What Selling to Doctors Taught Me About Attribution

Most of us who build things online start from the same assumption: if you instrument everything, you can see everything. Wire up the events, tag the campaigns, watch the funnel. When something works, the dashboard tells you. When something doesn't, it tells you that too.

Then you sell into a market that doesn't behave that way, and you find out how much of that confidence was borrowed.

I run patient acquisition for medical and dental practices. It is not the most glamorous niche on this site, and I'd guess almost nobody reading this is thinking about how a periodontist fills next Tuesday. But it has been the most useful education in attribution I've had, because it's a market where a large share of the buying decision happens somewhere no analytics tool can reach — a phone call, a conversation in a waiting room, a neighbor's recommendation on the sideline of a kid's soccer game.

Years of running a medical marketing company have mostly been years of learning that the channel that gets the credit is rarely the channel that did the work. That lesson transfers to software more cleanly than I expected.

The last click is the cheapest click

Here's the pattern that broke my model.

A practice runs a campaign. Nothing obvious happens. Two weeks later, new patients start calling — and when you ask how they found the practice, most of them say some version of “I looked you up.” The dashboard records a branded search. Branded search gets the credit. By the numbers, the campaign did nothing and the practice's own name did everything.

Except the practice's name didn't do anything. The name was the last thing that happened. Something earlier put the name in their head, and that something is invisible because there was no click attached to it.

This is not a healthcare quirk. It's what happens any time a purchase involves more than one session and more than one device, which is most purchases above trivial cost. If your funnel shows that “direct” and “branded search” are your best-performing channels, you have not discovered that your brand is strong. You've discovered that your instrumentation stops working at exactly the moment the interesting thing happens.

The last click is easy to measure, which is precisely why it's overvalued. You end up optimizing the cheapest step in the chain and starving the step that actually created demand.

Some intent never becomes a click at all

The second thing that broke my model was the phone.

For a local service business, a huge share of the highest-intent contact arrives as a phone call. Someone finds you, decides, and dials. That's the best outcome in the entire funnel — and in a default analytics setup it produces almost no signal. A session with a short duration and no conversion event. By the numbers, a bounce.

So the pages that generate the most valuable behavior can look like the pages that are failing. I've watched people cut a page that was quietly driving the practice's best calls, because the page looked dead in the report.

Founders selling software hit a softer version of this constantly. Someone reads your docs, decides you're the answer, and mentions you in their team's Slack. Someone screenshots your pricing page and sends it to their manager. Someone watches a talk you gave. None of it is a tracked event. All of it is the actual purchase decision.

The general form: the more consequential the moment, the less likely it is to be instrumented. Casual browsing generates clean data. Serious evaluation happens in private.

What I do instead

I haven't found a way to measure what I can't see. What I've found is a way to stop pretending I can.

Measure at the level of the whole system, not the channel. Instead of asking which channel produced this customer, ask whether total qualified contact went up in the period after we changed something, and whether anything else changed at the same time. It's a blunter instrument. It's also honest, and it doesn't collapse the moment a customer switches devices.

Run holdouts when the stakes justify it. Turn something off in one market and leave it on in a comparable one. This is unglamorous and slow and it will tell you more in six weeks than a year of dashboard-reading. Most of the “obviously working” channels I've tested this way were working considerably less than the report claimed. One or two were working much more.

Instrument the invisible step where you actually can. Call tracking is the healthcare version — a distinct number per source, so the phone stops being a black hole. Your equivalent might be a specific link in your docs, a demo request field that asks an open question, or a signup step that captures how they heard about you before it captures anything else.

Then just ask people. The single highest-value question I've ever added to an intake process is an open text field: how did you hear about us? Not a dropdown. A dropdown returns your existing assumptions back to you. Free text returns the answer you didn't know to guess. A meaningful share of the responses I get name a source that appears nowhere in any report — a podcast, a friend, a post someone half-remembers.

The answers are messy and self-reported and people misremember. They are still better than a model that confidently attributes everything to the last thing it happened to observe.

The uncomfortable part

There's a version of this post that ends with a tidy framework. I don't have one, and I've come to think the wish for one is the problem.

Attribution feels like an accounting exercise. It isn't. It's an inference problem with most of the evidence missing, and treating it as accounting is how you end up defunding the thing that was working because it didn't leave a receipt.

The practices that grow fastest in my world aren't the ones with the best dashboards. They're the ones that got comfortable saying “we think this is working, we're not certain, and we're going to keep doing it for another quarter and watch the total.” That's an uncomfortable sentence for anyone who got into this because they liked measurable things.

But your buyers were never obligated to leave a clean trail. They were only ever obligated to decide.

Steve Bunker is CEO of Target Patients MD, a healthcare-only agency working on patient acquisition for medical and dental practices across the United States and Canada.