AI is not a strategy. It's a capability you apply to a problem. The teams getting real value are the ones that stopped asking 'where can we use AI?' and started asking 'which of our existing problems is AI unusually good at?'
Four categories that pay back
- Repetitive knowledge work — drafting, summarising, extracting, classifying.
- Decision support — surfacing the right context to a human at the right moment.
- Personalisation at scale — messages, offers, and interfaces tuned per customer.
- Software leverage — smaller teams shipping more, faster, with fewer defects.
Three categories that usually don't
- Full autonomy over high-stakes decisions with no human in the loop.
- Replacing systems of record with a chatbot on top.
- Novelty features bolted onto a product that isn't working.
A simple test before you fund anything
For every proposed AI initiative, answer three questions in plain English: What decision or task does it change? Who is accountable for the outcome? How will we know within 90 days whether it worked?
If any answer is fuzzy, the project isn't ready — no matter how good the demo looks.