Fewer Projects, Faster Results

Here’s what separates the organisations getting real returns from everyone else: they do less. BCG found the top performers achieve a 76% closer match between where they put AI and where it actually pays, and they get to impact inside twelve months rather than eighteen (BCG, 2025). Spreading budget across a dozen experiments feels like progress. Concentrating it on one or two is what creates it.

The difference shows up most starkly in what reaches daily use. Among those top performers, 62% of AI initiatives are live and working. Among the organisations struggling, it’s 12%, with most of the rest still sitting in planning. Same enthusiasm, same market, wildly different outcomes. Running eight small experiments doesn’t spread your risk, it divides your attention eight ways and gives none of them enough support to survive contact with real work.

Choose the Work, Not the Technology

Don’t start with a tool demo. Start with your own week. Somewhere in it sits work that’s repetitive, rule-following and eating hours: the reports assembled by hand, the data moved between systems, the same questions answered again and again. That’s your shortlist, and it was there before any vendor rang.

A vendor-led start puts the answer before the question, which is how organisations end up with capable software solving a problem nobody was losing sleep over. Turn it round and ask the people doing the work where their week disappears. You’ll hear the same four or five answers, and the good news is that the tasks named most often, pulling information out of documents, keeping records current, writing up notes and drafting routine replies, are precisely the ones this technology handles well today.

A Pilot Without a Destination Is a Hobby

Only a quarter of organisations have moved even 40% of their AI experiments into live use (Deloitte, 2026). The trap is comfort: pilots are cheap and safe, so it’s easier to fund another one than to scale the last one. Break the pattern before you start. Define what success looks like in numbers, give the project a hard deadline of a few weeks, and decide upfront that it either goes live or gets stopped.

It’s worth knowing why the step from pilot to live catches people out. A trial runs with a handful of willing users on tidy data in a corner of the business. Going live means integration, security sign-off, monitoring and someone owning it on a Monday morning, and work scoped at three months can stretch to eighteen once that reality arrives. None of it is insurmountable, but it does have to be planned for at the start rather than discovered at the end, and that single piece of foresight is most of the difference between a pilot that lands and one that quietly joins the pile.

Ninety Days Is Enough

Choose one piece of work, set the finish line first and there’s no reason your first project can’t be live and proving itself inside ninety days. Not everything, not everywhere. One thing, done properly, showing results you can point to.

And here’s the quiet advantage of starting this way: your second project inherits everything the first one taught you. The data questions are already answered, the review habits already exist, and the people who were sceptical in January have now watched something work. Momentum, it turns out, is the cheapest thing in AI, and the only way to get it is to finish something.