Where AI Actually Moves Your P&L (And Where It Doesn't)
Most AI initiatives fail quietly, not because the model is bad, but because nobody named the line it was supposed to move.
Every AI project we've picked up after someone else stalled out has the same root cause: nobody could say, in one sentence, which number on the P&L it was supposed to move.
Not "efficiency." Not "customer experience." A line. Gross margin. Working capital. Cost of goods sold. Response time on a specific queue.
The three places AI reliably pays back
After running enough of these engagements, the pattern holds:
- Where a human is the bottleneck on a repeatable decision. Pricing 500 SKUs across ten cities by hand. Reconciling accounts payable by inbox. Answering the same guest question for the two-hundredth time. If a competent analyst could do it with enough hours, an agent can usually do it with none.
- Where the data already exists but nobody can query it. We've opened more than one company's ERP and found the answer to "why is cash so tight" sitting in inventory data nobody had connected to purchasing. The insight wasn't new. It was just locked behind SQL nobody on the team could write.
- Where the cost of a slow response is measurable. A three-hour reply window on a booking inquiry isn't a UX problem, it's a conversion problem. Cut it to five minutes and the P&L moves whether or not anyone calls it "AI."
Where it doesn't
AI is a bad fit for judgment calls with real ambiguity, one-off decisions that won't repeat often enough to justify the build, and (this one is unpopular) most "strategy" work. If the deliverable is a deck, you don't need an agent, you need a decision.
We say no to a meaningful share of the AI work that comes to us for exactly this reason. If we can't trace the initiative to a line on your P&L within the first conversation, we'll tell you that directly instead of selling you a roadmap.