Which channels filled the service lines?

In a health system the majority of episodes arrive by physician referral, insurance network, or proximity — not by advertising. Any tool reporting a high attributed share for a hospital is redistributing that volume, and the resulting numbers cannot support a budget decision.

A large Direct / Unknown bucket is the correct answer here, and seeing its true size is what makes the attributed portion meaningful. The question worth answering is not what share advertising produced, but which sites it moved and by how much.

System-wide attribution averages away everything that matters. Orthopedics, cardiology, maternity and imaging have entirely different economics, different audiences and different competition, and a channel that works for one may do nothing for another.

A site column mapped as Location turns one uninformative number into a comparison the marketing team can act on. A service-line column would be the other obvious cut, but which department treated somebody is health information, and it does not belong in a file sent to a vendor that signs no business associate agreements.

A health system's revenue divides into episodes that would have arrived regardless and episodes that can be influenced. Emergency presentations, transfers and referred admissions are not a marketing outcome in any useful sense of the word. Elective procedures, screening, diagnostics and the service lines patients choose for themselves are, and they are frequently where the contribution margin sits. Reporting one attributed share across the whole book buries the only part of it a campaign could ever have changed. Averaged together, the influenced volume is small enough to vanish into the rest, and the report reads as though marketing did nothing worth funding.

Separating them happens before the export rather than inside it. Filter the billing extract to elective episodes, so the file that leaves the hospital never names an episode type or admission route, and the report stops trying to answer an unanswerable question — what share of all hospital revenue advertising produced — and answers a narrower one instead: within elective volume, which channels produced booked and completed episodes, and at what value each. The episode type is already recorded for billing, so the filter needs nobody to add a field to a clinical system, and the file that results carries a contact detail, an amount and a date and nothing else.

The attributed share inside that slice looks nothing like the system-wide figure, and it is the figure somebody can act on without qualifying it. It also survives a finance review, because nobody is being asked to accept that a paid search campaign caused an emergency admission. Narrowing the denominator is not a way of flattering the numbers; it is how the numbers become about a decision somebody is actually making. The system-wide number is still worth printing, because it shows how much of the book was never in play. It should simply not be the number anybody plans against.

Very little hospital demand converts on a website. A campaign produces a call or a request, the access center receives it, and between that moment and a completed episode the patient has to be scheduled, verified, cleared and then actually attend. Every one of those steps loses volume, and the losses are not distributed evenly across the channels that produced the inquiries in the first place. A campaign aimed at a service line with a six-week wait and one aimed at a clinic with capacity next Tuesday look identical at the point of inquiry and nothing alike at the point of billing.

This is why inquiry counts mislead so badly in a health system. A channel producing large numbers of requests that cannot be scheduled inside an acceptable window is producing frustration rather than revenue, and it will still look like the best performer on any report built from form submissions. One producing fewer requests that become billed episodes is doing the job. Only a match against the billing record separates them. The same logic covers non-attendance, which concentrates in some referral patterns far more than others and is invisible in any count taken before the appointment.

The access center's own records are the second file. Whatever logs inquiries and where they came from — a call platform, a scheduling queue extract, a campaign's form submissions — is a list of people with contact details, and that is the entire requirement. Anything you cannot source stays in Direct / Unknown, which for a hospital is correctly the largest block on the page and should stay that way. Several sources are combined into the one lead file, each row tagged with its source, and they do not have to be de-duplicated against each other first: a contact that appears twice is credited to its earliest row.

The match needs three things from the revenue file: a contact detail, an amount and a date. Nothing else is read at any point. Diagnoses, procedure codes, clinical notes, payer detail and anything resembling a record of care play no part in the calculation and should not leave the systems holding them. Restraint in the export is not a limitation to be worked around here; it is the correct way to run this, and it shortens every internal approval that follows.

A site name is the one addition worth making, because it turns one uninformative network figure into a comparison somebody can act on. It is an administrative category rather than a clinical fact and says nothing about an individual's care, and it is already present in the finance extracts most systems produce for internal reporting every month. A service-line or department label is a different matter: which department treated somebody is health information, so it stays out of the file. If the site is genuinely unavailable the report still runs without it. It simply has less to say, and it says the same thing about every site at once.

Everything uploaded sits in its own workspace, encrypted in transit and at rest, and can be deleted outright rather than marked as deleted. A retention window can be set so that imports purge on a schedule without anyone having to remember to do it. We act as a processor and a DPA is available, and the smaller the file you send, the shorter that conversation is with whoever has to approve it. Access is by role, so the people who upload a file and the people who read the report do not have to be the same set of people.