Dynamic incentives lifted web bookings 69% and added $39,637 of incremental booked revenue across five ProMD locations in one month.
A 5.53x return on incentive spend, measured against a randomized holdout rather than a before-and-after baseline, with no cannibalization detected.
If you incentivize customers to book slots that would otherwise go unsold, do you actually increase net bookings, or do you just hand a discount to people who would have booked anyway?
What Booko does
Booko grows revenue for appointment businesses by making sure fewer slots go unsold. Its demand-forecasting models train on a location's historical and live booking data plus real-time signals (seasonality, weather, local events, geography) to predict, slot by slot, which appointments are likely to go empty, and its pricing models size the right incentive for exactly those slots.
The same models drive the marketing layer, choosing which lapsed patients to reach and which promoted slots to feature. Full-price demand is left alone.
The challenge
An appointment slot that goes unfilled is revenue that never existed. In medical aesthetics the average ticket is $527 and the average practice sees about 245 visits a month (AmSpa, 2024), so even a handful of empty slots per provider per week adds up fast. ProMD Health operates medical aesthetics clinics across Maryland, Virginia, and Florida, and like every practice it has predictable soft spots: certain days, certain hours, certain providers with open time.
Blanket promotions are the usual answer, and they have a known failure mode: they discount patients who were going to book anyway. ProMD wanted the opposite, a way to fill the slots that needed help without touching demand that was already there. The group has built its reputation on adopting new technology ahead of its industry, and it approached AI the same way: start with a rigorous experiment, then scale what proves out. So ProMD and Booko designed a joint research pilot to prove it properly.
What Booko did
At ProMD, Booko's models learned the group's demand patterns from its booking history, then ran live: forecasting demand for every upcoming appointment slot, predicting the ones most likely to go unsold, and pricing a right-sized incentive for exactly those slots. Incentives surfaced as loyalty rewards inside the web booking flow patients already use. High-demand slots got no incentive. Nothing changed for ProMD's front desk, and no new system was added to the stack.
How we measured it
Of the slots the engine flags as worth incentivizing, a randomized share is deliberately held out: the widget stays off and patients book at full price. Treatment and holdout are drawn from the exact same pool of incentive-eligible appointments. The only difference is whether the patient saw the reward. That is what makes the read causal: the holdout arm tells us what would have happened without the widget, stripping out seasonality, demand shifts, and operational noise that a year-over-year or pre-pilot baseline would fold in.
Booking counts for whichever arm the slot was in at the moment the patient booked. Phone and walk-in bookings are excluded; the widget cannot influence them. Method details are in the fine print at the bottom of this page.
Results
The counterfactual is what treatment slots would have earned if they had only booked at the holdout's pace, with no widget. The gap between that and what treatment actually earned is the revenue the program created.
Of the $7,172 incentive cost, $5,285 (74%) was already credited to patient accounts at report time; the remainder is owed only if patients attend.
Per location (Conservative window)
| Location | Treatment rate | Holdout rate | Relative lift | Uplift | Incentive cost | Return |
|---|---|---|---|---|---|---|
| ProMD Annapolis | 7.91% | 5.06% | +56% | +$8,229 | $1,854 | 4.44x |
| ProMD Baltimore | 6.77% | 4.09% | +66% | +$15,453 | $2,021 | 7.65x |
| ProMD VA | 7.31% | 2.37% | +208%* | +$9,116 | $1,290 | 7.07x |
| ProMD Wellington† | 2.24% | 1.41% | +59% | +$496 | $161 | 3.09x |
| TOX BAR | 9.35% | 7.25% | +29% | +$6,342 | $1,846 | 3.44x |
| Pooled (all locations) | 7.11% | 4.21% | +69% | +$39,637 | $7,172 | 5.53x |
*ProMD VA's relative lift comes from a small holdout denominator (3 holdout bookings); treat it as directional rather than precise. †ProMD Wellington is a location still ramping volume; its read rests on a single holdout booking and is likewise directional. Columns may not sum exactly due to rounding.
Did it cannibalize existing demand?
The skeptical question deserves a direct audit, so we ran negative-control checks across ProMD's full booking environment. If Booko were stealing bookings from nearby slots, those slots should have declined. They did not: nearby unrewarded slots rose by roughly 430 bookings versus expected, and no control cohort showed an offsetting drop.
| Check | Evidence | Result |
|---|---|---|
| Nearby unrewarded slots | +427.5 bookings vs expected within +/- 1 hour; +585.1 within +/- 2 hours | Rose, not fell |
| Ineligible and non-visible inventory | 6 control cohorts: 4 supported incrementality, 0 showed displacement, 2 inconclusive; combined +620.75 vs expected | No offset detected |
| Staff and non-web channels | Web bookings rose by 1,224 vs the pre-period rate while staff and non-web bookings did not decline | No channel shift |
Fractional counts come from time-weighted exposure. The booking-side audit is the strongest cannibalization read; revenue-side evidence is supportive but directional. The environment-wide rise in web bookings affects both arms equally, which is what the randomized holdout nets out.
The marketing layer: reaching lapsed patients
The same engine now also drives email outreach to lapsed patients through ProMD's existing email marketing platform, pointing them at the slots that need filling. The first cycle sent 5 campaigns (one per location) to 874 lapsed patients, with a blended open rate of 44.9%, roughly 1.4x Constant Contact's Health & Wellness industry average of ~33%, and 0 unsubscribes. The Annapolis send reached 276 patients lapsed 9+ months and opened at 52.7%.
One lapsed patient booked the exact promoted slot six hours after the email, noting they came in because the email told them it was time for a filler refill. The filler campaign also drove a cross-service halo: 1 Sculptra, 2 Dysport, and 2 Botox bookings from patients who opened the email.
One cycle proves the pipeline end-to-end: audience selection, send, click, and booking. A publishable conversion rate needs 6 to 8 more cycles, and we do not claim one yet. The layer complements ProMD's existing retention work; it does not replace it.
“We have it set up as a last-hour push. It's not competing with our organic bookings, and it's working really well.”
The research pilot answered the question it was designed to answer. It has since grown into an ongoing enterprise partnership now expanding across the ProMD network, with ProMD continuing to lead its industry in putting applied AI to work.