Over a decade ago, I was working with the VP of supply chain at a multi-hospital system out west. The problem he had was one almost every hospital had at the time: he couldn’t tell you what surgery cost.
Not the billed price — the actual supply cost. What went into the room and got used up. Back then, surgical supplies were a single line item. At month-end, someone tallied what had been ordered and booked the total. There was no way to see what a given procedure consumed, no way to compare one type of surgery to another, and certainly no way to look at next quarter’s surgical schedule and predict what it would cost to stock. The hospital was driving by looking in the rear-view mirror.
We built a model to change that. It’s less impressive than it sounds — an Excel workbook, formulas, data imported by hand from the ordering system and the surgical case log. But it let us estimate supply cost forward, from the schedule, broken out by procedure type, by case volume, and by one variable most people didn’t think to include: the individual surgeon.
I included the surgeon because I already knew what the data would show. I’d spent enough time in surgical services to know that two doctors in the same practice, performing the same procedure, will use different supplies, different tools, different quantities. The model just made it visible in dollars. Surgeon A’s knee replacement cost materially more than Surgeon B’s identical one — and nothing clinical explained the gap. It was habit, priced.
That’s the moment the project stopped being about cost and became about something harder.
The card is the problem
Every surgeon has a preference card — a standing list of what to pull for their cases. It’s supposed to encode the procedure. In practice, it encodes the person: their preferences, their training, the reps who called on them, twenty years of accumulated habit. Multiply that across a department — then across every hospital in the system — and you get hundreds of subtly different recipes for what is clinically the same operation, each one treated by the supply chain as if it were a legitimate requirement.
The fix is conceptually simple. You move from a preference card — one per surgeon — to a procedure card — one per operation, with the supplies standardized to what the procedure actually needs. You keep the variation that’s clinical and you cut the variation that’s arbitrary. In one department we reviewed more than four hundred cards and found that a third of the differences were pure redundancy.
Simple to say. The hard part is that standardizing a surgeon’s card means telling a surgeon their personal way is now non-standard. Surgeons did not become surgeons by deferring to spreadsheets.
You don’t convince the surgeon. You convince the room.
Here’s what I learned doing this over and over, and it’s the part that transfers to work that has nothing to do with hospitals.
You cannot standardize a surgeon by arguing with the surgeon. A consultant with a cost model has no standing in that conversation. What worked — the only thing that worked — was to start with the nurses and the surgical techs.
A nurse or a tech usually works across several surgeons in a practice, or across a whole specialty. They are the only people in the building who see the variation from the inside. They already know that Surgeon B gets the same outcome with fewer supplies, because they’ve scrubbed in on both. So we worked with them first, card by card, before a single surgeon was in the room. And when it came time to talk to Surgeon A, the message didn’t come from us. It came from the person who stood next to them in every case.
“The surgeon will not take ‘you’re non-standard’ from an outsider with data. They will take it from someone whose judgment they already trust, holding the same data. The evidence was necessary. It was never sufficient. It had to be carried by someone with standing.”
Why this outlived me
When I left Deloitte, the model and the method stayed. Other consultants used them on similar engagements. For a long time I told that story as a point of pride — a deliverable that survived its author.
But the reason it survived is the actual lesson. It wasn’t the Excel. Anyone can rebuild the Excel — and today you wouldn’t; you’d wire an API into the ordering and scheduling systems, train an agent to surface the patterns, and generate the estimate in real time. The spreadsheet was always the disposable part.
What survived was the method: find the variation that’s arbitrary rather than clinical, make it visible in terms that matter, and route the change through the people who already hold the trust. That doesn’t age, because it isn’t technical.
The part that isn’t about hospitals
Every organization has surgeon preference cards. They’re rarely about surgeons.
They’re the four definitions of ‘active customer’ living in four departments, each certain theirs is the real one. The general ledger where every plant books the same cost a different way. The master data recorded five ways by five teams, none wrong exactly, all incompatible. We call it a data problem and reach for a data tool, and then the cleanup stalls — not because the tool failed, but because standardizing how someone records their work means telling them their way is now non-standard.
The instinct is to standardize everything, because uniformity is easy to administer. That instinct is wrong. Some of that variation is clinical — it’s there for a real reason, and flattening it does damage. The work is telling the difference, and then standardizing only what’s arbitrary.
You need the data to know which is which. You need someone the holdout already trusts to make the change stick. Miss either half and the cards drift back to preferences within a year — in a hospital, or anywhere else.