Blog · Strategy

Stop running AI pilots: pick one process and finish it.

Seven in ten smaller companies now use AI. Even so, the gains mostly stall at loose conveniences. Research points to the cause, and it is good news for anyone willing to choose.

Seven in ten smaller-company owners now use AI, accountancy software maker Exact calculated this summer; a year earlier it was one in three. Experimenting has become the norm: a subscription here, a smart assistant there, a pilot in every team. Yet in most companies the returns stall at loose conveniences. New research shows why, and what the small group that gets a lot out of it does differently.

Shallow and wide is the default.

In Harvard Business Review, researchers from BCG described late last year the pattern they saw at hundreds of companies: dozens of AI initiatives at once, each defended with its own little sum. They call that approach shallow and wide. Every application wins a few percent of time somewhere, and there it ends: apply AI to five to ten percent of a task and no process gets genuinely redesigned.

There is a second problem: every loose application is simply for sale. Whatever you bolt onto a tool this month, your competitor has next quarter. Time savings anyone can buy are no head start.

The underlying BCG research among a thousand executives shows that only 4% of companies take the other route: pick a few priorities and transform deeply there. Over time that small group gets twice the return on AI compared with the broad approach.

What choosing delivers.

Consumer goods giant Reckitt aimed generative AI exclusively at marketing. With over two thousand marketeers it mapped three hundred tasks and plotted them by automatability and time spent; around one hundred proved suitable. The result: product concepts come together 60% faster and the communication processes became over 30% more efficient. IKEA did the same with interior design: what took days and around a hundred dollars per room through an external agency, AI now does in ten minutes.

Big names, small principle: depth beats breadth. And a company of five to fifty people applies that principle more easily than a corporation.

Ten pilots change no process at all. One process genuinely redesigned changes your whole week.

Isometric plate: on the left loose AI pilots on islands that connect to nothing, on the right one process redesigned from request to invoice.
Loose pilots stay islands. One process, redesigned from request to invoice, delivers double over time.

The real work: 70-20-10.

The same researchers sum up where the effort goes as 70-20-10. Seventy percent is change management: taking the team along into a new way of working. Twenty percent is getting the data in order. Ten percent is technology. That explains why a stack of AI subscriptions changes so little: the heavy lifting sits with the process and the people who will work with it, and you only do that work well in one place at a time. It matches what we wrote earlier about why software projects fail: the failure factor is almost always adoption.

A smaller company can choose.

A company of five to fifty people has an edge here. You have a handful of core processes, short lines and an owner who can see the whole. Where a corporation spends months debating which domain goes first, you choose on a Tuesday afternoon. Three questions point out the process:

  • Which process touches every client? Improvements there count double, because they come back with every job.
  • Where does most time leak away? Find the place where people retype, forward and gather things; that is probably where you are the glue.
  • What lives mostly in heads and mailboxes? Work that is recorded nowhere is the first thing to jam when someone is out or when you grow.

The answers usually land on the same line: from request to invoice at services firms, from site survey to handover at installers.

This is how you finish it.

Take that one process and redesign it from start to finish. Record where the work comes in, which steps AI may handle by itself and what a person sees first. Take the team along from day one; that is the seventy percent. Measure the effect after a few weeks, in hours and in calm. And start on the next process only once this one runs and gets used.

It feels slower than ten pilots at once. The research says it delivers double over time, and it produces something a competitor cannot copy with a credit card: a process built around your company.

Wondering which process to start with in your company? The quickscan points it out in five minutes, or email hello@bravio.nl. You will have an answer within one working day.

Further reading

More on working without noise.