Gartner says 40% of AI projects will fail by 2027.

Gartner says 40% of AI projects will fail by 2027.

I think the number's low. And the reason is boring: most companies can't tell you how long a task takes before they try to automate it.

I've seen enough AI vendor demos to spot the pattern. Every one of them shows you the "after." Beautiful dashboards. Agents routing tickets. Workflows humming. Nobody shows you the "before" (because the "before" is your problem, and they don't want you thinking about it).

A company wants to automate customer onboarding. Great. But can you tell me:

How long does onboarding take today? (Not the guess. The real number.) Where does it break? Which step has the highest drop-off? What does a rep actually do during those steps that isn't in the SOP? What does a mistake cost you? If you can't answer those, you're not ready to evaluate an AI tool. You're ready to measure your own process.

That's the unsexy version of "wait." You're building the scorecard that makes the eventual decision obvious. When you know your baseline, the vendor demo either beats it or it doesn't. No hand-waving. No "it should save time." Show me the number.

I've watched teams skip this step and buy a $40K platform that automated a process nobody had measured. Six months later they couldn't prove it helped. Not because it didn't, but because they had no "before" to compare against.

Build your benchmarks before you build your agents.