Read Your Front Office Numbers

There is no shortage of front-office metrics. Between a phone system and a practice management system a small practice can generate 80 of them without trying, arranged on dashboards nobody has looked at since the week they were set up.

The problem is not that the numbers are wrong. It is that almost none of them change a decision. A metric that goes up or down without anyone doing anything differently is a cost with no return, and a dashboard made mostly of those trains people to ignore the whole screen, including the parts that matter.

Eight lessons below, producing one page. Six numbers, each with a written definition, a stated denominator, and a note on what you would actually do if it moved. Plus, deliberately, a list of what you have decided to stop looking at.

Why most front-office dashboards go unread#

A dashboard gets ignored for a specific reason: the numbers on it are not attached to actions. Nobody can say what they would do if answer rate fell 4 points, so it falling 4 points produces a shrug, and after 3 months of shrugs the screen stops being opened at all.

The fix is not fewer numbers for their own sake. It is that every number on the page has to have a written sentence beside it naming what you would do if it moved in the wrong direction. If you cannot write that sentence, the number does not go on the page.

That test does most of the work in this guide. It is also the reason lesson 7 exists, because deciding what to stop tracking is harder than deciding what to track and nobody ever does it.

If you cannot write down what you would do when a number moves, that number does not belong on the page.

01Pick six numbers, and know what each one is for#

Six is not a magic figure, it is roughly the number a person can hold in mind and genuinely act on. The point is the pairing: every metric has an owner and a stated response, or it does not qualify.

Read the middle column as the reason each row survived the test in the section above. Every one of them answers a different question, and a page where two rows answer the same question has a redundant row rather than a corroborated one.

Two of these come from your phone system, three from your practice management system, and one needs both. None of them needs new software, and the setup cost is the definitions rather than the collection.

Assign an owner per row before you go further. A number without a named person becomes everybody's and therefore nobody's, and the monthly review turns into a reading exercise.

The third column is the one to argue about, and it is worth arguing about now rather than during a month when the number has moved. Writing "listen to 10 calls, do not adjust the number" beside enquiry to booking rate is a commitment made calmly, and it is what stops a bad month producing a panicked change to a definition or a target.

Note also which rows you cannot yet produce. Most practices can compute 4 of these 6 in an afternoon and need a small amount of work for the other 2, usually the ones joining two systems. Start the page with the 4 rather than waiting until all 6 are available, because a partial page in use beats a complete page in a plan.

Two of the six have published comparators, which is rare enough to use. A 2022 study of 6 Veterans Affairs medical centres chosen for above-average primary care access records the standards those sites were held to: an average speed of answer of 30 seconds or less, and call abandonment under 5% [4]. They come from staffed call centres working to a mandate, so they are a direction rather than a target for a practice with 2 people on the desk. They are still the only external numbers on this page that were not produced by someone selling software.

02Fix the denominator before you trust the numerator#

Most front-office metrics are ratios, and almost all the confusion in them comes from the bottom half. A practice that changes its denominator quietly has produced a trend that is entirely an artefact of definition.

1. For answer rate, decide whether the denominator is all inbound calls or all inbound calls excluding those under 10 seconds.
2. Decide whether repeat calls from the same number within 48 hours count once or several times.
3. Decide whether internal and supplier calls are excluded, and how you identify them.
4. Write each decision on the definitions page, with the date.
5. Never change one mid-year. If you must, restate the prior months rather than creating a step in the chart.

Step 1 has the biggest effect. Misdials and wrong numbers are typically several percent of inbound volume and they inflate your apparent miss rate, and a practice that starts excluding them halfway through the year appears to improve without doing anything.

Step 5 is the discipline that makes a scorecard worth keeping. A number that is comparable across 12 months is worth more than a better-defined number that is comparable across 3.

A comparable number beats a better number. Change a definition and you have started a new series, whether you meant to or not.

The 48 hour window in step 2 is a judgement rather than a standard, and any window between 24 hours and 7 days is defensible depending on how long your patients take to decide. Pick one, write down why, and stop revisiting it. The consistency is what carries the meaning; the specific number carries almost none.

Step 3 is the one that quietly corrupts small practices. Supplier calls, lab calls, and internal transfers can be 10% or more of inbound volume at a practice taking 40 calls a day, and they arrive during staffed hours, which means they inflate exactly the band you are trying to read.

Where a trend comes from nothing

A real change

• A receptionist left and was not replaced
• Opening hours changed
• Marketing spend went up and volume followed
• A rota gap was closed

An artefact of definition

• Sub-10-second calls started being excluded
• Repeat attempts began collapsing into one
• Supplier calls were filtered out mid-year
• The phone vendor changed how it counts abandoned

03Split answer rate by staffed and unstaffed hours#

A single answer rate is the least useful number your phone system produces, because it averages two situations with completely different responses. Splitting it is the highest-value 10 minutes in this guide.

1. Build a table of your actual opening hours, including lunch closures.
2. Classify every inbound call as staffed, rota gap, or unstaffed.
3. Compute answer rate separately for each of the 3.
4. Track the 3 separately, forever, and never average them back together.

The middle band is what makes the split pay for itself. A practice missing calls in a 45 minute lunch gap has a rota problem it can solve this week at no cost, and that finding is completely invisible inside a single blended figure.

Sort the unstaffed band by hour once, when you set this up, and then leave it. Most practices find their out-of-hours volume concentrated rather than spread, often in the 2 hours after close, and that shape changes what covering it is worth. A gap that is really 10 hours a week of concentrated demand is a very different proposition from 128 hours of thin trickle, and the monthly number cannot tell them apart.

A standard schedule of 8 hours a day, 5 days a week covers 40 hours out of 168 hours in a week, so 128 hours sit in the unstaffed band. Your answer rate there is a fact about your coverage arrangement rather than about your staff, which is why mixing the two produces a number that reflects on the wrong people.

A blended answer rate hides the one finding you could act on for free.

04Measure time to first human response honestly#

This is the number practices most often measure wrongly, and the error always runs in the flattering direction because the systems report the convenient version rather than the real one.

1. Define it as the time from the patient first trying to reach you until a person actually spoke to or messaged them.
2. Not until a voicemail was left. Not until a message appeared in a queue. Not until someone opened it.
3. Measure the median rather than the mean, because a handful of multi-day outliers will otherwise dominate.
4. Report the 90th percentile beside it, because that is the experience of the patients you are losing.
5. Track it separately for enquiries arriving in and out of hours.

Step 4 is where the useful information lives. A median of 40 minutes with a 90th percentile of 2 days describes a practice that handles most things quickly and abandons a tail completely, and the tail is where the complaints and the lost patients are.

The gap between the median and the 90th percentile is itself worth watching as a number. A widening gap almost always means a queue is being worked from the top rather than in order, which is a process fix rather than a resourcing one.

Two practical notes on collecting it. Most phone systems will not compute this for you, because the second timestamp lives in whatever system your staff actually work from, so this is usually a monthly join of two exports rather than a report you can subscribe to. And 20 sampled requests a month is enough: this does not need to be measured exhaustively to be useful.

Median against 90th percentile

40 min

median time to first human response

the experience most patients have

2 days

90th percentile

the experience of the patients you lose

Widening

gap between the two

a queue worked from the top, not in order

05Track the recall list as a rate, not a count#

For any practice with a recall cycle, this is frequently the largest recoverable number on the page and the one least often tracked. The count of recall-due patients tells you almost nothing. The share contacted is the whole story.

1. Count patients whose recall date fell in the month.
2. Count how many were contacted at all, by any channel, in that month.
3. Divide the second by the first. That is the number.
4. Separately count how many of those contacted booked.
5. Track both, since they fail for different reasons and have different fixes.

Step 2 needs a decision about who counts as due. A patient 3 months past their recall date and one 3 years past are both technically due, and lumping them together produces a denominator that grows forever and a rate that only ever falls. A reasonable rule is to count patients due within the last 12 months and hold the older ones in a separate reactivation list with its own number.

Steps 3 and 4 measure different things and get conflated constantly. A low contact rate is a capacity problem: nobody had time. A high contact rate with a low booking rate is a messaging or timing problem, and buying more capacity will not touch it.

The context here is real. In 2025 one third of dentists reported they were not busy enough, up from one quarter in the fourth quarter of 2024 [1], which is a lot of practices with available capacity and a list of people who are due.

Expect the first measurement to be uncomfortable. Practices that have never computed this routinely find a contact rate well under half, and the value is that it is a problem with a known solution rather than a mystery.

Define contacted before you count it, because the definition decides the number:

That is a deliberately strict definition and it produces a lower, more useful number. A practice reporting 80% contacted on a definition that includes unanswered voicemails is reporting its own activity rather than any effect on a patient, which is the same error as counting links sent instead of bookings completed.

Counting messages you sent is measuring your activity. Counting replies is measuring your reach.

06Split no-show rate by booking horizon#

An overall no-show rate is a number you can watch and cannot act on. Split by how far ahead the appointment was booked, it becomes diagnostic, because it tells you whether your problem is forgetting or something else entirely.

1. Bucket completed appointments by lead time: same week, 2 to 4 weeks, over 4 weeks.
2. Compute no-show rate within each bucket.
3. Compare the buckets rather than watching the total.
4. Note your overall baseline, because it caps what any intervention can achieve.

If misses concentrate in the long-horizon bucket, forgetting is genuinely your problem and reminders address it. If the rate is flat across all 3 buckets, something other than memory is driving it, and reminders will disappoint you regardless of what a vendor's case study says.

There is a cheap follow-up worth running once when the buckets come out flat. For 2 weeks, have whoever rebooks a no-show ask a single question and record the answer: what stopped you coming in. Sort the replies into forgot, could not, and changed my mind. The could not column frequently points at something you can genuinely fix, such as appointment times that assume nobody works, and that fix costs nothing.

Step 4 sets your ceiling honestly. Pooled trial evidence puts attendance at 67.8% with no reminder and 78.6% with a text [2], an improvement of about 10.8 percentage points from a very high starting miss rate. A practice already at 6% does not have 10 points available to recover.

No-shows by booking horizon

	Concentrated in long horizon	Flat across all three
What it suggests	Forgetting	Circumstance or intent
Do reminders help	Yes, this is what they treat	Very little
First thing to try	A second reminder earlier in the cycle	Ask 20 patients what stopped them
What to expect	Several points, more if the baseline is high	Almost nothing

07Watch repeat callers per distinct caller#

This is the number nobody tracks and it changes the interpretation of everything above. Divide total missed calls by the count of distinct numbers behind them, using the same 48 hour window you defined in lesson 2.

1. Take the month's missed calls after cleaning.
2. Group by caller number inside a 48 hour window.
3. Divide rows by distinct numbers.
4. Track the ratio monthly alongside the raw count.

A ratio near 1.0 means most people who could not reach you tried once and stopped, which points at a coverage problem and at callers with no particular loyalty. A ratio of 1.7 or higher means fewer people are affected but they are trying repeatedly, which points at capacity during busy periods.

The two situations have opposite fixes and identical missed-call totals, which is why the raw count on its own has misled so many practices into buying the wrong thing.

Two practices, identical missed-call totals

	Ratio near 1.0	Ratio 1.7 or higher
What happened	Most callers tried once and stopped	Fewer people, trying repeatedly
What it points at	Coverage, and low-loyalty callers	Capacity during busy periods
First thing to try	Cover the unstaffed hours	Look at hold times and staffing peaks
What the raw count says	Identical	Identical

Watch the ratio's direction as well as its level. A ratio that is falling over several months while the raw count holds steady means more distinct people are being affected, which is a worse situation than the flat total suggests and is invisible without this row.

08Write down what you are not tracking#

The list of abandoned metrics is the part that makes the scorecard durable, because without it the page grows back. Everything removed should be removed on the record, with a reason.

• Total call volume on its own, since it moves with marketing and says nothing about performance
• Average call duration, which is not better when it is shorter or when it is longer
• Any vendor-reported metric with no denominator you control
• Anything reported as a relative improvement rather than an absolute rate
• Satisfaction scores based on fewer than 30 responses a month
• Any number nobody could name an action for during lesson 1

Row 4 is worth internalising beyond this page. A 14% relative improvement and a 14 percentage point improvement are wildly different claims, and the first is routinely presented as the second, a distinction worked through in what the evidence on no-show rates actually supports.

Row 3 catches a specific problem with vendor dashboards. A number whose denominator lives in someone else's system cannot be checked, cannot be compared with your own records, and will change definition without notice when they ship a release.

Row 5 is not a criticism of asking patients what they think. It is that a satisfaction score built from 11 responses moves several points when 2 people answer differently, so it reports sampling noise as though it were a trend. Either collect enough responses for the number to mean something or collect the comments and read them, but do not put a volatile average on a page people are meant to act on.

Deleting a metric is harder than it sounds, and the reason is social rather than analytical. Somebody built that chart, and removing it reads as a judgement on their work. Doing the removals in one deliberate session, on the record, with a reason written beside each, is much easier than doing them one at a time.

The finished page

• ✓Answer rate, split into staffed, rota gap, and unstaffed
• ✓Time to first human response, median and 90th percentile
• ✓Enquiry to booking rate
• ✓Recall contacted rate, and booked rate within it
• ✓No-show rate by booking horizon, in 3 buckets
• ✓Repeat callers per distinct caller

One number belongs on the list of things you are not tracking, and it explains more variance than most of the six. MGMA polled 357 practices in May 2025 and found front-office roles, receptionists and patient service representatives, were the most frequently cited turnover hotspot, including at practices whose overall turnover was improving [3]. A page of front-office metrics that does not record who was on the desk will attribute a staffing change to a process change sooner or later.

Build the page, then leave it alone#

One page, six numbers, three months of history beside each, and the definitions on the back. Thirty minutes on the first working day of the month, done by the person who owns the row rather than by whoever has a spare afternoon.

Who reads it matters as much as who fills it in. A page produced monthly and seen by one person is a private record; a page walked through in 10 minutes at an existing meeting is a management tool. Attach it to a meeting that already happens rather than creating one, because a new recurring meeting is the first thing to be cancelled in a busy month.

Keep the history on the same sheet rather than in a separate archive. Three months of prior values beside the current one is what turns a number into a reading, and the moment history lives somewhere else it stops being consulted, which returns you to reacting to single months.

Resist adding a seventh number. Every scorecard grows if permitted, and a page with 14 numbers has the same problem as the dashboard it replaced. When something genuinely warrants tracking, the honest move is to swap it for a row you can retire, not to append it.

One caution on the annual definition review. It is genuinely tempting to improve a definition you now understand better, and the improvement is usually real. Weigh it against the cost, which is that you lose comparability with everything before it, and a scorecard whose main value is a 3 year trend should change definitions rarely and loudly.

Review the definitions once a year rather than the numbers, and treat a definition change as an event: restate the history or start a new series. The numbers are only worth anything because they are comparable, and comparability is the thing that quietly breaks.

If you have not measured your baseline yet, size your missed-call gap produces most of the phone-side inputs in one afternoon, and running a 30 day pilot explains how these same numbers behave when you are evaluating something new.