WORKING TOOL / HUMAN REVIEW
Restaurant voice AI evidence log
Answer in briefLog what a human reviewer actually heard, link it to a test case and keep the source recording under appropriate access controls. The summary starts at zero. A blank test history is not a passing result, and a vendor’s published metric is not an entry in your own test log.
Loman Reference · Editorially reviewed November 9, 2026.
No published test results. This is your local working log. All summary numbers come from records you enter. We have not independently tested or ranked vendors here.
Review a local recording
Choose only audio you have permission to review. It stays in your browser and is not uploaded. This player does not transcribe or score it.
No recording selected.
Records use this browser tab’s session storage. Audio files are not stored. Download before closing the tab. Import replaces this tab’s log only after you confirm in the tool.
How this summary is calculated
The pass rate is passed cases divided by passed plus failed cases. Blocked cases are shown separately and excluded from that rate. Every row requires a case ID, configuration label, date, speaker code, language, reviewer, evidence reference and observation. Required fields do not establish that the underlying evidence is authentic; you remain responsible for the record.
The mean clarification count uses completed cases only. The optional mean response delay uses only completed cases with a timing value, and its own sample size is displayed. Neither figure is a benchmark of the whole product. Review the reporting guide before interpreting a small or uneven sample.
Protect the recording and the participant
Use pseudonymous codes and consented test speech. Do not upload customer calls, card details or confidential staff conversations here. The local player reads the file in your browser; it does not send it to a server or save it in the exported log. Exported logs contain the evidence reference and your notes, not the audio.
The tool prevents duplicate case IDs within the same run label. To rerun a case after a configuration change, use a new run label and keep the original result. Clear or replace records only when that is intentional. Session storage may be cleared by your browser, so keep a downloaded copy of work you need to retain.
Customer evidence for your voice test
Big Apple’s early call-review habit is a useful model for an evidence log: record what happened after a call ended, including calls that produced no order.
Results and operating context from Loman’s published customer case studies.
Big Apple Pizza & Pasta
- 100%
- missed calls eliminatedThe case-study headline reports this result; measurement period not specified.
- First couple of months
- operator review of every callThe operator followed up on hang-ups and calls without an order.
Add an outcome for calls that ended without an order, then record the follow-up. Big Apple’s operator reviewed every call and contacted hang-up or non-order customers during the first couple of months.
Read the Big Apple Pizza & Pasta case study ↗Watch Big Apple Pizza & Pasta on YouTube ↗
Record the first outcome before retesting
Use a test ID such as V001, a configuration label and a consented recording reference. Enter the observed outcome and a short reason: “Quantity correction acknowledged; final spoken request matched.” That sentence is an illustrative note, not a result from this site.
A later pass belongs in a new, clearly labelled run so the first failure remains visible. If you include a product number in a report, cite Loman’s company-published benchmark context separately; do not import it as a local observation.
Published November 8, 2026 · Substantively reviewed November 9, 2026