Nine different answers to the same question, sitting in one query history, and not one of them knew the other eight existed.
I was going through a client's query history last month, looking for one thing: how many ways this team had defined "who counts as a patient on this drug." I expected two, maybe three. The usual mess.
There were nine.
Nine different cohort definitions for one medication. Built by different analysts, at different times, for different requests. None of them flagged as conflicting, because no single analyst had ever seen all nine side by side. They lived in nine separate queries, run by nine separate people, months or years apart, each one confident it was the definition.
That's not a horror story. That's a Tuesday. That's what's sitting in your query history right now, whether you've looked or not.
Every analytics team has a black box
It's called query history. Every metric calculation, every business rule buried in a WHERE clause, every cohort definition, every join nobody wrote down: all of it gets recorded, silently, whether anyone means it to or not.
Aviation figured this out decades ago. Airlines don't wait for a crash to read the flight data recorder. They read it after every single flight, scanning for drift before it becomes a failure. Most analytics teams are still waiting for the crash. They open the query history when a client challenges the numbers, when an analyst leaves and nobody can rebuild what they built, when a dashboard breaks and nobody knows why.
Our 46-KPI Health Assessment is us reading the black box before the crash.
How concentrated is the knowledge
Not "do we have documentation," but for every metric, how many people actually know how it's built. We've run this across healthcare data platforms, medical device distributors, and analytics services firms — different sizes, different maturity levels. The number doesn't move. 90 to 97 percent of metrics are known to exactly one person.
How much of it survives
Most logic runs once, produces a number someone uses, and disappears. At one company, 95 percent of a quarter's worth of analytical work vanished the moment the query finished executing. It existed. It was correct. It's gone.
Where one idea quietly became several
This is where my nine cohort definitions live. Zoom out from one medication to a whole analytics practice and you get a core usage metric with three competing definitions, each producing a different number, each defended by whoever built it. None of them technically wrong. Just built at different times, by different people, for different questions, and never once put next to each other.
Your best analyst knows which of the nine is right. They didn't learn that from a dictionary. They learned it by getting burned.
What that costs you
We measure this as an efficiency gap: the difference between an analyst rebuilding logic from scratch every time and one who's reusing what already exists. Across the same companies, that gap runs 5x to 61x between the best and worst case. That's not a productivity nice-to-have. That's the reuse problem, with a number attached.
And underneath all of it: is any of this AI-ready
When a text-to-SQL tool gets asked what "monthly active users" means, it doesn't know there are nine definitions fighting for the answer. It picks one. Your best analyst knows which one is right. The AI doesn't, because nobody ever wrote down which one was right in a form a machine could read.
Of course it picks wrong sometimes. Of course your board ends up with a number that traces back to a definition nobody reviewed. Nobody documented which of the nine was true, so how would it know.
None of this is a governance problem you can schedule for next quarter. It's the actual shape of your risk today, sitting quietly in query logs nobody's opened. The assessment doesn't fix it by itself. What it does is make it visible: exactly where the concentration sits, exactly what it's costing, so the conversation stops being "we should probably look into this" and starts being "here it is."
Nine definitions for one drug. I still think about that one.
Your analytics team already has a black box. Someone just has to read it.
We're building the context layer that reads it for you, before the client challenges the number, before the analyst leaves, before your AI agent has to guess.
Implementing AI in your analytics team? Let's talk.