The quarterly numbers are on the screen, and the owner says the line that ends most of these conversations early: “We’re better than the benchmark.” It is true, and it answers a question almost nobody is asking.
A benchmark tells you where your practice sits next to other practices. That is useful once a year, in a board meeting, when somebody wants context. It tells you nothing about whether something in your own operation changed last week. Those are different questions, and only one of them is urgent.
Here is where this sits on the path from a visit to money in the bank. A visit becomes a charge, the visit written up as a billable line, and the charge becomes a claim, the bill sent to the insurance company. Then the insurance company answers.
Every step of that path throws off a number: how long a note waits for a signature, how long an insurer takes to decide, how much of a patient’s share was collected at the desk. Drift detection is the habit of watching those numbers against their own past. It is one of the five checks that watch the money path instead of waiting for a report.
What is drift detection in a medical practice?
Watching a number against its own recent history instead of an outside standard, so a change shows up as a signal early. The comparison is you against you.
At one practice, the days from visit to payment slid by about half in one quarter. Against a market benchmark of thirty it still looked excellent. Against its own past it had lost half its advantage, and the cause had a date.
Why a benchmark hides movement
Take that nine-day practice drifting to fourteen. Against the benchmark nothing would raise a question. Against its own history something specific caused that, and it is still fixable. The same logic runs the other way. A practice sitting at forty days in a market that averages thirty looks poor on every report and may be completely stable.
Nothing is changing there, so nothing needs urgent attention, even though the number is uncomfortable. Benchmarks describe position. Drift describes change. Position is a strategy conversation, and change is a this-week conversation.
Why change is the more useful signal
Because change has a cause, and the cause is usually recent, specific, and still fixable. An insurance company that starts taking longer to decide on claims did something. A note that now waits twice as long for a signature has a reason. A denial, the insurance company’s refusal to pay with a reason code, showing up at four times its usual rate traces to a rule somebody changed.
Each of those is findable while the trail is warm. Three months later the same search is archaeology, and the pile of affected claims has multiplied.
What to watch it on
Not everything. Four or five numbers, chosen because a change in them means something specific happened.
Days from the visit to the payment. The whole distance from care to cash. It moves before anything else, and it moves for reasons that are always findable.
How long each insurance company takes to decide. Per insurer, never blended. One insurer slowing from twelve days to thirty barely moves a blended average while it moves your bank balance, which is why the average is the wrong instrument.
Charge lag. The days between the visit and the charge being created. A rise means something changed in the notes or in how visits get turned into charges, and it shows up in cash three weeks later. Charge lag is the first of the twelve places a claim can get stuck.
Denial rate by reason. The total is noise. One reason code appearing at four times its usual rate is an insurance company changing a rule, and it is worth knowing this week.
Patient payment at the desk. What patients paid at the time of the visit. It falls quietly when the front desk routine slips, and it predicts the size of your unpaid patient balances better than any later number.
Setting a rule you can actually run
Three decisions, none of them sophisticated. First, what is normal: a rolling average over enough weeks to smooth ordinary variation, long enough to be stable and short enough to still be current. Second, how far is far enough: a percentage change or a standard deviation both work. What matters is picking one and writing it down, because an undefined threshold means the alert fires whenever somebody feels concerned.
Third, how long before it counts. One day outside the range is noise. Three days running, or two weeks running, is a signal. That last decision is the one that kills most attempts, because an alert that fires constantly gets ignored, and an ignored alert is the same as no alert at all.
Real situations, and what the monthly number said
When one insurance company slows from about twelve days to about thirty, the blended average barely moves. The blended average moved by less than two days, so nothing on the monthly pack changed color. Per insurer, that one company was sitting on a month of cash.
One denial reason jumping to four times its usual rate is the classic sign of an insurer applying a new rule. The total denial rate barely moved, because the other reasons were flat. Grouped by reason, it was one insurance company applying a new rule, and it had a fix once somebody knew to look.
Where this comes from
Manufacturing worked this out decades ago. A production line does not judge each part against an industry standard. It plots measurements against the line’s own recent behavior and reacts when the pattern changes. A process that starts producing slightly different acceptable parts is telling you something before it produces rejects.
The idea that transfers is this. A value inside acceptable limits can still be a problem if it got there by moving. The movement is the information, and waiting for the number to become unacceptable means waiting until the cost has already been paid. A practice at fourteen days is inside every acceptable limit. It was at nine.
Why a monthly review does not do this
A monthly review looks at a value. Drift detection looks at a direction, and the two find different things. A number moving steadily one way reads as normal in any single month, because the change from one month to the next is small.
It only looks alarming across five or six months, by which point five or six months of claims have been affected. This is why practices are surprised by a number that has been moving for a quarter. Nobody was hiding it. Every individual month looked fine.
What this means for you
Take one number. Days from the visit to the payment is the best place to start, because everything else feeds it. Plot it weekly for the last twelve months instead of reading it as a monthly figure. If the line is flat, you have stability, and that is worth knowing.
If it has been climbing since March, you have found something, and the cause will have a date. Then add the per-insurer version, because that is where a single quiet insurer hides inside a healthy average.
Grab 30 minutes with us. Prep nothing. You will see whether your numbers have been moving and since when.
Questions people ask
What is drift detection?
Watching a number against its own recent history instead of an outside benchmark, so a change surfaces as a signal instead of becoming a bad quarter. The comparison is you against you. It catches the practice that is still better than average and quietly losing ground, which a benchmark never will.
Why are benchmarks not enough?
Because they describe position, and drift is about change. A practice turning care into cash in nine days that drifts to fourteen still looks excellent against a market average of thirty. It has lost more than half its advantage for a specific and recent reason, and nobody has looked for it.
Which numbers should a practice watch for drift?
Days from the visit to the payment, how long each insurance company takes to decide, the lag between a visit and its charge, the denial rate by reason, and patient payment at the desk. Five numbers where a change means something specific happened and is still fixable.
How do I set a drift threshold?
Define normal with a rolling average, decide how far outside it counts, and require the condition to hold for three days running before it fires. Without that last part you get constant alerts, and an ignored alert is the same as no alert. Write the threshold down, so it does not fire on a feeling.
Why does a monthly review miss drift?
Because it looks at a value, and drift is a direction. A number moving steadily reads as normal in any single month, since the change from one month to the next is small. It only looks alarming across five or six months, by which point the cost has been paid.