A clinic dashboard should fit on one screen and take ten minutes. Five numbers: how much you collected, how long the money takes to arrive, how many people didn't show up, how full the schedule was, and whether patients came back. That's it. Everything else is a detail you look at only when one of those five moves.
But before the list, the thing that took me by surprise while researching this — and that no dashboard company will ever put in writing:
I sell dashboards. It would be easier for me to tell you to stare at one every morning. Instead, here are the five numbers — and then the honest account of what actually makes looking at them worth anything, which turns out to have nothing to do with how often you look.
Why revenue isn't one of the five
Start here, because it's the most common mistake. Revenue belongs on the dashboard — but not on the list of things you check and act on. Revenue is the scoreboard. By the time it moves, the thing that moved it happened weeks ago.
The five below are the levers. Each one is upstream of revenue, each one is something you can do something about on a Monday, and each one has a whole guide behind it if you want to go deeper.
The five numbers
Net (adjusted) collection rate
Of the money you were actually entitled to collect — after contractual write-offs — how much did you get? This is the single most important number on the page, because it is the only one that tells you whether work you already did turned into money.
A practice can be busy, well-reviewed and fully booked, and still sit at a 92% adjusted collection rate — three points below the bottom of the range the AAFP calls average. Every one of those points is money you already earned and simply did not chase.
Days in A/R
Total unpaid A/R divided by your average daily charges — a proxy for how long the money takes to arrive. Collection rate tells you whether you get paid; this tells you when. Two honest caveats: it is not literally the age of your oldest claims, so if your charges are growing it will flatter you; and a chunk of the balance is usually owed by patients, not payers. A healthy collection rate sitting next to a 62-day A/R still means cash is arriving too slowly — often because claims are being reworked and eventually paid, which a collection rate happily hides.
No-show rate
Capacity you paid for and did not use. Staff, rent and equipment cost the same whether the chair is full or empty, so what you lose is the contribution margin on the slot — in a business this fixed-cost, that is most of it. It's also the fastest number on this list to move: appointment reminders are cheap, and unlike most of what gets sold to practices, they have actually been tested in randomised trials.
Provider utilisation
What share of your available clinical hours were actually booked and delivered. This is the number that tells you whether your problem is demand (not enough patients) or supply (not enough hours offered) — and owners routinely misdiagnose which one they have.
Patient retention
Whether the patients you already paid to acquire are coming back. It is the slowest of the five to move and the most expensive to ignore, because it compounds — every retained patient is an acquisition cost you don't pay again.
Those five, on one screen, weekly. If you want the longer list — the twelve that matter over a quarter rather than a week — that's the full medical practice KPI guide, and there's a dental-specific version too.
Now the part that matters more: how often should you actually look?
I went looking for the evidence behind "check your numbers weekly." Here is what I found.
Nobody has tested it where it counts
No one has randomised a medical or dental practice to weekly versus monthly review of its business metrics. I looked hard and could not find a single trial — not a weak one, not a flawed one. Anyone who hands you a cadence is handing you a preference dressed as a finding.
The nearest thing to a real test comes from outside healthcare entirely, and it is worth knowing because it cuts against the "look more often" instinct: in a controlled experiment that actually manipulated how often people got feedback, performance followed an inverted U — it improved with more frequent feedback up to a point, and then declined. The authors set out explicitly to challenge the "more is better" assumption. That was a lab task, not a clinic's P&L, so I won't pretend it settles your Monday. But it is the only randomised evidence on frequency anyone has, and it does not say what a dashboard salesman would want it to say.
The closest evidence can't make up its mind
The nearest relevant literature is audit and feedback: the study of what happens when you show professionals data on their own performance. It's the most-reviewed behaviour-change intervention in healthcare — Cochrane has now published four versions of the same review, most recently across 292 studies. On the question of frequency, the same research group has said three different things:
| Paper | Finding on frequency |
|---|---|
| 2012 Cochrane · 140 studies | Feedback may be more effective when it is "provided more than once." |
| 2014 J Gen Intern Med | Feedback appears most effective when "presented frequently." |
| 2025 Cochrane · 292 studies | "Contrary to expectations, repeated delivery was associated with lower effect size." |
Two point one way. The newest and largest points the other — and note the authors' own three words: contrary to expectations. They were surprised too.
Now here is the part that actually settles it, and it's the reason I'd rather write this section than the one my headline promised. In the 2025 review, that frequency result comes from an exploratory meta-regression — a hunt through the data for patterns, not a planned test. And when the authors wrote their conclusions, listing the seven things that make feedback work, frequency appears nowhere on the list. Not as a positive. Not as a warning. They dropped it.
So the honest summary is not "Cochrane now says don't repeat feedback." It is this: after 292 studies, the largest review of feedback in medicine declined to tell you how often to do it. Not because the answer is uncomfortable — because the evidence won't support one. If frequency were the lever everyone assumes it is, it would not keep changing sign, and it would not have been left out of the conclusions.
One more honesty note, since I'm asking you to trust the rest of this page. None of these findings come from randomising anyone to a cadence. They are comparisons across trials — studies that happened to repeat feedback versus studies that didn't. A repeated-feedback trial is probably tackling a harder, more entrenched behaviour in the first place, which would drag its effect size down all by itself. That's confounding, not causation. It's a good reason to disbelieve the 2025 result as a claim that repetition is harmful — and no reason at all to believe the frequency story anyone is selling you.
And the field knows it's stuck
In 2014 the same research group audited their own literature and concluded that the effect size "became stable in 2003" — after just 51 comparisons from 30 trials — and that new trials were "contributing little further information" about what actually makes feedback work.
That's the paper in the middle row of the table above, and I want to be straight about it: it is also the paper that says feedback works best when "presented frequently." It cuts against me. I'm including it because leaving it out is precisely the move I spend the rest of this article complaining about — and because a reader who checks my sources should find them saying what I said they say, including the parts I'd rather they didn't.
Meanwhile, in the real world, benchmarking is mostly an annual event
When MGMA polled medical groups on how often they benchmark themselves against external data, the median answer was once a year:
How often medical groups compare themselves to benchmarks
And the ADA's own guidance on dental KPIs? Hedged, but clear: "it may be best to start by tracking a limited number of factors and to reassess them at least quarterly" — while noting that KPIs tied to clinical production, appointments and overhead should be tracked more frequently than the rest.
Note what that clause does not say. It says those KPIs should be tracked more often than the rest — not weekly. Weekly is my inference from it, not the ADA's recommendation, and I have no evidence that a week beats a fortnight. I'd be a hypocrite to spend eight hundred words saying nobody has tested cadence and then quietly present my own as a finding.
What I will defend is the direction, and the reasoning is ordinary rather than scientific: the operational numbers — no-shows, schedule fill, collections — move fast enough that a short loop can still catch the cause, while the structural ones — retention, payer mix, overhead — barely budge in seven days, so watching them weekly mostly trains you to react to noise. Hence: five numbers on a short loop, the wider set quarterly. Where exactly you set that loop is a judgement call. I'm making it, not proving it.
Monday, then, is a forcing function, not a finding. A fixed slot means the review happens at all, before the week eats it. That's a perfectly good reason to pick a day — and it is not a scientific claim, so I won't dress it up as one.
The finding that should actually change how you do this
Here is the number that reframed this entire article for me.
The landmark meta-analysis of feedback — Kluger and DeNisi, Psychological Bulletin, 1996 — pooled 607 effect sizes across 23,663 observations. On average, feedback improved performance (d = 0.41). But:
First, what this is and isn't about. A "feedback intervention" means someone being shown data about their own performance. That is not you reading your own collection rate on a Sunday night — it's what happens the moment you show a clinician their no-show column, or a front-desk team their booking numbers. So read the next three numbers as a warning about how you run the review with your team, not about whether you personally should look at a screen. That's the domain the evidence actually covers, and it's the part of the Monday review most likely to go wrong.
Read the first and third cards together. Showing someone their performance data is not automatically good: over a third of the time it made performance worse. And when it does help, it helps modestly. Cochrane's headline estimate is a 6.2% mean absolute improvement — that's the figure with the confidence interval and the moderate-certainty rating behind it, and it's the one I'd defend. But the distribution is skewed: the median outcome improved by 2.7%, and the interquartile range starts at zero. Both numbers are real. The mean is the better estimate; the median is the better description of a typical case. Neither is a transformation.
One boundary worth drawing, because it's where articles like this usually cheat: that 2.7% is the improvement in the specific clinical behaviour being audited — prescribing, test-ordering — not a general uplift a practice "sees." Don't let anyone, including me, turn it into a revenue promise.
Still, this is the strongest argument I know of for being deliberate about how you run a review — and against the "real-time visibility" pitch that every dashboard company, including the category I sell into, runs on.
Dashboards specifically? The evidence is thin
A 2022 systematic review in JAMIA looked at the randomised trials of clinical dashboards — 11 of them — and concluded there is "limited evidence indicating the positive impact of introducing clinical dashboards into routine practice" on medication use and test ordering. The results were conflicting, and the heterogeneity was too great to pool them at all.
Then, reaching for an explanation of why the trials came out so flat, the authors offer a line that ought to be on a poster in every health-tech office: "Another possible reason for the nonsignificant effects might be a lack of dashboard use by clinicians or patients." Note the hedge — might be. They didn't measure it. They just couldn't rule out that the dashboards were built, deployed, studied, and never opened. Which is its own kind of indictment.
What the evidence does support
The 2025 review also found several design features that did hold up. Feedback works better when it:
- Compares you to a benchmark or to top performers. The meta-regression found a benefit for comparison against top peers or a benchmark, and did not find a significant benefit for comparison against average peer performance. Be careful with that second half, though — "no significant effect found" is not the same as "doesn't work", and the review's head-to-head trials do support peer comparison over no comparison at all. The safe reading: any comparison beats none, and a target beats an average.
- Comes from someone with an existing relationship with you — a local champion, not a report from head office.
- Is interactive, rather than a written or didactic one-way document.
- Uses individual-level rather than team-level data. "The practice's no-show rate is 9%" changes nothing. "Your Tuesday afternoon column has a 19% no-show rate" changes something.
- Comes with an action plan containing specific advice. This is the one everyone skips.
Notice that every single one of those is about the design of the review, not its frequency. A weekly ritual with no owner and no action plan is not a neutral use of thirty minutes — on this evidence, it carries roughly a one-in-three chance of making things worse.
The Monday review that's actually worth doing
Open one screen, not seven reports
Five numbers, each against its own trend line. If assembling the view takes longer than reading it, you will stop doing it by week three — which is the real reason most metric routines die.
Compare to a target, not to last week alone
Comparison to a benchmark or to top performers is one of the few design features the 292-study review actually backs. Where a real benchmark exists — collection rate ≥95%, A/R 30–40 days — use it. Where it doesn't (retention, utilisation), set your own target and compare against that. A number with nothing to compare it to is decoration.
Look at one number per person, not one number per practice
Individual-level data outperformed team-level data. Split each metric by provider, by site, by day of week. A practice-wide average is the single best way to hide the one column, one clinician or one Tuesday that's causing the problem.
Give every red number an owner and a date
Not "we should look at no-shows." Instead: "Maria will call every Friday-afternoon appointment two days ahead, starting this week, and we'll look again on the 28th." An action plan with specific advice is one of the few things the 292-study review actually supports.
Change nothing when nothing moved
Most week-to-week movement in a small practice is noise. If a number wobbled 2% and nobody knows why, that's not a signal — that's a Tuesday. Chasing it burns the credibility you'll need when something real does move. Write down what you're watching, and wait. (This one is my own judgement, not a finding — I'm flagging it because the rest of this section isn't.)
Does any of this actually make you more money?
The honest answer: for a small practice, nobody has measured it. There is no study linking measurement and reporting to the financial performance of a US medical or dental practice. Every confident claim you'll read to that effect is marketing.
What does exist is suggestive, and comes from outside healthcare. In a randomised field experiment on Indian textile firms (Quarterly Journal of Economics, 2013), plants given free consulting on modern management practices — a bundle that explicitly included recording quality problems by type and analysing those records daily — raised productivity by 17% in the first year through better quality, higher efficiency and lower inventory.
That is a genuine causal estimate, and it is the best one anybody has. Now let me hold it to the same standard I've held everyone else to on this page, because it does not survive it comfortably:
- The sample is twenty plants. Fourteen treated, six control, across seventeen firms. The authors are candid about it: the small sample was, in their words, "the major challenge of our experiment."
- The treatment was not a dashboard. It was five months of on-site implementation consulting, on top of a month of diagnostics that the control plants also received — an intervention the consultancy prices at roughly $250,000 per plant.
- The famous profit figure is imputed, not measured. The authors say so.
- And it is textile plants, not clinics.
So: the most expensive management intervention ever properly tested, run on twenty factories, produced a 17% gain. You cannot attribute that to looking at a dashboard, and I'm not going to pretend otherwise. In healthcare itself the equivalent work — hospital management scores correlating with outcomes like heart-attack survival — is explicitly correlational; those authors state plainly that they cannot demonstrate causation.
So the case for a Monday review isn't "studies prove it pays for itself." It's narrower and more defensible: you cannot fix a collection rate you have never calculated, and the five numbers above are the ones that most often turn out to be quietly broken.
The numbers you'll be handed that aren't real
One last thing, because it will save you from acting on fiction. Three figures circulate constantly in this space, and none of them means what it's presented to mean:
| The claim | What's behind it |
|---|---|
| 80–89% "optimal utilisation" | It appears in a vendor-sponsored article on MGMA's site, written by an employee of a room-scheduling software firm. The survey it cites (of health-system executives) only supports the claim that clinics run ~20 points below the range — where the 80–89% range itself came from, nobody says. No sample size, no date, no method. Its single footnote points to the vendor's own whitepaper, and that link is now dead. |
| ~14% "dental no-show rate" | The only published US figure near it is 14.3% — a visit-level rate at an academic dental school clinic (7,379 visits among 825 patients aged 0–19, over half self-pay). Not a private-practice benchmark. Reported prevalence runs 5% to 38%. The best private-practice number in circulation is 7.4% (a vendor dataset across ~3,400 practices) — which at least states an n and a date, and is still a vendor's own book of business. |
| 60–70% "patient retention" | Vendor blogs. All of them. No dataset, no method, no primary source anywhere. |
Then there are the two figures you'll meet most often: "no-shows cost US healthcare $150 billion a year" and "every no-show costs you $200." I had these down as untraceable. They aren't — and the truth is more useful. Both come from a single 2017 trade-magazine article written by the chief marketing officer of an appointment-scheduling company, which asserts them with no methodology and no citation of any kind. That is their entire provenance. Every "studies estimate…" you have ever read on this leads back there.
The one cost-per-no-show figure with an actual method behind it is not a universal $200 — it is $196 per missed appointment at a US Veterans Affairs medical center (Kheirkhah et al., BMC Health Services Research, 2016). A real study, a real number, and one that tells you very little about your own practice. Which is rather the point.
Which raises an obvious objection to this whole article
If my rule is "no dataset, no benchmark" — then what about the AAFP numbers I told you to use? Because the AAFP publishes no dataset either. No sample, no method, no year. By the standard I just applied to everyone else, "collection rate should be 95%" is a slogan too.
That objection is fair, and the answer is not that AAFP is nicer. It's that there are two different kinds of number, and the whole game is telling them apart:
- A professional target — what a body like the AAFP or the ADA says you should aim for. It is a judgement, openly presented as one. AAFP files its figures under "best practice tips." Nobody is pretending they measured anything.
- A fake measurement — a number presented as an observed fact about the world ("the optimal range is 80–89%", "60–70% of patients stay") when no one ever observed it.
The first is honest and useful: aim at it, and be held to it. The second is a claim about reality with nothing behind it, and it will send you chasing a gap that may not exist. Use targets as targets. Don't accept a measurement that never happened.
Where to build it
Five numbers do not need a platform. You can build this from the exports your practice management system already produces, and there are three sensible routes: a KPI dashboard in Excel if you want it done this afternoon; Power BI if you want it to refresh itself; or a template if you'd rather not build anything.
All five numbers, one screen, ten minutes
Clinic Vitals is a Power BI template built around exactly this review: collections, A/R, no-shows, utilisation and retention — each against its own trend, split by provider and site, from the exports you already have.
View Clinic Vitals →Frequently asked questions
What should be on a clinic dashboard?
Five numbers are enough for a weekly review: net (adjusted) collection rate — how much of what you earned you actually collected; days in A/R — how long the money takes to arrive; no-show rate — the capacity you paid for and didn't use; provider utilisation — how full the schedule was; and patient retention — whether people come back. Revenue itself is deliberately not on the list: it's the scoreboard, not a lever. The five above are what you can act on this week to move revenue next quarter.
How often should I review my clinic's metrics?
Honestly: nobody knows, and anyone who tells you otherwise is guessing. No study has randomised practices to weekly versus monthly review of business metrics. The closest evidence — Cochrane's reviews of audit and feedback — cannot make up its mind: the 2012 review (140 studies) found feedback may work better when "provided more than once", and a 2014 paper by the same group said it works best when "presented frequently", but the 2025 update (292 studies) found, "contrary to expectations", that repeated delivery was associated with a lower effect size. All of these are observational comparisons across trials, not head-to-head tests of cadence — and, tellingly, the 2025 review left frequency out of its conclusions entirely. The ADA advises reassessing KPIs at least quarterly. Weekly is a useful forcing function for the operational numbers — not an evidence-based prescription.
Can looking at a dashboard make performance worse?
Showing someone else their numbers certainly can. Kluger and DeNisi's meta-analysis (1996), covering 607 effect sizes and 23,663 observations, found feedback improved performance on average (d = 0.41) but that over one-third of feedback interventions decreased performance. That literature is about data given to a person about their own performance — so it applies the moment you hand a clinician their no-show column, rather than to you reading your own P&L. Separately, a 2022 JAMIA review of 11 randomised trials of clinical dashboards found only limited evidence of positive impact, and speculated that one reason the results were so flat "might be a lack of dashboard use by clinicians or patients". A number with no owner and no action plan isn't neutral; it's a distraction with a cost.
What is a good benchmark for these numbers?
Separate the targets from the measurements. For two of the five there are professional targets: the AAFP says the adjusted collection rate should be at least 95% (average 95–99%, top performers 99%+), and days in A/R should stay below 50, with 30–40 preferable. Those are best-practice judgements, openly presented as such — AAFP publishes no dataset behind them, and that's fine, because a target isn't claiming to be a measurement. For the others, be sceptical. The widely quoted "80–89% optimal room utilisation" appears in vendor-sponsored content with no sample size, date or methodology — and no stated origin for the range itself — while there is no credible patient-retention benchmark at all, since every figure in circulation traces to vendor blogs. Compare yourself to your own trend line first.
What makes a metric review actually work?
The 2025 Cochrane review of 292 studies found audit and feedback is more effective when it compares performance to top peers or a benchmark; when it's delivered by a local champion who already has a relationship with the recipient; when it's interactive rather than just written or didactic; when it uses individual- rather than team-level data; and when it comes with an action plan containing specific advice. Every one of those is about the design of the review. Frequency appears nowhere in the review's conclusions, in either direction.
Every figure here was checked at its primary source, and the draft was then attacked by a second fact-check pass — which changed it. Where a number turned out to have no method behind it I have said so in the sentence rather than the footnote, including where that cost me the stronger claim. The audit-and-feedback frequency findings are meta-regressions across trials, not randomised tests of cadence, and the 2025 review's full text sits behind a paywall. Lucid Vitals is not affiliated with MGMA, AAFP, the ADA or Microsoft.
Sources
- Ivers et al., Cochrane Database of Systematic Reviews (2025) — Audit and feedback: effects on professional practice. 292 studies / 678 arms overall; the effect estimates come from 177 of them (558 dichotomous outcomes): mean absolute improvement +6.2% (95% CI 4.1–8.2, moderate certainty), median +2.7% (IQR 0.0–8.6). The frequency result — "contrary to expectations, repeated delivery was associated with lower effect size" — is an exploratory meta-regression and does not appear in the authors' conclusions · plain-language summary
- Ivers et al., Cochrane Database of Systematic Reviews (2012) — The earlier review (140 studies): feedback may be more effective when "provided more than once"
- Kluger & DeNisi, Psychological Bulletin (1996), 119, 254–284 — The effects of feedback interventions on performance (607 effect sizes; over one-third of interventions decreased performance)
- Ivers et al., Journal of General Internal Medicine (2014) — "Growing literature, stagnant science?" — the effect size stabilised in 2003 and new trials add little
- Xie, Chen, Hincapié et al., JAMIA (2022) — Systematic review of 11 RCTs: "limited evidence" for clinical dashboards. The authors speculate the flat results "might be" due to "a lack of dashboard use by clinicians or patients" — a hypothesis, not an observation
- AAFP — Finances and your practice: adjusted collection rate 95% minimum / 99%+ for top performers; days in A/R below 50, 30–40 preferable; denial rate 5–10% average (AAFP guidance, not a published dataset)
- MGMA Stat (Jan 2025) — No-show rate 6.81% in 2023 (MGMA DataDive single-specialty aggregate), vs a 7% pre-pandemic 2019 benchmark
- MGMA Stat (Dec 2023) — Benchmarking cadence poll: 41% annually, 24% at least monthly, 15% quarterly, 15% never, 4% other (n = 332). "Weekly" was not an answer option
- American Dental Association — Key performance indicators: "it may be best to start by tracking a limited number of factors and to reassess them at least quarterly"; recommended target of 5% or less for cancellations and no-shows combined (recommended targets; no dataset published)
- MGMA (Nov 2020) — Where the "80–89% optimal room utilisation" range circulates: a sponsored article authored by a room-scheduling software vendor. The Porter Research survey of health-system executives it cites supports only the "~20 points below" claim; the range itself is asserted, with no sample size, date or method, and its lone footnote points to the vendor's own (now dead) whitepaper
- International Journal of Dentistry (2025) — The source of the "14% dental no-show rate": a visit-level rate at an academic dental school clinic (University at Buffalo; 7,379 visits, 825 patients aged 0–19, 52% of visits self-pay); reported national prevalence ranges from 5% to 38%
- Bloom, Eifert, Mahajan, McKenzie & Roberts, Quarterly Journal of Economics (2013), 128(1):1–51 — Does management matter? Evidence from India (17% first-year productivity gain; note the earlier NBER working paper reports 11%)
- Bloom, Lemos, Sadun & Van Reenen, Review of Economics and Statistics (2020), 102(3):506 — Healthy business? Management practices and hospital outcomes (correlational)
- Brehaut et al., Annals of Internal Medicine (2016), 164:435–441 — 15 suggestions for optimising audit and feedback (expert consensus, not trial evidence; the authors concede "there is still much to be learned about optimum methods")
- Lam, DeRue, Karam & Hollenbeck, Organizational Behavior and Human Decision Processes (November 2011) — The impact of feedback frequency on learning and task performance: the one experiment that actually manipulated how often feedback was given, finding an inverted-U ("challenging the 'more is better' assumption"). A laboratory task, not a clinic
- Kheirkhah, Feng, Travis, Tavakoli-Tabasi & Sharafkhaneh, BMC Health Services Research (2016), 16:13 — Prevalence, predictors and economic consequences of no-shows: $196 per missed appointment — at a US Veterans Affairs medical center, which is the only cost-per-no-show figure here with a published method
- Gier, Health Management Technology (April 2017) — "Missed appointments cost the U.S. healthcare system $150B each year" — the single origin of both the "$150 billion" and the "$200 per no-show" figures, written by the chief marketing officer of a scheduling vendor, with no methodology and no citations