Reece

Repeatable QA, not demo theater

Test an AI receptionist with calls it can fail.

A polished routine call proves very little. This scorecard uses repeatable prompts, observable outcomes, stop conditions, and saved evidence so a team can decide whether a narrow call flow is ready—and retest it after any meaningful change.

Written by Reece Editorial · Reviewed against Reece product truth · Updated 2026-08-14

See how Reece works Hear Reece answer live

7 days free · 60 minutes · No credit card required

Current Reece product assurances

  • Keep your business number: Use your provider's forwarding instructions for all calls or missed calls only.
  • Cover configured calls 24/7: Answers configured business calls 24/7, including after-hours and overflow calls.
  • Review call results: The web dashboard keeps call recordings, transcripts, AI summaries, caller details, and outcomes together.
  • Test without a card: 7 days and 60 call minutes are included.

Freeze the test inputs

Write the approved greeting, hours, services, service area, transfer destinations, appointment type, and fallback language before calling. Use the same caller facts and expected outcomes on every vendor or configuration so the result is comparable.

Score observable facts

For each call, record whether contact details were correct, the intent was classified, approved questions were asked, unsupported facts were avoided, the next step was truthful, and the saved record matched the conversation. Do not score charm or voice preference as factual accuracy.

Include correction and interruption

Change a phone number mid-call, interrupt the answer, pause, speak over the receptionist, and correct a name. The transcript and final summary should contain the corrected value, not both versions presented as equally valid.

Force each integration failure path

Call when a transfer destination does not answer and request an appointment time. The appointment request must remain unconfirmed for staff follow-up. A failed transfer should preserve context and use the approved fallback rather than imply success.

Use explicit stop conditions

Stop the pilot for fabricated prices, invented availability, unsafe advice, lost corrections, false booking confirmation, repeated routing loops, exposure of restricted information, or a summary that materially changes the caller's request. Fix and rerun the full affected set.

Retest on a defined cadence

Rerun affected calls after changes to knowledge, prompts, forwarding, transfer destinations, appointment rules, provider credentials, languages, or pricing. Run the complete suite before launch and periodically for high-value paths.

Define success for how to test an AI receptionist

Begin with one written caller job and one observable business outcome. For this page, the working scope is: A polished routine call proves very little. This scorecard uses repeatable prompts, observable outcomes, stop conditions, and saved evidence so a team can decide whether a narrow call flow is ready—and retest it after any. Turn that scope into a acceptance record that names the caller's question, the allowed information source, the fields worth collecting, the next-action owner, and the evidence that proves what happened. Avoid goals such as “handle calls better” because they cannot distinguish a useful intake from a polished conversation that leaves the employee without a usable next step. A successful call can still end in human review; success means the boundary and ownership were truthful, not that automation completed every request.

Translate approved knowledge into caller-safe answers

Write the approved greeting, hours, services, service area, transfer destinations, appointment type, and fallback language before calling. Use the same caller facts and expected outcomes on. Use that source material to write short approved answers, but preserve conditions, exclusions, dates, and ownership. Website import can propose knowledge for review; it does not make every sentence current or safe to say on a call. Mark facts that change often—hours, service areas, prices, availability, provider connections, and transfer destinations—and assign the person responsible for the next action to approve them. When two sources conflict, keep the item inactive until a person resolves it. When a caller asks beyond the reviewed source, Reece should acknowledge the limit and take the approved conservative action instead of filling the gap from general knowledge.

Label every outcome by evidence, not by intention

For each call, record whether contact details were correct, the intent was classified, approved questions were asked, unsupported facts were avoided, the next step was truthful, and the. Define outcome states as captured, attempted, provider-confirmed, failed, or human-owned. A requested appointment is not a booked event, a configured transfer is not a connected call, and a captured phone number is not a qualified lead. For each state, specify what appears in the recording, transcript, summary, caller fields, action result, and employee queue. This vocabulary should be shared by the call flow, dashboard review, analytics event, and operating playbook. Consistent labels let a reviewer distinguish product behavior from staff follow-up and prevent an attempted action from becoming an unsupported completion claim in reporting or customer-facing copy.

How Reece handles the call

  1. Forward the calls you choose. Keep the number customers already know, then use your phone provider's instructions to forward all calls or missed calls only to Reece.
  2. Reece answers with approved knowledge. Reece answers the configured call and uses business knowledge the owner has reviewed before it becomes active.
  3. The caller explains what they need. Reece answers approved FAQs and gathers the caller, request, and intake details needed for a clear next step.
  4. A supported handoff happens. Eligible Scale and Growth setups can use configured direct or warm transfers. Otherwise, Reece records details or an appointment request for staff follow-up.
  5. Your team reviews the result. The recording, transcript, summary, caller details, and outcome stay together in the responsive web dashboard.

Your team stays in control. Reece handles approved call intake and configured routing. Your team still owns diagnosis, safety decisions, pricing, dispatch, appointment confirmation, and any follow-up that needs a person.

See how Reece works

12-call qualification sheet

Check a call only after reviewing its saved evidence. Use the note fields for the observed result, record ID, and repair owner.

Routine accuracy

Prove ordinary facts before testing integrations.

  • Hours and service-area question uses approved facts
  • New lead captures corrected name, number, location, and need
  • Existing customer is separated from a new-sales path

Conversation resilience

Create realistic speech and audio difficulty.

  • Caller interrupts and the receptionist resumes without repeating a promise
  • Caller corrects a key fact and the final record uses the correction
  • Background noise or silence triggers a conservative clarification

Boundaries

Ask for something the system must not invent.

  • Unsupported price receives an honest next step
  • Sensitive or safety-related request reaches the approved boundary
  • Spam or sales call is filtered without blocking a legitimate unknown caller

Handoffs

Test success and failure, not just initiation.

  • Answered transfer gives the recipient useful context
  • Unanswered transfer uses the fallback and preserves the call record
  • Unavailable appointment is not reported as booked

References and retrieval dates

Know the handoff before you forward a call

Can I keep my current number?

Plans include 1, 3, or 10 Reece phone numbers. Businesses can keep an existing number and forward selected calls to Reece.

Can Reece take missed calls only?

Yes. Choose missed-call forwarding when staff should have the first chance to answer, or forward all calls when Reece should answer first. Your provider controls forwarding behavior.

What if the caller needs a person?

Scale and Growth support owner-configured direct and warm transfers. Reece does not include a staffed human fallback, so the team must define the follow-up path for calls that are not transferred.

Does an appointment request mean booked?

Reece can capture appointment requests, preferred times, caller details, and approved intake answers for staff follow-up. The current production UI does not offer customer-configurable calendar connections or direct provider-backed booking, so Reece does not report an appointment as confirmed.

Setup is guided, but completion time varies with business rules, transfer routing, testing, and carrier forwarding. Call results are available in the web dashboard. Production email delivery and SMS notifications are not currently offered.

Privacy and sensitive calls. Review the current Privacy Policy and Terms before forwarding calls that may include sensitive information. Reece does not claim HIPAA compliance; medical, legal, emergency, financial, and other consequential calls may require a qualified person. Privacy Policy · Terms

Questions, answered

How many calls should I test?

Twelve is a useful minimum for the distinct paths here. Add calls for every material service, language, location, routing rule, and high-consequence exception in your business.

What evidence should I save?

Save the input script, timestamp, recording, transcript, summary, outcome, provider result for a booking or transfer, expected result, actual result, and repair owner.

Does this worksheet send or save my answers?

No. The checklist and notes stay in the current browser page. Printing uses the browser's print dialog; this page has no lead form, email action, or gated download.

Does a completed worksheet mean the call flow is ready?

No. Completion means the decisions were documented. Run realistic calls, review the actual record, and obtain any legal, safety, or professional approval the workflow requires before relying on it.