Lesson objective
Measure volume, cycle time, rework and error for a practice workflow before it is redesigned, and record the method precisely enough that the same measurement can be repeated afterwards.
Plain-language explanation
Why this matters now
A practice that changes a workflow without a baseline can never answer the only question that matters afterwards: did this help. It will have opinions. It will have a vendor dashboard. It will have a member of staff who says the mornings feel calmer and another who says the afternoons are worse. It will not have a comparison, and it will not be able to detect the thing it should fear most, which is a change that improves speed while quietly degrading accuracy.
Four measures cover most practice workflows, and all four are needed because they trade against each other.
| Measure | What it is | Why alone it misleads |
|---|---|---|
| Volume | Items entering the process per period | Says nothing about how well they are handled |
| Cycle time | Elapsed time from start to finish, and the waiting inside it | Can fall while rework rises |
| Rework | Items that go back a step, are re-sent, reopened or resubmitted | Invisible unless someone counts it deliberately |
| Error | Items that were wrong and were found to be wrong | Depends entirely on how hard anyone looked |
Professional guidance (voluntary, nonbinding): OIG's Compliance Program Guidance for Individual and Small Group Physician Practices, published at 65 Fed. Reg. 59434 on October 5, 2000 and now listed in OIG's archive, puts internal monitoring and auditing first among its seven components, and unusually says it "does not suggest that physician practices implement all seven components of a full scale compliance program." It is voluntary, it is dated, and it says nothing about AI. What it does supply is the habit: a practice that already audits its own billing has the instinct this lesson needs, applied to operations instead.
A question to sit with
Your practice changed something last year that everyone agrees was an improvement. What number, measured the same way before and after, would convince a sceptic, and does that number exist?
What goes wrong
The baseline is measured after the change. Somebody remembers to ask in month two and reconstructs "before" from memory. Memory reliably reports the worst day as typical.
Only the flattering measure is taken. Cycle time is easy to pull from a system and looks good. Rework needs somebody to count, and error needs somebody to check, so neither gets measured and the story is told with the one number that moved.
The method is not recorded, so the follow-up measures something else. The baseline counted business days and the follow-up counted calendar days; the baseline counted all referrals and the follow-up counted only new patients. The comparison is now worthless and looks fine.
Error is confused with what was caught. A practice reports two errors last quarter. It found two. Nobody knows the denominator, because nobody sampled.
The window is too short or the wrong shape. A two-week baseline that misses month-end, a holiday week or the annual plan-year change describes a season, not a process.
The measurement changes the work. Staff who know they are timed work differently. This cannot be eliminated and should be recorded, not ignored.
The practical answer
Write the measurement plan before collecting anything. Six lines are enough, and this is this programme's recommended minimum.
| Line | Example content |
|---|---|
| What is counted | Prior authorization requests initiated by this practice |
| Inclusions and exclusions | All payers; excludes requests cancelled by the patient |
| Clock start and stop | Starts when the order reaches the authorization queue; stops at the payer decision |
| Unit of time | Business days, practice holidays excluded |
| Who collects, how | Named person, from the queue export plus a manual rework tally |
| Window, and why | Six weeks, chosen to include one month-end |
Measure rework and error deliberately, because neither shows up by itself. Rework is a tally staff keep at the moment it happens: one mark each time an item goes back a step. Error needs a sample checked against a source. Take a fixed number of completed items per week, have a qualified reviewer check them against the record they should reflect, and record both the number checked and the number wrong. Module 13 builds this into a full sampling plan; the baseline needs only enough to give an honest denominator.
Record the conditions, not just the numbers. A baseline taken during a staff shortage, a system migration or a plan-year change is still useful if the conditions are written beside it. The same baseline with the conditions omitted becomes a misleading benchmark the moment somebody quotes it in a memo six months later.
Decide in advance what would count as worse. This is the discipline that separates measurement from marketing. Before the redesign, write down the result that would make the practice stop: a rework rate that rises, an error rate in the sample that rises even while cycle time falls, or an increase in items that reach a patient, a chart or a claim without a named person having reviewed them. Module 14 develops this into stop conditions; putting the sentence in the baseline memo is what makes it possible later.
Do not import somebody else's numbers as your starting point. A figure from another practice, a vendor case study or a published study describes that setting, with those staff, those payers and that product. It cannot be your baseline. Lesson 4 takes the published documentation studies apart to show exactly what they can and cannot be used for.
Remember this
- Measure volume, cycle time, rework and error together, because the first two can improve while the last two worsen.
- Write the six-line measurement plan before collecting anything, and keep it for the follow-up.
- Rework needs a deliberate tally and error needs a sample checked against a source; neither appears on its own.
- Record the conditions, the window and anything unusual, so the baseline cannot become a benchmark later.
- Write down in advance what result would count as worse, and include the review rate among the measures.
Medical practice example
Brightfield Pediatrics, a fictional three-clinician practice, plans to change how it handles school and camp forms, a process the staff describe as a nuisance that consumes a morning a week. The office manager, Tomas Reyes, resists the temptation to fix it immediately and spends six weeks measuring instead. He writes the plan first: what is counted is completed form requests; inclusions are all forms requiring a clinician signature and exclusions are immunisation record printouts, which need no signature; the clock starts when the request is logged at the front desk and stops when the completed form is released to the family; time is counted in business days; he collects the volume and dates from the request log and keeps a paper tally for rework; the window is six weeks, chosen because it spans the start of the school year. The results surprise him twice. The volume is 214 requests, far higher than the "morning a week" story implied. The median cycle time is 4 business days, which nobody complains about. But the rework tally records 71 items going back a step, which is 71 divided by 214, or 33.2%, and almost all of it is one cause: the form is completed and then returned because a required field was on a second page the front desk had not scanned. The error check, 10 completed forms a week for six weeks with a clinician reading each against the chart, finds 4 forms of 60 with a date or a measurement that did not match the record, which is 6.7%. Tomas writes all of it down, including that week three covered a public holiday and week five ran with one coordinator absent. The redesign that follows is not an AI project: it is scanning both sides of every form at intake. He keeps the baseline, because the next proposal to arrive will be an AI one, and he now has numbers to hold it to.
Responsible-use warning
Do not report an error count without a denominator. "We found three errors last quarter" says only that three were found, and the number depends entirely on how hard anyone looked; a practice that checks nothing will honestly report zero. An error figure means something only as a fraction of a stated sample, checked against a stated source, by a stated reviewer. Without those three elements the number cannot be compared with anything, including the same practice's own figure from the quarter before.
Human-review requirement
The practice manager or named administrative owner owns the measurement plan and the baseline record, and signs it with a date and the conditions of the window. Where the error sample requires reading clinical documentation, the reviewer must be qualified to judge it: the physician or advanced practice lead for documentation content, and the coding professional together with the clinician whose documentation supports the claim for anything touching codes. The practice owner approves the sentence stating what result would count as worse, because that sentence is what obliges the practice to act on its own measurement later.
Practical exercise
Use fictional or fully anonymised material only. No real patient information.
- Choose the workflow you mapped in Lessons 1 and 2. Write the six-line measurement plan before you collect anything: what is counted, inclusions and exclusions, clock start and stop, unit of time, who collects and how, and the window with the reason for its length.
- Choose a window of at least four weeks that includes whatever is seasonal in your practice: a month-end, a plan-year change, a school-year start or a known busy period. Write down why you chose it.
- Collect volume and cycle time from whatever system already records them. Report cycle time as a median and as the slowest tenth, not as an average alone.
- Keep a deliberate rework tally: one mark each time an item goes back a step, is re-sent, reopened or resubmitted. Express the total as a percentage of volume.
- Design the error sample: a fixed number of completed items per week, checked by a named reviewer qualified to judge them, against a stated source document. Record both the number checked and the number found wrong.
- Record the review rate as well: the proportion of completed items where a named, qualified person reviewed the output and the review was recorded.
- Write down every condition that affected the window: absences, system changes, holidays, unusual volume. Put them in the same document as the numbers, not in a separate note.
- Before proposing any redesign, write one sentence stating the result that would make you stop or reverse it, naming the measure and the direction. Have the practice owner sign that sentence.
Knowledge check
-
Why must volume, cycle time, rework and error be measured together?
Answer guidance: They trade against each other. Cycle time can fall while rework rises, and both can look good while error rises, because a faster process may simply be checking less. Volume alone says nothing about quality, rework is invisible unless someone tallies it deliberately, and an error figure depends entirely on how hard anyone looked, so only the four together describe what happened.
-
What belongs in the six-line measurement plan, and why must it be written before collecting anything?
Answer guidance: What is counted, the inclusions and exclusions, the clock start and stop, the unit of time, who collects it and how, and the window with the reason for its length. Written first, it lets the follow-up measurement repeat the baseline exactly; reconstructed afterwards it usually differs in some quiet way, such as business days becoming calendar days, which makes the comparison worthless while still looking sound.
-
What three elements make an error figure meaningful?
Answer guidance: A stated sample giving the denominator, a stated source document the items were checked against, and a named reviewer qualified to judge them. Without all three, an error count reports only how many were found, which a practice that checks nothing can honestly report as zero, and it cannot be compared with any other figure including its own from the previous quarter.
-
Why should a practice write down in advance what result would count as worse?
Answer guidance: Because after a redesign everyone involved has an interest in it having worked, and a threshold set afterwards will be set where the result already sits. Deciding beforehand that a rising rework rate, a rising sampled error rate even alongside a faster cycle, or more outputs reaching a patient, chart or claim without a recorded review would trigger a stop is what makes the measurement capable of showing failure.
Sources and further reading
- U.S. Department of Health and Human Services, Office of Inspector General (HHS-OIG) — OIG Compliance Program Guidance for Individual and Small Group Physician Practices (65 FR 59434, October 5, 2000) (2000-10-05). Accessed 2026-09-15.
- National Institute of Standards and Technology — Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (2023-01-26). Accessed 2026-09-15.
- Thomson Reuters Institute — 2026 AI in Professional Services Report (2026-02-09). Accessed 2026-09-15.
- AI Coalition Network — Certification Educational Standard: Human Review of AI-Assisted Professional Work. Accessed 2026-09-15.