Parkview Health Services runs behavioral health pharmacies. Two years ago it had about five of them. It now has sixteen.
Eleven more locations in twenty-four months. Some of them arrived carrying their own pharmacy management system, which is normal, and which everybody filed under integration work to be scheduled later.
Here is the part worth sitting with. Every one of those systems reported accurately. Not roughly, not close enough. Accurately, against the definitions the software itself shipped with.
A date meant written, or filled, or checked, or shipped, or billed, depending on which report produced it. Every one of those reports was correct. The company still could not say how many patients it served, because a patient who used two Parkview pharmacies counted as two patients.
No data quality tool flags that. Every record is valid. Every total ties. The numbers were accurate the entire time they were wrong.
Nothing broke. The question changed.
Before the growth, none of this mattered, and it matters to be precise about why. The systems never had to agree with each other, because until each deal closed they never reported together.
Five pharmacies is a company where the tribal answer is the correct answer. Somebody in operations knows that the Tuesday report counts billed date and the clinical one counts filled date, and that person is in the room when it matters.
Sixteen pharmacies is a different company wearing the same reporting layer.
Nothing was broken on the day the definitions stopped being true. There was no outage, no failed load, no red light on a dashboard. The acquisitions closed, the reports kept rendering, and the gap between what the numbers said and what the business was opened quietly, one deal at a time.
That is the shape of this failure and the reason it goes unnoticed for years. It does not announce itself as a data problem. It announces itself as a slow executive meeting where two people quote different patient counts and the conversation moves on without resolving it.
Definitions are a deliverable nobody assigns
Mapping the workflow was the only way to find out what had actually been getting measured. Not profiling the tables. Walking the process and asking what the business means by a word.
Those are business questions before they are technical ones, and that is exactly why they sit unowned. The data team cannot answer what "patient served" means. The pharmacy operation can, but does not experience it as a thing that must be written down, because for five locations it never was.
So the definition stays tribal, and the reporting layer quietly becomes the constraint on how fast the company can buy the next pharmacy.
Parkview expects to keep acquiring for another three to five years. The reporting layer is no longer the thing that decides how fast, because somebody finally treated the definitions as work with a name on it rather than an assumption everybody shared.
Your Monday move is small and slightly uncomfortable. Take the three numbers your leadership quotes most often. For each one, go to every system that produces it and ask what the field actually means, not what the dashboard label says. Write the answers next to each other.
If two of them disagree, you have not found a bug. You have found the date your company outgrew its own vocabulary, and the fact that nobody was watching for it.
Growth is the thing that invalidates a data model, and it never files a ticket. Your definitions were written for a company you may no longer be running. When did you last check which one you are in?