumbers and machine mechanics, and rates the labor market tightness for bookkeeping staff as average. So in field service this is a shortage; in administration it is more a margin question.

The American investor Sequoia wrote in March 2026 about newcomers who use AI to deliver the work you now hire a service provider for. Sequoia invests in companies like that, so this is also a pitch. For you, it describes a possible competitor, but also what you could become yourself.

After all, you start with customers, expertise and a feel for what good work looks like. The question is how much of that is usable without someone having to call you.

My starting point: do not aim AI only at production. Annual accounts written faster still sit there if nobody chases the missing documents. The opportunity is also in gathering, routing and following up around the skilled work itself. So you need to look at your whole way of working.

Which work do you tackle first?

Choose work that someone is already paid for and that you can check for correctness. Those are the two questions you start with. Matching an invoice to the right project has a checkable outcome. Deciding whether to give an important client more time needs your judgment.

Then make the unit small: what exactly needs to be ready? With scheduling a service call, that is a proposal with the address, the complaint and a fitting time slot. The repair itself is a separate job. Drawing that line tells the reviewer exactly what they are approving.

Then arrange the intake. What data has to be in before the work can move forward? Invoices that arrive through three different channels need to land on the same work list, with recognizable fields. Let missing information show up as a visible open question. Otherwise you just speed up the retyping, and the searching stays.

Also write down when something counts as done. With assigning invoices to a project, that includes the backup for the decision. A filled-in project number on its own is not a delivery you can trust. You want to see which order or agreement supports that choice.

How do you get the rulebook out of your people’s heads?

You make expertise usable by recording, at real decisions, why something is right or wrong. Start with the colleague who checks the work now. What information do they look up? When do they deviate from the usual approach?

Take overdue reminders that keep sitting there. Before you send a reminder, you want to know whether there is a payment plan or an open complaint. Write down where that information lives and who decides when it is unclear. That is your rulebook: the agreements you use to judge what the next step should be.

Bombos interviews your people to get that knowledge out of their heads. After that, their decisions, approvals and corrections become rules for next time. The control layer is the approval: the AI agent prepares the work, a person reviews it. A correction also carries the reason, so the same judgment call can be found again later.

That rulebook belongs to your business, not to the vendor. So ask where you can read and improve the recorded agreements. That way, interviewing your people also sharpens your own way of working.

What changes about delivery and price?

You agree what work counts as finished and what the client pays for it. With annual accounts, for example, that also means clarity about the questions still open. Make progress visible to the client, so they do not have to call to find out which answer you are waiting on.

Compare two ways of working. You can have AI write a question email faster, after which someone still has to find it again and follow up. Or you can organize the file so it stays visible which question is outstanding, with whom, and what the answer sets back in motion. In that second approach, the delivery itself changes.

Price involves a similar choice. Geert Dreschler, an accountant at MKB Accountants (a Dutch accounting firm), told Accountant.nl in October 2025: “With hourly billing you don’t have that incentive. With a fixed price and a value proposition, you do.” (translated from Dutch)

So look into a price per defined job or a fixed monthly fee. Describe what is included and when extra work starts. Then working faster can create room within the same agreement. Without clear boundaries, a fixed fee promises an unknown amount of work.

Where does this go wrong today?

AI output can be wrong, and time pressure can strip away exactly the check that is supposed to catch it. Bert Vries, head of innovation and technology at law firm CMS, says in Accountant.nl: “Staff often work under tight deadlines, which creates the risk that they accept AI tool output too quickly.” (translated from Dutch)

So an approval is only worth something once someone can actually judge the underlying documents. A neatly worded reminder does not tell you whether the client made a payment arrangement yesterday. And a filled-in file does not prove the information in it is correct. Agree who checks it, and free up time for that.

Using AI is also not the same as changing how your business works. According to Statistics Netherlands (CBS), 29.8 percent of small and medium businesses used AI in 2025. That covers businesses with ten to 249 employees, a broader group than the reader here. The figure says nothing about hours saved or files handled well.

So also measure rework and review time. Only counting how much work the AI prepares leaves out exactly the costs that could make this approach fail.

And a rulebook by itself is not an advantage. Jason Lemkin of SaaStr describes companies where 50 to 80 percent of the work to switch to another vendor was done by copying and pasting the instructions, in a few minutes. What a competitor cannot simply take over are your people’s corrections on your real cases. That is where the value sits, not in a neatly written-down process.

What do I expect for September 2027?

My hypothesis: by September 2027, a documented way of working around one recurring task will save an existing service business more time than just adding AI to the writing and the number-crunching. Test that against total human time per completed task, including checking and rework, at the same quality level.

The reasoning starts with the difference between using AI and finishing the work. CBS measures use, ABN AMRO describes staff shortages, and the Dutch debate among accountants ties pricing to the incentive to work faster. Together they point me toward organizing the work as the next step. They are not a data series you can use to calculate how fast that shift will happen.

What would break my hypothesis: exceptions that demand so much review time that the gain disappears. Or intake and follow-up at your business barely costing any time, while the writing and calculating eat up almost all the hours. Then the gain sits somewhere else.

So write down now what a completed task looks like and how much attention it needs. A new pricing model for your whole business can wait until you know which work is demonstrably running better.

How do you start without turning everything upside down?

Start with one recurring task alongside your current way of working, for example sorting your shared inbox. Have the person who does that now review and correct the proposals. Track how much time is left after that review. That task is the beginning of handling more work with the same organization.

Bombos finds the matching data in the software you already use and prepares the work with the reason attached. We take you along, and after that you can do it yourself: your own team teaches the AI agent the next task.

The recorded knowledge stays in place when we swap the underlying AI for a better one. That way Bombos grows with you without you having to start your rulebook over. We follow developments daily; how much work becomes more independent in-house is up to you. No message to a client and no payment goes out without your approval.

Our promise: after three months, the work we start with is ready every day, without anyone having to think about it. Leave your number on the contact page, and we will call you back.

Start with the job that keeps waiting on someone.

Sources

Every source was opened on 21 September 2026 and every quote appears in it word for word.

The idea and the Dutch context

Price and control