Scope to one number: why we won't start a build without a metric
Most AI pilots die in the deck because success was never defined. Our fix is a rule we apply to ourselves first.
Here is how most AI projects die. A pilot is approved with enthusiasm and a vague mission — "explore how AI can improve operations." Three months later there's a demo, a deck, and a strange silence, because nobody can answer the only question that matters: did it work? Nobody defined what working meant. The pilot didn't fail. It was never aimed.
Our fix is a rule we apply to ourselves before any client feels it: every build is scoped to one business number, named in writing, before work starts. If we can't name the metric, we don't start.
What counts as a number
Not "efficiency." Not "better customer experience." A number someone in the business already feels:
- Response time — how long a new lead waits before anyone answers.
- Quote turnaround — inquiry to estimate-in-hand.
- Hours returned — admin time handed back to people hired for something else.
One number, not five. A dashboard with one honest metric changes behavior; a dashboard with twelve becomes wallpaper by Thursday.
Why the rule protects the client
The metric is a kill switch pointed at us. If the number doesn't move, the conversation about why is short, factual, and on our side of the table — that's the deal, and it's why we scope small first. Vague goals protect vendors; specific numbers protect clients. Any AI partner who resists naming one is telling you something.
Why it protects us too
Scope creep dies at the same door. When every request can be tested against "does this move the number?", the project stays a system instead of becoming a wish list. The discipline is symmetrical — which is the only kind of discipline that survives contact with a real engagement.
What we turn down
Builds where the honest answer to "what number?" is "we'll know it when we see it." Not because those businesses are wrong — because starting anyway would make us the vendor we warn people about. The kindest thing an engineer can say is sometimes "not yet — let's find the number first." Discovery exists for exactly that.