Every organisation feels pressure to ‘do something with AI’. The discipline is deciding which something — and being willing to choose none of them, for now.
Three honest options sit behind most AI decisions, and the right answer is rarely the most exciting one.
Build
Build when the capability is genuinely core to your differentiation, when you have (or can create) the data foundation, and when you can staff it with people who’ve shipped ML to production before. Building a commodity capability you could have bought is how budgets evaporate.
Buy
Buy when a mature product already solves the problem well and your advantage is in using it, not owning it. Most AI needs — transcription, classification, summarisation, standard automation — are better bought than built. The integration work is real, but it’s a fraction of the cost of a bespoke model.
Wait
Wait when the technology is moving faster than your use case is urgent. Sometimes the model that’s hard and expensive today is a cheap API call in six months. Spending heavily to be early, on a problem that isn’t time-sensitive, is a poor trade.
We’ve saved clients more by telling them what not to build than by anything we built.
The value isn’t in having an opinion about AI. It’s in having a clear-eyed one about your specific situation — and the confidence to recommend patience when patience is the right call.