Reading funnel conversion and engagement metrics

Two cardiology campaigns land on your desk with the same headline: 0.08% of impressionsimpressionsThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.View full definition → became a booked consult. One is losing people between the ad and the service page. The other is losing them inside the scheduler at insurance verification. The headline number cannot tell you which, and the two fixes share nothing: new creative and a bid change in the first case, three evening slots and a pre-check tool in the second. Reading a funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition → means reading the step rates, then arguing about which one is genuinely broken.

This lesson is the arithmetic and the diagnosis: how step conversion behaves, which engagement signals warn you early, and the specific ways hospital funnel numbers lie to you.

Step conversion versus overall conversion

Take the six stages the mapping lesson lays out, impression through kept visit. Each arrow carries its own rate, and the rates multiply. Multiply 2% click, 40% engaged, 20% lead action, 50% booked and 81% kept and you get 0.065% of impressions ending as a patient in a chair. That is why the end-to-end number is close to useless as a diagnostic: it moves whenever any of five things moves.

Multiplication also sets your priorities. A step sitting at 50% has 50 points of headroom; a step at 81% has 19. Ten points recovered at the booking step flows through untouched to kept visits, while ten points added to click-through gets diluted by everything downstream. Work the low rate in the middle, not the one that is easiest to report.

Why hospitals leak differently

Cardiology skews older and phone-driven. If your only lead event is a form submission, you can miss half the demand and then read the gap as a landing pagelanding pageA standalone web page built for a single campaign goal, designed to maximise conversions by removing distractions and focusing visitors on one action.View full definition → problem. The mirror failure is double counting: a patient who calls, hangs up, then starts the online scheduler registers two lead actions and one booking, which drops your lead-to-booking rate for reasons that have nothing to do with your intake team. Deduplicate on phone number or email before you draw conclusions from that step.

Instrumenting the funnel

Three instruments cover most of the gap between an ad platform's report and a kept visit.

1. Call tracking

Call analytics vendors assign a unique number per campaign or channel, and a dynamic number insertion (DNI) script swaps the displayed number according to how the visitor arrived.

// Visitor from paid search sees a tracked number
if (source === "google_cpc") {
  document.getElementById("phone").innerText = "(555) 010-2201";
} else if (source === "social") {
  document.getElementById("phone").innerText = "(555) 010-2202";
}

Calls from that first number now belong to Google paid search, and you can read volume, duration (a rough proxy for quality) and call-to-booking rate. Two known blind spots: patients who call the switchboard number from your Google Business Profile listing bypass DNI entirely, and patients who note the number and call three days later from a landline may land outside your session window.

The compliance layer is not optional. In the US, anything discussed on a tracked call is protected health information, so recordings and vendor data flows need a Business Associate Agreement making the vendor legally accountable. HHS Office for Civil Rights issued a bulletin in December 2022 on online tracking technologies in healthcare, and several US health systems have since settled class actions over marketing pixels that passed identifiable data to ad platforms. In Europe, GDPRGDPREU regulation governing how organizations collect, store and use personal data, with fines tied to global revenue for breaches.View full definition → requires a lawful basis and, for recording, explicit consent. A funnel you cannot legally join end to end is a design constraint, not an excuse.

2. Form fills

Track submissions, and track abandonment with equal care. A form asking for insurance details, referral source and symptom description before it asks for a name will bleed users; measure completion field by field and you usually find one question doing most of the damage.

3. Scheduling drop-off

Instrument each scheduler step: slot selected, details entered, insurance verified, confirmation. Insurance friction stops people cold, which is why this is often the deepest leak in a specialty funnel. Marketplace and vendor-hosted booking (Doctolib-style widgets, and Doctolib sells that booking layer, so read its own conversion claims accordingly) adds a wrinkle: the confirmation event lives in the vendor's system. Unless you pass it back, your ad platforms never learn which clicks produced bookings, and their optimisation runs blind.

For event-based funnel tracking, see Google's free documentation on measuring conversions in GA4.

Reading the drop-off: a worked example

All figures below are illustrative, chosen to show the arithmetic, not sector data.

Overall conversion from impression to booked appointment: 32 / 40,000 = 0.08%.

The signal is the 50% at the booking step. Sixty-four people asked for care and thirty-two got an appointment. Before blaming intake, check the denominator: at 64 lead actions, a handful of duplicates or one wrong-number call moves that rate by several points. Below roughly 30 events a month, step rates are noise, and you should read a rolling quarter instead of arguing about last week.

Turning a leak into a number you can act on

InboundInboundA strategy that attracts prospects organically via valuable content (blog, SEO, social) rather than interrupting them.View full definition → calls in this campaign convert to bookings at 65%; the online scheduler converts starts at 30%. On 40 scheduler starts:

• At 30%: 12 bookings
• At 50% after fixing insurance verification: 20 bookings

Eight extra consults from the same traffic. The second-order effect is the one to flag to operations: eight consults land downstream imaging and follow-up load on a clinic that did not ask for it, and if the schedule is already full, the fix converts a booking leak into a wait-time problem and a fresh crop of no-shows.

Benchmarks: what "good" looks like

Treat everything here as directional estimates that vary by specialty, geography, channel and year.

• Landing page conversion (visit to lead) in healthcare paid search is commonly cited in the 3% to 8% range. High-consideration specialty care sits at the low end.
• Call-to-appointment rates for inbound healthcare calls are frequently reported around 25% to 40%. Phone leads usually beat web forms on intent.
• No-show rates in outpatient specialty care are widely estimated at 15% to 30% depending on setting and reminders. Doctolib built much of its practice pitch on automated SMS reminders for exactly this reason.

A benchmark earns its keep when a stage is wildly off. A call-to-appointment rate of 8% against a 25% to 40% range points at hold times, staffing and slot availability, not at your ads.

Engagement metrics that predict conversion

Engagement signals give you lead quality before bookings arrive: physician bio and video views, insurance page views (high intent and a frequent abandonment point), scroll depth on a procedure page.

The counter-example matters more. Cleveland Clinic's consumer health content pulls traffic at a scale most hospital service lines never approach, and almost none of it is booking intent: symptom-checkers and condition explainers attract readers from everywhere, including outside the catchment area entirely. Engagement predicts conversion only inside a single intent class. Compare time on page across an editorial article and a "find a cardiologist" page and you will optimise toward readers.

AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition →: connecting the leak to the source

Stage data says where you lose people. Attribution says which channel those people came from, so you stop funding traffic that fills the top and never converts.

• Last-touch gives the final channel full credit: simple, and it starves awareness activity.
• First-touch credits the opener: useful for demand generationdemand generationMarketing activities designed to attract and capture contact information from prospects interested in your offer, creating a pipeline of potential customers.View full definition →, useless for judging what closes.

Your tooling constrains the choice more than theory does. Google retired several of its multi-touch models in GA4 in 2023, leaving a narrower set, so a model you read about in a 2019 playbook may not be available to you.

Watch for the channel that looks strong on clicks and weak on kept appointments. The failure mode is quiet: optimise ad bidding toward form fills and the platform gets very good at finding people who fill forms, including the ones who never attend. Feed the booking event, and ideally the kept-visit event, back into the platform and the same budget starts buying patients rather than events.

Tag campaigns with UTM parameters (?utm_source=google&utm_campaign=cardio_q1), then join that to call tracking and scheduler events. Kept-visit status usually lives in the EHR, on the far side of a consent and governance boundary, so agree the join key and the legal basis before you promise leadership an end-to-end number.

Putting it together

The campaign in the opening scene was losing half its qualified demand at the booking step. Once the step rates were visible, the answer was not more impressions. It was evening slots, an insurance pre-check, and a booking event fed back to the ad platform. Same spend, more consults.

Key Takeaways

• Step rates multiply, so a healthy top of funnel hides a broken bottom. Work the lowest rate in the middle, not the metric that reports best.
• Track calls and deduplicate them against forms and scheduler starts, or your lead-to-booking rate will be wrong for structural reasons.
• Below roughly 30 events a month, a step rate is noise. Read a rolling quarter.
• The deepest specialty leak is usually insurance verification and slot availability, and fixing it puts real load on clinic capacity.
• Optimise ad platforms toward bookings and kept visits, not form fills, and settle the EHR join and its legal basis before promising end-to-end numbers.