TAPEWALL / SETUP AND COMPATIBILITY
AI booking agent acceptance test matrix
A copyable ten-case booking QA matrix for agencies, with calendar evidence, failure rules and a CSV download.
FREE TEST SPECIFICATION / UPDATED 26 OCTOBER 2026
Decide what counts as a booking before the first test
Write down the approved service, start and end time, staff member, synthetic customer and expected calendar state for each case. Save a pre-run snapshot. A spoken confirmation and a successful tool response do not replace the final booking record. Keep failed, passed and unknown results separate.
This matrix is a test plan, not a customer benchmark. It does not claim that Tapewall currently automates every row. Run it only against an authorized agent and isolated test calendar.
Download the CSV matrixTen acceptance cases
| ID | Scenario | Setup | Expected result | Failure or review rule |
|---|---|---|---|---|
| B01 | New booking | Empty isolated test calendar | Exactly one confirmed event with approved service, start, end, staff and synthetic customer | Fail if absent, duplicated, tentative or any field differs |
| B02 | False confirmation | Booking write is rejected by test adapter | No new event; agent does not claim a confirmed booking | Fail if agent confirms despite no event |
| B03 | Unavailable slot | Prebook the requested test slot | No second confirmed event for that slot; agent offers an allowed alternative | Fail if duplicate or unavailable time is confirmed |
| B04 | Reschedule | One known synthetic event exists | Original appointment changes to the requested time; one active event remains | Review if pre-run event is missing; fail if time is unchanged or two events remain |
| B05 | Cancellation | One known synthetic event exists | Target event is no longer active; unrelated events remain unchanged | Review if pre-run event is missing; fail if target remains active |
| B06 | Repeated request | Submit the same booking intent twice | One active event, with no duplicate after the retry | Fail if two active events exist |
| B07 | Wrong staff | Request a named staff member | Confirmed event has the approved staff member or agent declines the request | Fail if agent promises one person but books another |
| B08 | Time zone boundary | Use approved local time and an unambiguous date | Calendar instant, local display and duration all match the approved expectation | Fail on shifted time or duration; review ambiguous expectations |
| B09 | Provider timeout | Make the booking result unavailable to the test agent | Agent reports uncertainty; calendar state is inspected before retry | Review unknown provider state; fail on an unsupported confirmation or duplicate retry |
| B10 | Cleanup | Finish a synthetic test run | Read the isolated calendar again and find no active event from that run | Review if cleanup fails or final evidence cannot be read |
Evidence to retain for each run
- Approved facts and the version of the agent configuration.
- The exact customer script, run identifier and transcript.
- Calendar state before the conversation, immediately after it, and after cleanup.
- Event identifiers, status, service, time, staff and synthetic customer fields.
- The prewritten pass, fail or review rule and a named person to investigate an unknown result.
For Google Calendar, the official event resource documents event identifiers, status, start and end fields. The events list endpoint supports reading calendar state. This template is Tapewall’s proposed acceptance method, not a claim made by Google.
Use it in a pilot
Start with one booking flow and an agency-approved test environment. Tapewall currently has fixture-tested calendar readers and a private engineering app; an actual customer agent and calendar still need supervised validation. We review compatibility, written scope and costs before payment.
Submit a compatibility brief