The five numbers worth tracking on a clinic front desk
Most clinic messaging dashboards report volume, which nobody can act on. Five numbers actually change decisions, and one popular metric quietly causes harm.
Updated 1 September 2026 · 4 min read
A dashboard that reports messages handled is telling you how busy the inbox was, which you already knew. These five tell you what to do next.
1. Median first response time, split by hour
The single most predictive number in patient messaging, and it must be a median — a single holiday weekend destroys a mean — and it must be split. A clinic with a nine-minute average often has a two-minute figure during opening hours and a fourteen-hour figure overnight, and only the second one is a problem.
Count the first substantive reply, not an acknowledgement. “Thanks for your message, we’ll get back to you” is not a response; treating it as one is how clinics report excellent numbers while losing patients.
2. Conversation-to-booking rate
Of the conversations where a patient asked about a treatment, what share ended in a booked appointment? Measure on conversations, not people, and set a window — seven days is usually right, because most patients who book do so in the same session or the next day.
Segment it three ways and it starts telling you things: by channel, by hour of arrival, and by treatment. A treatment with high enquiry volume and low booking rate is usually a pricing or knowledge problem, and reading ten of those conversations will tell you which.
3. Escalation precision and recall
Two numbers, and clinics almost always track only the easy one.
- Precision — of the conversations that escalated, how many genuinely needed a person? Visible from the escalation queue. Low precision means staff learn to ignore the queue.
- Recall — of the conversations that should have escalated, how many did? Not visible from the escalation queue, by definition. The only way to get it is for someone clinical to read a random sample of conversations the agent handled alone.
Recall is the safety number. Twenty sampled conversations a week is enough to catch a boundary that has drifted, and it is the one measurement that cannot be automated away.
4. Knowledge coverage
The share of questions answered from an existing source, versus those where retrieval came back empty. Every empty retrieval is a named, prioritised piece of documentation work — the system is telling you exactly what to write next, ordered by real demand.
Watch the trend rather than the level. Coverage should climb steeply in the first two months and then flatten. A sudden dip usually means something changed in the world — a new treatment, a new campaign, a competitor’s claim patients are asking about.
5. Escalation response time
How long a patient waits after being told a person will reply. This is the number that quietly undoes everything else: a system that answers in forty seconds and then leaves escalated patients for nine hours has produced a worse experience than a consistently slow human one, because it raised the expectation first.
It has a different owner from every other metric here — it is a staffing and rota question, not a software one — which is precisely why it goes unmeasured.
The metric to stop reporting
Percentage of conversations fully automated. It looks like efficiency and behaves like a hazard. Set it as a target and it will be met, by automating conversations that should have reached a person — because the only way to move it is to make the boundary looser.
If you want a number in that shape, use knowledge coverage instead. It rises for the right reason: because someone wrote the card.
| Metric | When it moves the wrong way, you | Owner |
|---|---|---|
| First response time (out of hours) | Extend automated coverage | Operations |
| Conversation-to-booking rate | Read ten conversations for that treatment | Clinic manager |
| Escalation precision | Tighten a topic rule that is too broad | Clinic manager |
| Escalation recall | Widen a rule, and review the sample weekly | Clinical lead |
| Knowledge coverage | Write the top missing cards | Knowledge owner |
| Escalation response time | Fix the rota, not the software | Operations |
Questions
Asked often enough to answer here
Under a minute is achievable with automation and is what patients comparing several clinics in one sitting reward. The more useful target is closing the gap between your in-hours and out-of-hours figures, because that gap is where enquiries are actually lost.
Sample. Twenty randomly chosen conversations a week that the agent handled without escalating, read by someone clinical, is enough to detect a drifting boundary long before it becomes an incident. It is the one measurement that has to stay human.
No. It is the one metric here that causes harm when targeted, because the only lever that moves it is loosening the clinical boundary. Track knowledge coverage instead — it improves for the right reason and correlates with the outcome you actually want.
Keep reading
Article
The real cost of the message nobody answered
Most clinic enquiries arrive when the clinic is shut. Here is how to size that loss from your own inbox, and what actually changes when it is covered.
Article
When an AI should stop and get a human
Escalation design for clinic messaging: the topic rules, the signals, the handoff payload, and why confidence thresholds alone are the wrong mechanism.
Solution
AI patient messaging for multi-branch clinic groups
One brand, several branches, different prices and different staff. How to run patient messaging centrally without giving every branch the same wrong answer.
See it answer your patients
Bring forty real messages from your own inbox. We will run them live, show you the source behind every answer, and tell you which ones we would refuse to automate.