Start with your most boring process: an operations guide to a first AI win

When companies ask me where to start with AI, they almost always point at the exciting things. The customer-facing chatbot. The slick sales assistant. The feature that would look good in a press release. I understand the pull, but it is usually the wrong place to begin.

The best first AI project in a smaller company is rarely the glamorous one. It is the dull, repetitive back-office process that everyone has quietly learned to tolerate. The order intake that someone retypes by hand. The quotes rebuilt from scratch every week. The data copied from one system into another because the two were never connected. These are unglamorous. They are also where the money is.

Why boring beats exciting

A boring internal process is the safest possible place to learn. If a customer-facing chatbot gives a wrong answer, your reputation takes the hit in public. If an internal quote-drafting helper makes a mistake, a colleague catches it before it leaves the building. The stakes are lower, so you can move faster and learn in private.

You also already understand the rules. Nobody needs to explain how your order intake works, because your team does it every day. That shared knowledge is exactly what makes a process automatable. The exciting projects usually fail on the opposite problem: nobody can quite say what “good” looks like, so the AI has nothing solid to aim at.

And a boring win compounds. It builds internal trust, it produces clean data, and it gives you a concrete result you can point to when you propose the next, bigger project.

The four-question filter

Before you automate anything, run it through four questions. I use these in every AI-waardescan, and they sort the prime candidates from the time sinks fast.

1. Structured input. Does the work start from something predictable, like a form, an email in a known format, or a file with consistent fields?
2. Predictable output. Is the result something you could describe clearly to a new hire on day one?
3. Rule-based decisions. Are the choices mostly logic and policy, rather than gut feel or delicate negotiation?
4. Repeated frequently. Does this happen many times a week, every week?

All four yes is a prime candidate. Three out of four is often still worth it. Two or fewer, and you are probably looking at the exciting project that will quietly drain a quarter.

What this looks like in practice

Three examples I see again and again in 30 to 200-person companies.

Order intake. Orders arrive by email or PDF and someone retypes them into the ERP. Structured input, predictable output, clear rules, dozens of times a day. A textbook candidate.

Quote drafting. A salesperson rebuilds a quote from old documents, copying prices and terms by hand. The logic is well understood and the format barely changes. A helper that drafts the first version turns 40 minutes into 5.

Data re-entry between systems. Your webshop and your accounting tool do not talk, so someone bridges them by copy-paste every morning. Pure mechanical work, high frequency, zero judgment. Almost always automatable.

Estimate the real cost first

Before you build anything, put a number on the pain. The arithmetic is simple and it keeps you honest: hours per week, times a loaded hourly cost, times the weeks in a year.

Say order intake takes one person 8 hours a week at a loaded cost of 45 euros an hour. That is 360 euros a week, roughly 18.000 euros a year, on a task nobody enjoys. Even if automation only removes 70 percent of it, you are looking at around 12.000 euros a year freed up, plus fewer typos and a happier team. Now the project justifies itself on a single line, and you know what a sensible build budget looks like.

This number also protects you from the opposite mistake: spending weeks automating something that runs twice a month and saves an hour. Boring is good. Trivial is not.

A boring win earns the next project

The pattern I see in companies that succeed with AI is almost always the same. They do not start with a moonshot. They pick one tolerable-but-wasteful process, automate it cleanly, and bank a result everyone can feel. That win earns trust, frees up time, and produces the clean, structured data that more ambitious projects depend on. Six months later they are doing the exciting thing, but from a position of strength rather than hope.

If you are not sure which of your boring processes is the right first one, that is exactly what the AI-waardescan is for. We map your real processes, run them through the four-question filter, quantify the time and cost, and hand you a prioritized shortlist. No hype, no vendor pitch, just a clear view of where AI earns its place first. Starting boring is not a compromise. It is the fastest route to a result you can build on.