Which ads produced patients who actually attended?

Someone researching a procedure may inquire in March, think about it for six months, discuss it with a partner, and book in October. Every part of that except the first click is invisible to an ad platform, so the campaign that actually produced the patient looks like it produced nothing.

Matching on the person rather than the session makes the October booking count against the March inquiry. For clinics with high-value elective procedures this routinely reverses which channel appears to be working.

The match needs a phone number or an email address, an amount, and a date. It does not need a diagnosis, a procedure code, or a clinical note, and those should not be in the file you upload.

Data is encrypted in transit and at rest, isolated per workspace and deletable in one click, and we act as a data processor with a DPA available. The match itself needs a contact detail, an amount and a date — nothing clinical — so the narrowest export that works is also the safest one to send.

Two patients arrive for the same appointment with the same complaint. One is self-pay at the posted price. One carries a plan whose allowed amount is roughly half of it. A third has a deductible that means the clinic waits four months and then writes part of it off anyway. None of this was visible at the moment somebody clicked an advert, none of it appears in any report an advertising platform produces, and counting booked appointments treats all three as one identical unit of success. The clinic's own books know better, and they are in a different building from the marketing decision.

Coverage and channel are not independent of each other, which is what makes this an attribution problem rather than a billing one. Campaigns built around a named procedure pull self-pay demand from people who have already decided to spend their own money. Campaigns built around symptoms and same-day availability pull whoever lives nearby with whatever coverage they happen to have. A clinic ranking its marketing on appointments is comparing those two campaigns as though a filled slot were a filled slot, and then wondering why a strong month for bookings was a poor month for the bank balance.

Ranking on what was collected takes the guess out. Export posted payments rather than charges, and the report shows the revenue each source produced after whatever the payer adjusted away, so the effect of coverage is already in the money. Do not add a payer or plan-type column to see the split: how somebody's care was paid for is health information, and CloseRev signs no business associate agreements. One campaign can carry a respectable share of visits and a much smaller share of money. That is a bidding decision rather than a reason to stop: the campaign is working, it is simply worth less per booking than the budget assumed, and now there is a figure for how much less.

Every clinic has a number for this, and most treat it as one operational figure — twelve per cent, say, applied evenly across the schedule and reviewed once a year. It is almost never even. Appointments booked at three in the morning through a social advert do not attend at the rate of appointments booked by somebody who rang during office hours and asked two questions first. The spread between sources can be wide enough to reverse a ranking built on bookings entirely, and it is wider in clinics that have made online booking genuinely frictionless.

Nothing in an advertising account can see this, because the conversion fires when the booking form submits and nobody returns a week later to amend it. The clinic knows, in the sense that the front desk knows and has said so repeatedly, and that knowledge does not reach the person setting next month's budget in any form anybody can act on. The gap between booked and attended is where a cheap lead source quietly becomes an expensive one, and it stays quiet because the two numbers live in two systems that were never asked to agree.

Because only attended, paid visits appear in the billing export, the report measures the far side of that gap by construction. Use the appointment export as the lead file, each booking tagged with its source, and booked and paid counts sit beside each other per channel, which turns the front desk's impression into a figure somebody can defend in a meeting. A clinic finding that one source books well and attends badly then has a specific thing to fix — a reminder policy, a deposit, a landing page that sets expectations — rather than a general suspicion that some leads are worse than others.

A clinic running at capacity is in an unusual position: more inquiries do not produce more revenue, they produce a longer wait, more cancellations and a worse experience for the patients already booked. At that point advertising has stopped being a growth lever and become a selection mechanism. The question is no longer which channel brings the most patients, because the clinic cannot see more patients. It is which channel brings the patients worth giving a slot to, and which one can be switched off tomorrow without the schedule noticing it has gone.

Lead counts cannot answer that at all — they are a measure of the thing that is no longer scarce. What answers it is revenue per patient by source, which means joining the billing export to the inquiry file and letting the averages fall out of it. A source delivering two hundred inquiries a month at a low average is a different proposition from one delivering forty at four times the value, and when appointments are rationed the second is the one that deserves the budget. In a half-empty clinic the first would be perfectly sensible.

Clinics that run this exercise usually end up spending less and earning more, which is an uncomfortable result to present because it contradicts the direction every marketing conversation assumes it is heading. It holds up under challenge for a structural reason: the unmatched revenue stays in Direct / Unknown rather than being shared out across the channels under discussion. Nobody defending the campaign that lost can argue the model quietly moved its revenue somewhere else, because the report never redistributes anything it could not prove belonged to a source.