Conversion Rate Optimization for Clinics and Health Platforms

The number you want to improve is recorded somewhere we cannot read. A patient books in a scheduling product, or calls the desk, or turns up because a doctor sent a letter. Session data sits on one side of that wall and the booking sits on the other. Joining them is a legal question before it is an engineering one. Your data protection officer or privacy lead decides it, in writing, before any identifier crosses. Which fields may be joined, on what legal basis, and who may see the result. Until that answer exists, the two sides stay apart.

So the first honest conversation is about what we are permitted to count.

The measurable funnel ends early

We instrument up to the last point it is safe to instrument. A visit to a service page, a click on the booking button, a form opened, a form submitted. After that we hand off. What we do not do is pass a third-party script the field naming the condition the appointment is for. At that moment the event is health information about an identifiable person, and it has left your control. In the United States the framework is HIPAA. The HHS bulletin on online tracking technologies says a tracking vendor that receives protected health information is a business associate. The provider needs a signed business associate agreement and a Privacy Rule permission first. Otherwise it needs a HIPAA-compliant authorisation from each patient. The same bulletin says a cookie banner is not that authorisation. A federal court vacated one part of it in 2024, in American Hospital Association v. Becerra. That part covered an IP address joined to a visit to a public page about a condition. HHS still publishes the rest, including its guidance on patient portals and its appointment booking example. In the EU, Article 5(3) of the ePrivacy Directive requires consent before a tag stores or reads anything on the device. The exception that matters here is storage strictly necessary for a service the visitor asked for. Article 9 of the GDPR makes health data a special category. Processing it is prohibited unless an exception applies, such as the patient’s explicit consent.

The consequence is that every event we keep is first-party and written by your own backend, with the health-specific fields never leaving it. That is slower to build than dropping a tag in a tag manager. It is also the only version that survives an ad blocker and a legal review in the same week.

Telephone bookings are the part nobody tests

A booking made by phone is invisible to every test running on the page. If calls are a real share of your appointments, a variant that wins on form submissions can lose on appointments. It pushes people away from the number, and the dashboard congratulates you. So calls count as an outcome from the first week, at minimum as a call started from a page. A design that buries the phone number is a guardrail breach, and gets stopped.

The cheap test here is the slow one

Everywhere else, changing a headline is the first thing anyone tries because it takes an afternoon. Here a headline is a claim about a treatment or an outcome. A claim goes through clinical and legal sign-off before a visitor sees it. Two weeks of review for a change that runs three and comes back flat is a poor trade. It lands on the first test of any programme that was not planned around it.

So the review queue goes into the plan. Clinical copy is batched, submitted once, and tested as a set after it clears. Nothing sits waiting on a reviewer one sentence at a time.

What we test while the clinical copy is frozen

Everything the clinic does that is not a clinical claim, which is a longer list than it sounds.

• What the booking form asks for, in what order, and how much of it could be collected after the appointment is held
• Whether real availability is visible before someone commits to filling anything in
• Insurance and self-pay clarity, which is where people leave without telling you why
• Current wait times and what happens next, stated on the page instead of in a confirmation email
• Which location a page defaults to for a visitor who has not chosen one
• The route for somebody who already holds a referral letter, which is a different path from somebody starting cold

None of it touches a clinical claim. All of it clears review quickly. And it is where the abandonment sits.

When we tell you not to run a programme

One clinic does not generate the weekly volume an A/B test needs. A group that does has service lines with separate owners and separate release trains. The test you designed cannot ship to all of them at once. When the arithmetic says your first test finishes in fourteen months, you hear that in the first meeting. What we sell instead is the research, the instrumentation and the defect list. Those are worth buying. An underpowered test that reports a winner is reporting noise, and acting on it leaves you worse off than having run nothing.