Your AI Booked It. Was the Appointment Actually Confirmed?

A booking request and a confirmed booking are different outcomes. Count a request when a new appointment record is saved. The confirmed-booking goal counts when that new linked appointment is saved as confirmed or explicitly marked completed within the measurement window. Confirmation alone does not prove attendance; an explicitly completed record provides separate evidence of completion. An automatic move to pending_completion never counts as confirmation.

That distinction is easy to miss when your team says, “The AI booked twelve appointments.” Were twelve requests recorded? Were twelve times accepted? Did twelve people attend? Each statement needs different evidence.

When experiments are enabled in your paid Receptionist workspace, choose the booking event that matches your decision. Access depends on your workspace and team permissions. Examples are hypothetical and do not describe customer performance.

Which booking outcome matches the business decision?

Choose the stage you want to improve before changing the receptionist. If staff must review every request, a request goal may be useful. If the assistant can secure an accepted appointment, confirmed booking is closer to the desired result. If the problem is attendance, neither goal alone answers it.

Evidence	What it supports	What it does not prove
A saved new appointment	A booking request exists	Confirmation or attendance
Saved as confirmed or explicitly marked completed	A confirmed booking; completion recorded separately where present	Attendance from confirmation alone, or payment
An automatic status change after the appointment time	Time passed	Customer confirmation
A later cancellation or no-show	A separate downstream event	That no earlier confirmation happened

A test can improve one stage while leaving the next unchanged. More requests may create more work for staff if the requests are incomplete. More confirmed appointments may still produce no-shows. Pick the primary goal that matches your decision, then inspect the other stages as context.

Microsoft’s experimentation guidance recommends measuring a broad set of effects so an improvement in one metric does not conceal a regression elsewhere. This is the reason to watch downstream outcomes alongside the selected goal, not a reason to replace the goal after results arrive. Microsoft Research

Why should the record outrank the wording of a chat?

A message saying “you’re booked” is not the appointment record. The save may have failed, the appointment may still need approval, or the message may refer to an existing booking. An experiment should count the event it can verify.

Receptionist uses saved business events rather than an AI judgment that a chat “looks successful.” The appointment must be new, linked to the assigned participant and within the allowed window. Repeated updates do not turn one appointment into several successes for the same participant and goal.

The source also matters. An old appointment imported into the system is not a new conversion. A voice call, an advertising booking page or an unrelated calendar entry must not be credited to a messaging test merely because the same customer exists in the workspace.

This is narrower than measuring the value of your whole business. It measures the outcome attributable under a specific test’s rules. Measuring AI return on investment needs that wider financial view; the experiment supplies one defined part of the evidence.

What does pending_completion mean for the test?

The current appointment lifecycle can move both pending and confirmed appointments to pending_completion after their scheduled end. That timer cannot establish which appointments were confirmed beforehand.

The confirmed-booking goal requires a new linked appointment to be saved as confirmed or explicitly marked completed within the measurement window. A pending appointment that only moves automatically to pending_completion does not count. Each participant can count once for this goal. AI confirmation wording alone does not change the saved appointment status.

This correction affects how success is measured. It does not turn a pending appointment into a confirmed one, rewrite old bookings or claim that the customer attended.

How long should you wait for the outcome?

Choose a window that gives your process time to work. You can choose 24, 72, 168 and 336 hours. These are available measurement windows, not recommended test durations or guarantees of sufficient evidence.

The enrollment period and outcome window solve different problems. Enrollment decides how long new participants enter. The outcome window decides how long an assigned participant has to reach the goal. If enrollment ends today, a participant enrolled near the end may still have time left.

Do not collapse that waiting period into a loss. Keep pending outcomes visible until their windows close. If the source event cannot be observed reliably, keep the uncertainty visible too. An unknown outcome and a measured failure are different facts.

Small teams may need patience. A full calendar week is not automatically enough traffic for a reliable conclusion. The decision also depends on the existing success rate, the difference you need to detect and whether your tracking remains complete.

How should a clinic review a finished test?

Start with the goal label, target account and measurement window. Then compare participant counts, successes, outcome rates, unresolved records and spend. Open the linked conversations for context where your permissions allow it.

Look for practical problems that a summary rate can hide: customers being offered an unsuitable service, staff having to correct details, or bookings created for a target the business cannot serve. These observations can justify a new hypothesis. They do not justify selectively removing inconvenient participants from the existing test.

Keep the result together with its saved configuration. If your team adopts B, record why. If evidence is inconclusive, retain A or design a new test. A later restoration must still check the current target and permissions.

Attendance needs its own review. Our guide to reminders and confirmation calls addresses what happens after a booking. A receptionist conversation test should not claim that evidence as its own measured effect.

What happens when the test ends or evidence is missing?

At the planned end, normal routing is the default. You may instead explicitly preselect saved A, B or C where available. That scheduled choice still needs experiment availability, current target and dependency checks; it is not an automatic winner selection. A blocked choice needs attention and must not be reported as applied. Existing participants can continue through their outcome windows.

Unknown outcomes have a strict consequence: even one unresolved unknown can prevent a statistical recommendation for that run. Descriptive results and an explicit user choice remain available. A manually stopped run without a completed planned analysis horizon is not proof of a winner. On a downgrade to Free, experiment access and new enrollment close, and experiment handling is interrupted safely. Ordinary Inbox records remain.

FAQ

Does a customer saying “yes” count as a confirmed booking?

Only if the new linked appointment record is saved as confirmed or explicitly marked completed within the test’s window. The wording of a message by itself is not the saved outcome. The automatic pending_completion status never counts.

Can one customer create several successes?

The primary goal is binary per participant. Several qualifying events can supply evidence, but they do not turn one participant into multiple primary successes.

Is a confirmed booking the same as an attended appointment?

No. Confirmation alone does not prove attendance or payment. An explicitly completed record provides separate evidence of completion. A confirmed-booking test is still not a standalone measure of attendance or revenue.

Can the test count existing appointments?

The booking goals require new, attributed appointments. Existing appointments and unrelated sources are excluded. This prevents routine updates from being counted as new business.

Bring the appointment statuses your team actually uses to a Receptionist review. Book a demo to agree on what a successful booking should mean before comparing versions.

Sources and measurement definitions checked 10 October 2026. No uplift or attendance improvement is claimed.