AI strategy vs. AI flows: what Norwegian SMBs actually need

Most AI projects start with a report.

Thirty pages, clean charts, a maturity ladder, and a recommendation to "begin the journey." It looks good in a board meeting. Then it sits in a folder.

I've started to distrust that deliverable.

A strategy report describes what someone thought at a particular moment. A workflow that runs on Monday morning describes what actually works. The two are nowhere near each other in value — but we pretend the first is groundwork for the second. Usually it's just groundwork for more groundwork.

The report survives because it's easy to sell and safe to buy. Nobody gets criticised for commissioning a study. It looks thorough, it's easy to show off, and it pushes the responsibility for actually doing something a little further into the future. A flow that doesn't work, on the other hand, is visible immediately — and someone has to own it. That's more uncomfortable. It's also why it's worth more.

Let me make it concrete.

Picture an SMB with 40 employees. The bookkeeper spends a few hours each week on the same thing: pulling in invoices, matching them against orders, pulling out the ones that don't add up. You can write a strategy about "AI in the finance function." Or you can set up one flow in Tripletex that does the matching and drops the exceptions into a list for a human. One gives you a recommendation. The other gives you back your Thursday.

What makes the flow better isn't that it's faster to build. It's that it's honest.

A strategy can't be wrong for another six months. A flow is wrong right away, if it's wrong — and then you know. You find out the data was messier than anyone admitted. That there were more exceptions than expected. That the real problem sat somewhere completely different from where the report pointed. That's information you'll never get out of a slide.

This doesn't mean strategy is wasted. It means the order has flipped.

We used to build the strategy first and hope reality would fall in line behind it. Now you build the smallest flow that works, watch what it teaches you, and let the strategy grow out of that. The strategy becomes a summary of what you already know works — not a bet on something you haven't tried.

For a large company with its own IT department and time to wait, the old order can be defended. For a Norwegian SMB it can't. You don't have twelve months and a project office. You have a managing director already doing three jobs, and a couple of systems that almost talk to each other.

The best investment isn't a map of the terrain. It's a path through it.

So if you're weighing up "an AI strategy" for your company, turn the question around. Don't ask what AI could do for you in the long run. Ask what eats the most time this week — and whether one flow could take a bite out of it. Pick something small and boring. Get it running. Measure whether it actually saved time.

It won't look as good in a board meeting as a thirty-page report. But you can run it on Monday, and you'll know whether it works by Friday.

A strategy gives you a plan. A flow gives you an answer.

It depends on what you need most. But most people need the answer.