Hypothesis
If people booking online are asked for a deposit, stated before payment with how it is refunded or taken off the bill, attended appointments per visitor who starts a booking will rise, because fewer booked slots are missed, unless the deposit loses more bookings than it saves from no-shows. The decision behind it is concrete: take a deposit when people book an appointment online, or take no payment until the visit.
The guide to booking and appointment flows calls a deposit a business decision that balances the cost of missed appointments against the effort it adds to booking, and says it works best when people see it early and understand why it exists. It also makes the attended appointment the unit that matters, not the click or the form. Those two points define this test. The deposit has two effects pulling in opposite directions: some people will not book once payment is asked for, and some of those who do book will keep a slot they would have missed. Only a randomised comparison, counted at the appointment, shows which effect is larger for this business.
Write the deposit wording, the services in scope, the reading date and the limit on each guardrail into the experiment hypothesis template before launch. The experimentation loop puts this step first because the reading date in this test sits well after the last booking, and a team that has not fixed it in advance will be tempted to read early.
Metrics and guardrails
The primary metric is attended appointments per visitor who starts the booking flow. Count every visitor who reaches the first step of the flow in each arm, follow each booking they make to its appointment, and divide the appointments that actually took place by the visitors who started. The denominator is set at the start of the flow, so bookings lost at the payment step count against the deposit arm, and the numerator is counted at the appointment, so missed slots count against the no-deposit arm. A reading at the confirmation page would capture the first effect and miss the second.
Three guardrails explain the result and catch harm:
- No-show rate per booking. The No-Show RateThe share of booked appointments, demos, consultations or event registrations where the person does not attend. is the mechanism the deposit is meant to move. Read it by arm to see whether it moved, but do not let it decide the test, because a deposit that deters bookings can lower it without filling more slots.
- Booking completion rate. The share of visitors who start the flow and reach a confirmed booking, the booking CVRShare of visits or clicks that convert. at the payment step. The conversion rate calculator turns each arm's step counts into rates. A large drop here, even with attended appointments holding, means the deposit is turning people away who might have come.
- Complaints and support contacts about the deposit. Calls, messages and reviews that mention the deposit, its refund or its terms, counted by arm. A deposit that confuses people creates work for reception and can harm reviews.
Each Guardrail MetricA metric an experiment must not worsen beyond an agreed bound, watched beside the primary metric that decides the test. gets a limit written before launch, and a break means the deposit is not adopted in its tested form, whatever the primary metric shows.
Design
Randomise at the level of the visitor who starts the booking flow, and hold each visitor in one arm across sessions, so that someone who leaves and returns sees the same terms. Store the arm on the visitor and on any booking they make. In the deposit arm, the deposit is stated before payment: how much it is, whether it comes off the final bill, when it is refunded and how much notice a cancellation needs. The no-deposit arm sees the same flow, the same price and the same cancellation terms, with no payment taken until the visit.
| Design element | Choice | Reason |
|---|---|---|
| Unit of randomisation | Visitor who starts the booking flow, held in one arm across sessions | The deposit acts on the decision to book and on turning up |
| Arms | Deposit stated before payment; no payment until the visit | A simple A/B TestControlled comparison of two variants. answers the business decision |
| Held constant | Prices, cancellation terms, confirmations and reminders | Only the deposit differs |
| Phone bookings | Matched to an arm where the caller can be identified; reported separately where not | Keeps phone demand visible without breaking the randomisation |
| Reading point | After every booking made during the run has reached its appointment date | No-shows can only be counted once the slot has passed |
| Primary metric | Attended appointments per visitor who starts the flow | Counts bookings lost and slots saved together |
How the deposit is shown matters as much as whether it exists. The booking flows guide states, drawing on the hospitality industry guide, that the price shown at the start should be the price the person pays, and that the Competition and Markets Authority is among the bodies that address prices that do not show what the customer will have to pay. Its practical rule for deposits is to state the amount, the refund and the terms before payment, in plain words. Both arms meet that standard; the deposit arm simply has more to state. This is general direction, not legal advice.
Reminders stay identical. The booking flows guide and the healthcare industry guide treat appointment reminders as service messages, while promotional texts need consent. If the deposit arm received an extra reminder about its deposit, the test would measure the reminder as well. The confirmation also stays the same apart from stating the deposit taken; the guide to confirmation pages covers what it shows and how each booking is counted once.
Both arms share one diary, which is the design's particular risk. If the no-deposit arm books more slots, the deposit arm's visitors find fewer times available, and the arms are no longer independent. Run the test in periods with spare slots, and stop the reading for any week in which the diary fills.
Sample size reasoning
The primary metric is a rate per visitor counted at the appointment, so the inputs are the current attended-appointment rate per visitor who starts a booking, the share of bookings that are missed and the traffic to the booking flow. The sample size and significance guide shows how many visitors are needed to detect the smallest change in attended appointments that would be worth acting on, and the conversion rate calculator gives each arm's rate once the counts are in.
The two effects of a deposit partly cancel, so the net change in attended appointments is often smaller than either the drop in completion or the fall in no-shows. Size the test on that net change, not on the change in no-shows, or it will be too small to answer the question it was built for. A clinic or salon with modest traffic may find that the test needs a long run; if so, decide whether a deposit for one high-cost service is the better question.
Duration
Run for whole weekly cycles of bookings, so that weekday and weekend patterns fall equally in both arms. Then hold the reading open until every booking made during the run has reached its appointment date, so that no-shows are counted in both arms. Avoid periods when the diary is already full, because shared capacity distorts the comparison. Fix the end of bookings and the reading date before launch.
Reading the results and the decision rule
Read the test once, on the reading date. Compare attended appointments per visitor who started a booking between the arms, with the interval around the difference. Then check the guardrails against their limits, and only after that, read any service split named before launch, such as long first sessions against short follow-ups. Write the decision rule into the plan in advance:
| Result | Decision |
|---|---|
| Attended appointments per visitor rise reliably, and booking completion and complaints stay within their limits | Take deposits |
| Attended appointments per visitor do not rise | Take no deposit |
| The service split named before launch shows a rise for the services where a missed slot costs most | Consider deposits only for those services, and confirm with a test on them |
| Attended appointments rise but complaints or completion break their limits | Keep no deposit, rework how the deposit is explained and retest |
Pitfalls
Both arms competing for one diary. A full diary turns the test into a race for slots. Run when capacity is spare, and check weekly that both arms could see similar availability.
Phone bookings left out. A visitor who meets a deposit online may phone instead. The guide to measuring calls and chats sets out how to record whether a call became a booking; match calls to arms where possible and report the unmatched ones, so a shift to the phone is visible rather than counted as a lost booking.
Reading before appointment dates. Bookings made late in the run may fall weeks later. A reading taken at the end of bookings counts the deposit arm's lost bookings but not the no-deposit arm's missed slots.
A deposit that appears only at payment. The booking flows guide names a deposit revealed late as one of the surprises that loses people. If the deposit arm's terms appear for the first time at payment, the test measures the surprise rather than the deposit.
The card-required trial test is the closest design in software, asking for payment details at sign-up, and the demo scheduling test changes how a sales meeting is booked rather than whether a payment is taken.
Example: an invented physiotherapy clinic tests a booking deposit (illustrative)
Bezmorant Physio tests a deposit at online booking
Bezmorant Physio, an invented physiotherapy clinic, sees too many missed first sessions. It tests a £20 deposit, taken off the bill and refunded with a day's notice, stated on the first screen of the booking flow. Over six weeks, 3,000 visitors start the flow and are split at random, 1,500 to each arm; the reading waits until the last booked date has passed, four weeks later. The plan names one service split in advance: first assessments against follow-up sessions.
| Arm | Bookings | Completion | No-shows | Attended | Attended per visitor |
|---|---|---|---|---|---|
| No deposit | 780 | 52% | 94 | 686 | 45.7% |
| Deposit | 705 | 47% | 28 | 677 | 45.1% |
The deposit cuts the no-show rate from 12% to 4%, but loses enough bookings at payment that attended appointments per visitor are level, with an interval of roughly −4 to +3 points. Six people contact the clinic about the deposit, inside the limit. In the pre-named split, first assessments show attended appointments per visitor of 44% with a deposit against 41% without, while follow-ups show no gain. Bezmorant keeps booking without a deposit and plans a second test of a deposit on first assessments only.
Common mistakes
Deposit tests usually go wrong at the reading rather than the set-up. The experiment library lists the designs that sit beside this one.
- Judging on the no-show rate. It falls whenever a deposit deters bookings; read attended appointments per visitor.
- Counting at the confirmation page. Missed slots happen later; wait for every appointment date.
- Changing reminders or terms in one arm. The test then measures the message, not the deposit.
- Ignoring the phone. Visitors who switch to calling are not lost bookings, and calls must be matched or reported.
- Running when the diary is full. Shared capacity makes the arms compete for slots.
- Reading every service after the result. Name the service split before launch and treat any other as a lead.
