Attribution,and how long a patient takes to book

What you will have at the end.

A measured booking delay for each service category, a lookback window chosen from it rather than from a default, and a clean source report with the false sources removed.

Attribution is the argument about who gets the credit. It matters because the credit decides where next month's budget goes, and because the default settings were designed for people buying shoes, not for someone deciding whether to finally deal with their knee.

First, measure your delay

1. Find the real gap between first contact and booking

   Almost every attribution mistake in a clinic comes from assuming this gap is short. For a dental cleaning it might be two days. For an elective procedure someone has been avoiding for years it can be four months, and the campaign that started it was switched off long before the booking arrived.

   Do not guess it. Measure it, per service category, because they differ enormously.

   —Pull the last hundred new patients per category with their first-contact date and their first-appointment date.

   —Subtract. Build the distribution, not the average.

   —Read the median and the ninetieth percentile. The median tells you the typical patient. The ninetieth tells you how long your window has to be to catch most of them.

   In practice

   Use the median, never the mean. One patient who booked eleven months after first contact will drag an average into nonsense.

Choosing the window

1. Set the lookback from your ninetieth percentile, rounded up

   The rule is simple. Whatever your ninetieth percentile delay is, round up to the next available window setting and use that.

   Set it in each ad platform and in your analytics, and write down the date you changed it. Changing the window changes every historical comparison, so the date is the only way to explain a step in your own charts later.

   —Set it once, per category if the platform allows, and leave it for at least a quarter.

   —Do not compare a month on a thirty day window against a month on a ninety day window. That comparison is meaningless and it will be made by somebody.

2. Understand what each model does before you argue about it

   Attribution models are not right and wrong, they are different questions. Know which question each one answers and the arguments get shorter.

   —Last click: who was the last touch before booking. Answers where do I close. Undercredits everything early.

   —First click: who introduced them. Answers where do I find people. Undercredits everything late.

   —Linear: everyone shares. Answers nothing precisely but is hard to game.

   —Data driven: the platform decides, using its own model you cannot inspect. Answers the platform's question, and tends to flatter the platform.

3. Read two models side by side, always

   Never look at one model alone. Put last click and first click next to each other for the same period, and the shape of the difference tells you more than either number does.

   —A channel large on first click and small on last click is finding patients. Cutting it will quietly hurt everything else in two months.

   —A channel large on last click and small on first click is closing. It is real, but it is not where growth comes from.

   —A channel large on both is genuinely carrying the clinic.

   —A channel small on both, with meaningful spend, is the one to question.

   In practice

   Be careful with the platform's own data driven model. It has an interest in the answer, and when only one conversion action is defined it will attribute almost everything to itself. Cross-check it against last click before moving money.

False sources

1. Find the source taking credit for everyone else

   Open your source report and look for anything that is not a place a patient could plausibly have discovered you. Your own payment provider. Your booking tool. Your own domain. A generic direct bucket that is implausibly large.

   Each of these is a technical artifact where somebody left your site, came back, and the return was recorded as a new arrival from that service.

   —A payment provider appearing as a traffic source means people leave to pay and return. Every one of those bookings has had its real source erased.

   —Your own domain in the source list means a hop between two parts of your own site is being counted as an arrival.

   —A booking tool on another address does the same thing.

2. Fix them with the exclusion list

   Every analytics property has a referral exclusion list. Adding a domain to it tells the system that a return from there is a continuation of the same visit, not a new one.

   —Add your payment provider, your booking tool, your own domain and any subdomain you use.

   —Do this before you spend a month analyzing a source report, not after.

   —Note the date. Your source report will visibly change shape and somebody will ask.

   In practice

   One clinic we know had a third of its online revenue attributed to Organic Shopping, which turned out to be entirely its payment provider. Nothing was wrong with the marketing and everything was wrong with the report.

3. Tag every link you control

   Anything you can put a tag on should have one. Untagged links fall into the direct bucket and become invisible.

   —Every ad, every email, every link in a social profile, every QR code on a printed card.

   —One naming convention, written down, used by everyone. Source in lowercase, medium from a fixed short list, campaign named after the campaign.

   —Inconsistent casing creates two rows for the same thing, and the report then understates both.

4. Accept what you cannot attribute, in writing

   A share of your patients will never be attributable. Somebody saw a post, told a friend, the friend searched your name six weeks later. There is no honest way to assign that.

   The professional move is to state the unattributable share and hold it steady, rather than distributing it silently across the channels you can measure.

   —Report it as its own line, named plainly.

   —Watch whether it grows. A growing unattributable share usually means something broke in your measurement, not that word of mouth improved.

   —One direct question at intake, how did you hear about us, with a fixed list of answers, gives you a sanity check that no analytics can.

When it goes sideways