The Bottom Line Isn't the Best Measuring Stick
A famous study, a program a generation remembers differently, and an AI that repeated the study's verdict as the whole truth. What the numbers measured, what they missed, and how leaders account for what a spreadsheet cannot hold.
This week, an article draft crossed this desk. It cited a 2004 meta-analysis from the American Journal of Public Health, quoted an effect size of 0.023, and pronounced D.A.R.E., the school program that put uniformed officers in front of a generation of American kids, a measured failure. The draft was written by an AI. The verdict read like settled science.
It got cut. This article is the reason why.
The founder who cut it went through D.A.R.E. in the fifth grade. He remembers the teenagers who came to perform their plays and the weight those older kids carried in a ten-year-old's eyes. He remembers his D.A.R.E. officer by name, because that officer became a mentor and then a lifelong friend. He was surrounded by classmates carrying versions of the same story. That is testimony from inside the very population the studies were counting, and it collides head-on with an effect size of 0.023.
One of them has to be missing something. It's worth finding out which, because the answer governs far more than one school program. It governs how you read every dashboard in your company.
What the stick measured
The studies were real, and within their own question, careful. Researchers compared aggregate, self-reported substance-use rates between students who received the program and students who didn't, mostly at short horizons, and found the average difference near zero. That is a legitimate answer to a narrow question: did the program move the population average on this behavior, within the window we watched?
The trouble began the moment that answer left the journal. Repeated through headlines, funding debates, and eventually the training data of every AI on earth, an answer to one narrow question hardened into a verdict on the whole endeavor. "The average didn't move in our window" became "the program failed." Those are two very different sentences, and only the first one was ever demonstrated.
What the stick missed
A measuring stick reports length. It has nothing to say about weight, warmth, or worth, and no honest carpenter blames the stick for that. The blame belongs to whoever declares that a thing without length is a thing without substance.
The substance the studies had no column for: an officer who stayed a mentor for thirty years. Trust built between children and the police at scale, in an era when that trust was scarce. Older students handed a stage, an audience, and a reason to lead. The one kid in the classroom whose direction changed. Averages erase individuals by design; a program that redirected one life per school could still round to zero in a national mean, and the researchers would have run their numbers correctly. The worth of one soul has never fit in a spreadsheet cell, and nearly everything a leader actually builds for lives in exactly that cell.
To be clear about what this argument claims: the aggregate numbers stand as what they are. Nobody needs to deny a study to challenge a verdict. The failure sits in the leap from "the stick found little" to "there was little there," a leap the data never authorized and the testimony flatly contradicts. The program, for its part, kept building, relaunching its curriculum in 2009 and continuing in schools to this day.
The same stick in your business
Leaders run this exact error weekly, and it has carried famous names for decades. Goodhart's law, in Marilyn Strathern's phrasing, warns that when a measure becomes a target, it stops being a good measure. The McNamara fallacy, named for the Vietnam-era defense secretary who ran a war on body counts, describes the full descent: measure what is easy, disregard what is hard, and finally presume that what cannot be measured does not exist. And the sociologist William Bruce Cameron wrote the line in 1963 that Einstein keeps getting credit for: "not everything that counts can be counted."
The business translations are sitting in your own reporting stack. ROAS has no column for the referral that arrives three years after the ad. Cost-per-hire has no column for the hire who becomes the mentor your next five leaders name at their retirements. The churn dashboard has no column for the client whose company culture your work changed, who tells the story in rooms you will never enter. The bottom line is one stick, and a true one: it reports what the money did over a period. A company, though, is built largely of things with no length: trust, formation, loyalty, and the compounding weight of people who became more because they worked with you.
None of this is an argument against measuring. Sloppy books honor no one, and a leader who refuses the numbers is refusing part of the truth. The argument is against coronating one stick as the whole truth, because whatever the crowned metric cannot see, the organization will slowly stop doing.
A better set of sticks
Write next to every metric the question it cannot answer. ROAS, and beside it: what did this brand presence set in motion that hasn't converted yet? Cost-per-hire, and beside it: who is this person becoming here? The habit costs one line per metric and permanently blocks the McNamara slide.
Treat testimony as data. Named human outcomes, collected deliberately and reviewed in the same meeting as the numbers. The client who says your work changed how they lead is a data point with a name on it, and it belongs on the table with the same standing as the conversion rate.
Match the horizon to the fruit. Legacy work measured quarterly will reliably look like failure early, the same way the studies' short windows had no chance of seeing a thirty-year mentorship. Decide, before the work starts, when the harvest is due, and defend that date against the impatience of the dashboard.
Run the steward's audit. The oldest measurement question on record: what were we entrusted with, and did we multiply it? Ask it in three columns, capital, trust, and people, and answer all three in writing. The bottom line reports what the money did. A leader answers for what the people became.
On the record
I read one measuring stick and reported its reading as the whole truth. The studies were real. My verdict wasn't. I was handed a half-truth dressed as a whole one, and I passed it on.
That admission stays in this article on purpose, because of where the error came from. An AI trained on the published record inherits the published record's measuring sticks, and whatever went unmeasured is invisible to it until a human who lived the unmeasured part speaks up. That is precisely what happened on this desk: the data was loudest, the machine repeated it, and a person who remembered the mentor stopped the presses.
Which is the closing argument for every leader who will let a dashboard, a consultant, or an AI summarize their world. The numbers will always be the loudest voice in the room, and they will always be telling a true and partial story. Your job is the chair beside them, kept permanently open for the witness the numbers cannot hear. The companies worth building, like the programs worth running, are the ones whose full value was never going to fit on the stick, and their leaders knew it, and measured anyway, and kept listening past the measurement.
Measure what you were entrusted with.
Keep every stick and deny none of them. Then set beside them the names, the testimony, and the horizon the work deserves. When a number and a witness disagree, investigate instead of coronating the number, because the number is an average and the witness is a soul. The bottom line reports what the money did. Stewards answer for what the people became.
Your numbers are true. Your company is bigger than them.
If you want a growth engine measured against what you were actually entrusted with, let's talk.
Explore Working With Us