Key points
- Write each measure's baseline, denominator, window and threshold down on day 0, and the result on day 90 decides itself.
- In a randomised trial of 238 physicians, an ambient scribe cut time in notes by 9.5% against usual care.2
- On the front desk, text reminders cut non-attendance from 21% to 15% across 21 randomised studies.4
- Measure the front desk and the consult together, since the gains in one show up as revenue or capacity in the other.
- wai runs a 90-day pilot for Admin that ends with what changed in revenue, hours saved and patients cared for.1
Summary
Run a paid clinic AI pilot for about 90 days against baselines taken from the eight weeks before it starts. Fix each measure's denominator, window and success threshold on day 0, measure the front desk and the consult together, and apply decision rules written in advance so the result decides itself.
On day 90 the clinic knows what changed in replies, attendance, notes and revenue, measured on numbers both sides agreed before the first patient message was drafted. This is the clinic-wide version of our AI scribe pilot plan, covering the front desk and the consult together.
See it in your clinicThirty minutes on your own patients and doctors.
Request a demoSee the pilot scorecard filled in with your own baselines, and the measures wai reports at day 90, in fifteen minutes.
| Measure | Denominator | Window | Example decision rule |
|---|---|---|---|
| Median time to first reply | Per patient enquiry, all channels | Days 15 to 90, weekly | Widen if halved against baseline |
| Enquiries that become bookings | Per new enquiry | Days 15 to 90 | Widen if up by an agreed number of points |
| No-show rate | Per booked appointment | Days 30 to 90 | Widen if below baseline in both later windows |
| Deposits collected at booking | Per booking where a deposit applies | Days 15 to 90 | Adjust if below the agreed share |
| Drafts approved unchanged | Per drafted message | Days 15 to 90, weekly | Adjust voice settings if flat after day 30 |
| Notes signed on the day | Per consult | Days 15 to 90 | Widen if up against baseline |
| Documentation time after the last patient | Per clinician session | Days 15 to 90 | Widen if down against baseline |
| Substantive clinical edits | Per signed note | Days 1 to 90, sampled weekly | Stop if any unsafe draft reaches signature unedited |
| Revenue | Per clinician session | Days 61 to 90 against the same weeks last year | Widen if up after removing price changes |
| Patients seen | Per clinician week | Days 61 to 90 | Widen if capacity rises with no fall in note quality |
Decide the decision first
Write the rule for each outcome before day 0, whether to widen to the whole clinic, adjust and run another 30 days, or stop.
Pay for the pilot, so both sides treat the measures seriously and the vendor sets the clinic up properly in the first week. wai runs a 90-day pilot for Admin, shaped around the team, which ends with what changed in revenue, hours saved and patients cared for.1
Baselines and denominators
Reply time is per enquiry, no-shows are per booked appointment, documentation time is per consult and revenue is per clinician session. Pull eight weeks of each from the systems the clinic already runs, and use the same definitions throughout.
The AMA's 2024 organisational data, nearly 18,000 responses, put US physician administrative work at 7.3 hours within a 57.8-hour week.5 That is the pool of time the pilot aims to return, so measure the clinic's own share of it before starting.
Log anything that will move the numbers on its own, such as a new clinician, a holiday or a price change, so it is not mistaken for an effect.
Realistic effect sizes
In a randomised trial of 238 outpatient physicians published in NEJM AI in 2025, one ambient scribe reduced time in notes by 9.5% against usual care.2 A stepped-wedge randomised trial of 66 practitioners found time on notes fell by about 0.36 hours a day.3
These are category evidence from scribe trials, never a forecast for any clinic. On the front desk, a BMJ Open meta-analysis of 21 randomised studies found text reminders cut non-attendance from 21% to 15%.4
Set thresholds the clinic would be pleased to see and that the data could plausibly show in 90 days.
Measurement windows
Split the 90 days into three windows. Days 1 to 14 are setup and learning, reported but not judged. Days 15 to 60 are the main window, measured weekly. Days 61 to 90 confirm whether the effect holds once the novelty has gone.
With more than one site or several clinicians, stagger the start. One group starts on day 0 and the other on day 30, and for that first month the second group is the comparison, so seasonal swings or a bank holiday affect both alike.
Use the event log as the source of truth for the front desk. Every message, draft, approval and signature is logged and can be exported, so reply times and approval rates come from the record.6
The pilot scorecard
The table lists each measure with its denominator, window and an example rule, and the clinic sets the thresholds.
Where to start
Choose five measures from the scorecard, two from the front desk, two from the consult and one for revenue. Pull eight weeks of baseline for each, set the thresholds with the clinical lead and the operations lead, and sign them with the vendor before day 0.
Run the pilot to our 100-day implementation plan, review weekly from day 15, and hold the decision meeting on day 90 with the scorecard alone on the table.
Bring eight weeks of your own numbers and fill in the scorecard with us in the demo.
Frequently asked questions
How long should a clinic AI pilot run?
About 90 days. The first two weeks are setup, the main window runs to day 60, and the last month shows whether the effect holds. wai runs a 90-day pilot for Admin.1
Should we pay for a pilot?
Yes. A paid pilot gets the vendor's full setup effort and makes both sides treat the measures seriously.
What is a realistic saving on documentation time?
Randomised trials of ambient scribes report modest gains, such as 9.5% less time in notes in a trial of 238 physicians.2 Measure the clinic's own baseline and judge against that.
Why measure the front desk as well as the consult?
The front desk often moves faster. Text reminders alone cut non-attendance from 21% to 15% across 21 randomised studies.4
Can we use the vendor's example numbers as our target?
Set thresholds from the clinic's own baseline. Another clinic's figures, or a vendor's illustration, say nothing about the clinic running the pilot.
How is the scribe-only pilot different?
It measures the consult alone. Our AI scribe pilot plan covers that case, and this scorecard adds the front desk and revenue.
Sources
- 1wai, 2026PricingRead 14 October 2026Read source
- 2NEJM AI, Lukac and colleagues, 2025Ambient AI Scribes in Clinical Practice, A Randomized TrialRead 14 October 2026Read source
- 3NEJM AI, Afshar and colleagues, 2025A Pragmatic Randomized Controlled Trial of Ambient Artificial Intelligence to Improve Health Practitioner Well-BeingRead 14 October 2026Read source
- 4BMJ Open, Robotham and colleagues, 2016Using digital notifications to improve attendance in clinic, systematic review and meta-analysisRead 14 October 2026Read source
- 5American Medical Association, 2025Doctors work fewer hours, but the EHR still follows them homeRead 14 October 2026Read source
- 6wai, 2026Security at waiRead 14 October 2026Read source
A clear, measured answer on day 90.
Fifteen minutes on your own numbers, with a clinician who uses wai daily. We reply within one working day.
Partner with us
See it on your own patients, then build it with us
Thirty minutes on your own patients, doctors and rules, in your market. We reply within one working day, and the clinics that partner with us early shape what we build next.
See what wai does between the visits