Every few weeks I get a version of the same call. Leadership has decided the company needs 'an AI strategy.' There's a board deck with the word transformation in it. Someone has a budget, a deadline, and a low-grade panic that competitors are already ten steps ahead. And the question is always some flavor of: where should we use AI?
It's the wrong question. Not wrong to care about - wrong to start from.
When you start from the technology, you go shopping for places to put it. You end up with a chatbot nobody asked for and a pilot that demos well and changes nothing. When you start from the work - from the specific place your business quietly bleeds time and money every single day - the technology picks itself, and the return is obvious enough that you don't need a deck to justify it.
Follow the pain, not the hype
The best AI opportunities in your company are almost never exciting. They're boring, repetitive, and beneath everyone's job description. They hide in plain sight:
- The report someone rebuilds by hand every Monday morning.
- The inbox where three people spend their day copying details into another system.
- The approval that sits for two days because it's waiting on a person to forward an email.
- The spreadsheet that is, if we're honest, the actual system of record.
That's your strategy. Not 'adopt AI' - delete that phrase. Find the one process that costs the most and hurts the most, and fix it properly. The strategy reveals itself once you've done it once, because now you can see the pattern everywhere.
Why starting small is the sophisticated move
Starting with one painful process isn't playing it safe. It's the move that actually works, for three unglamorous reasons.
The ROI is legible - you're fixing something the business already measures, so nobody has to take the value on faith. The scope is bounded, so you can ship in weeks and learn on something real instead of debating architecture for a quarter. And you build trust: with the team whose day just got better, and with the leadership who can finally point to a number.
Do that two or three times and you don't have an AI strategy in a slide. You have one on the ground, made of systems that are already paying for themselves.
The part vendors won't tell you
Sometimes the honest answer is that you don't need AI at all. The report that eats all of Monday might just need two systems to talk to each other. The 'AI use case' might really be a tool you should cancel. I'll tell a client that, even when it means a smaller engagement - because the alternative, selling someone a model they didn't need, is how the whole industry earned its skepticism.
Build the smallest system that removes the most pain. Then do it again. That's the whole strategy, and it beats the deck every time.