{"resourceId":"nz-police-axon-auto-transcribe-interviews-80-percent-accuracy-verification-burden-20261102","versions":[{"version":"external-f35fed114c00014881201023b371f5acf19d7e1eb4c194c2b451f935d9dd9d60","resource":{"id":"nz-police-axon-auto-transcribe-interviews-80-percent-accuracy-verification-burden-20261102","title":"New Zealand Police begin using Axon Auto-Transcribe for interviews, with vendor-stated 80% average accuracy and barristers warning checking will shift to lawyers","organization":"New Zealand Police","sector":"Policing / AI transcription of investigative interviews","geography":"New Zealand","publishedAt":"RNZ, Nov. 2, 2026","publicationDate":"2026-11-02","eventDate":"2026-11-02","sourceName":"RNZ","sourceLabel":"New AI tool for police interviews 80 percent accurate","sourceUrl":"https://www.rnz.co.nz/news/science-and-technology/2292343/new-ai-tool-for-police-interviews-80-percent-accurate","evidenceClass":"independent-reporting","outcomeClass":"emerging","topics":["knowledge-work","data-security","operating-model"],"finding":"RNZ reported that New Zealand Police began using Axon Auto-Transcribe for interview speech-to-text on Nov. 2, 2026. Axon says the tool averages 80% accuracy on public safety data and over 90% on clearer audio; police acknowledged the transcripts are not always accurate and said investigators must corroborate information and record that verification before acting. Court transcripts will still be prepared and verified by people in the police Transcription Service. Prosecutors may give the auto-transcript, with the recording, to defendants. A barrister warned of breaches and shortcuts in an underfunded system; another said checking would shift to prosecutors and defence counsel, and raised lock-in and data-sovereignty concerns about overseas vendors. RNZ learned of the tests through an Official Information Act release.","sledRelevance":"Shows a police agency adopting AI transcription with an explicit human-verification rule while the accuracy figure comes from the vendor.","evidence":"Opened and read the full RNZ article. The 80% and 90% figures are Axon claims as reported; no police evaluation was published.","architectureImplications":"Transcripts are generated inside the vendor evidence platform, including from officer smartphones; Axon stores some NZ data in Australian datacentres.","governanceImplications":"Police say they followed their new-technology policy; verification must be recorded on file. Disclosure of auto-transcripts to the defence needs clear labelling.","securityPrivacyImplications":"Barristers raised breach risk and data stored offshore with an overseas vendor.","caveats":"Vendor accuracy claim not independently tested; New Zealand privacy and evidence law differ from U.S. state law; time savings are expected, not measured.","streamIds":["public-safety"],"roles":{"sales":"Sales framing: New Zealand shows police adopting AI interview transcription while openly saying the output is not always accurate. The buyer problem is transcription backlog without weakening evidence quality. A bounded engagement could measure accuracy on the agency’s own interview audio and design the verification record. Do not promise savings; the 80% figure is Axon’s average, and defence lawyers expect the checking burden to move to them.","engineering":"Engineering framing: Auto-Transcribe runs in the vendor evidence platform, including from smartphones. Pre-sales should ask where audio and transcripts are stored, how drafts are labelled, and how verification is recorded. A proof of value would score word and meaning errors on a sample of local interviews with accents and poor audio. Constraints: data location, vendor lock-in, and evidence rules requiring human-verified transcripts for court.","delivery":"Delivery framing: the work is a two-tier process where machine drafts aid investigators and court transcripts stay human-verified. Ownership sits with the investigations lead and the transcription service manager. Acceptance criteria are proposals: measured accuracy on local audio, a recorded verification step before action, clear draft labels in disclosure, and a review with defence bar input. Main risks: unchecked drafts used as fact and workload shifted to lawyers."},"roleDetails":{"sales":{"customerProblem":"Interview transcription backlogs slow investigations, but machine transcripts can be wrong in ways that matter in court.","stakeholders":["Investigations lead","Transcription service manager","Prosecutors","Defence bar representatives"],"discoveryQuestions":["What accuracy does the tool reach on our own interview audio?","How is verification recorded before investigators act?","Where are recordings and transcripts stored?"],"valueHypothesis":"Faster draft transcripts for investigators, if accuracy is measured locally and verification is recorded.","potentialEngagement":"Local accuracy test and verification-process design for AI interview transcription.","unsupportedClaims":["Vendor 80% average as local accuracy","Measured time savings","Court-ready transcripts without human review"]},"engineering":{"fit":"Relevant to agencies with interview recording and transcription backlogs.","architectureIntegration":"Speech-to-text inside the vendor digital evidence platform, including smartphone capture, feeding case files and disclosure.","prerequisites":["Recorded interview audio in the platform","Draft labelling","Verification field in the case file"],"constraints":["Offshore data storage","Evidence rules on transcripts","Vendor lock-in"],"security":"Confirm storage location and access controls, and keep drafts distinguishable from verified transcripts.","proofOfValue":["Error scoring on local interview sample","Meaning-changing error count","Verification-step audit"]},"delivery":{"work":"Pilot auto-transcription for investigation use while court transcripts stay human-verified.","dependencies":["Vendor platform configuration","Transcription staff","Prosecution disclosure process"],"ownership":"Investigations lead and transcription service manager.","skillsAdoption":"Investigators need training on checking drafts; transcription staff move to court-transcript work.","governanceCheckpoints":["Local accuracy measured","Verification rule in policy","Defence bar consulted"],"acceptanceCriteria":["Accuracy on local sample reported","Verification recorded on every file where a draft is used","Drafts labelled in all disclosures"],"risks":["Unchecked drafts used as fact","Workload shifted to lawyers","Data breach or offshore exposure"]}},"retrievedAt":"2026-11-03T03:54:34Z","enrichedAt":"2026-11-03T03:57:16Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Human transcription staff shift to court transcripts; checking work may move to prosecutors and defence counsel.","procurementImplications":"Accuracy should be measured on the agency’s own audio before relying on vendor averages; exit and data-location terms matter.","operatingModelImplications":"A two-tier model: machine drafts for investigation, human-verified transcripts for court.","sourceVerification":{"openedUrl":"https://www.rnz.co.nz/news/science-and-technology/2292343/new-ai-tool-for-police-interviews-80-percent-accurate","referenceExcerpt":"Before acting on information, investigators must confirm and corroborate it and record that verification on the file.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}