Essay · AI & Capital Discipline · Part 11 of 12
Most measurement I see would not survive fifteen minutes with a hostile CFO. It isn't dishonest. It's just built for a friendly audience, and the audience changed.
The standard to build against isn't whether the number is true. It's whether you can defend it to someone who'd rather it weren't.
Attribution is the whole fight
The number is rarely disputed. Handle time did drop. Throughput did rise. What gets disputed is whether your initiative is why, and that's where most cases fall apart.
Because something else was always happening. You also changed staffing. Volume was seasonal. A competitor had an outage. The team knew they were being measured. Any one of these gives a skeptic a place to stand, and once they're standing there, your number is decoration.
The organizations that win this argument set up the comparison before they start, when it's cheap. A holdout group. A staggered rollout by region. A pre-period long enough to establish a baseline nobody argues with. None of this is sophisticated. It's just impossible to add later, which is why it's usually missing.
The claim I've learned to distrust
Time saved. It's the most common claim in this cycle and the weakest.
Hours saved aren't money. They become money when headcount changes, when throughput rises against fixed cost, or when a cost line falls. If none of those happen, the hours were absorbed, and absorbed hours are real for the person and invisible to the business.
I'm not saying they don't matter. Absorbed time can show up as retention, or quality, or a team that stops burning out. Those are worth having. They're just a different claim, and you should make that claim directly instead of dressing it as a financial return, because the financial version gets tested and fails and takes your credibility with it.
Report the range
One habit separates people I trust on this from people I don't. They give me a range and tell me what would make it the low end.
"Between four and nine hundred thousand, and it's four if the seasonal effect explains more than we think." That's a person who's examined their own case. A single confident figure with no error bar tells me nobody's stress-tested it, and I'll find the weak point myself, and then I'll discount everything else they present for a year.
Ranges feel weaker in the room and are stronger over time. That trade is worth making every time.
Measure the failures too
The most useful number nobody reports is the cost of what didn't work.
A portfolio where three of ten initiatives paid and seven were killed early might be excellent. It depends entirely on what the seven cost and how fast they were stopped. Report that and you're telling finance something real about your process, not just your winners.
It also protects you. A team that only ever reports successes gets read as either lucky or selective, and the second reading arrives eventually.
On Monday
Take the strongest result you're currently claiming and try to break it. Write the best case a skeptic would make.
If you can't answer it, you don't have a result yet. You have a number waiting for someone to ask about it, and better it's you.
The operating principle. Build every measurement for the skeptic, not the sponsor. A number with no error bar hasn't been tested.
Juan Vegarra is the author of An Outsider’s Playbook. The views here are his own. More essays · Advisory · Write me