{"resourceId":"phoenix-pd-draft-one-90-day-pilot-metrics-20261007","versions":[{"version":"external-0ea8afa86fd006c13a9ae66287e29e5af94be7717a97d657e788258e60666135","resource":{"id":"phoenix-pd-draft-one-90-day-pilot-metrics-20261007","title":"Phoenix police start a 90-day Axon Draft One pilot with signed attestations, a planted error officers must find, and overtime and report-time metrics","organization":"Phoenix Police Department / Phoenix City Council (reported by ABC15 Arizona)","sector":"Municipal policing","geography":"Phoenix, Arizona, United States","publishedAt":"Oct. 7, 2026, 6:03 PM MST (ABC15)","publicationDate":"2026-10-07","eventDate":"2026-10-01","sourceName":"ABC15 Arizona","sourceLabel":"ABC15 report on Phoenix PD's Draft One pilot design and metrics","sourceUrl":"https://www.abc15.com/news/phoenix-police-launch-ai-pilot-program-using-body-worn-camera-audio-to-write-first-draft-police-reports","evidenceClass":"independent-reporting","outcomeClass":"emerging","topics":["knowledge-work","governance-procurement","accessibility-workforce","operating-model"],"finding":"ABC15 reported that the Phoenix Police Department launched a 90-day pilot of Axon Draft One, which generates first-draft police report narratives from body-worn camera audio, running Oct. 1 through Dec. 31, 2026 in the Mountain View Precinct Patrol Bureau, the Family Investigations Bureau and the Mountain View Assaults Unit. Councilwoman Ann O'Brien, who pushed for the pilot, framed it as a response to officer vacancies. Each draft must be reviewed and corrected and carries a signed officer attestation before entering the records system, and O'Brien said an intentional error is placed in every draft that the officer must find before signing. The department will track overtime, response times, calls answered, arrests, body-camera activation and report-writing time for patrol, and case-management and call-out overtime, case disposition and report time for investigations, then decide whether to extend or expand.","sledRelevance":"Directly relevant to policing AI: a large-city department is defining measurable success criteria and an anti-rubber-stamp control before deciding on department-wide AI report drafting.","evidence":"Local TV reporting based on interviews with the council sponsor and a description of the pilot design. No results exist yet; the metrics list and the planted-error control are as described by O'Brien, not from a published Phoenix PD policy or Axon configuration document.","architectureImplications":"Drafts are generated only after body-camera footage is uploaded and transcribed in Axon's environment, then edited and attested before entry into the department's records management system, so the integration point is the RMS hand-off.","governanceImplications":"Human accountability is enforced by attestation and a deliberately inserted error; the pilot has a fixed end date and a decision point, which gives the council a natural review gate.","securityPrivacyImplications":"Drafts derive from recorded audio of public encounters; the report does not describe retention of AI drafts, disclosure labels or prosecutor access to drafts, which should be confirmed.","caveats":"No outcome data yet. The planted-error safeguard is described by the councilwoman, not documented by the department. A 2024 randomized trial (not opened tonight) and prior Hays County experience in the archive suggest time savings may not materialize; Phoenix's results should be compared against a baseline.","streamIds":["public-safety"],"roles":{"sales":"Sales framing: Phoenix gives a template for selling AI report drafting as a measured pilot, not a productivity promise. Discovery should ask which units are included, what baseline report time and overtime are, how attestation and planted-error checks are configured, and who decides after Dec. 31. A bounded engagement is an evaluation design and baseline measurement for the pilot window; do not claim time savings or more officers on the street before data exists.","engineering":"Engineering framing: the integration point is the hand-off from Axon's draft to the records management system. Engineers should confirm how drafts are labeled and retained, whether the planted-error check is a product setting or a manual step, and how report-time is measured from system timestamps. Proof of value should compare pilot and non-pilot units on the same metrics, with care for family-violence and assault cases where transcription errors carry higher risk.","delivery":"Delivery framing: the department needs a pilot runbook covering training on editing drafts, attestation, the planted-error check, metric collection and an end-of-pilot report. Ownership sits with the police department's patrol and investigations leadership, with the council's public safety subcommittee as the reviewer. Risks are untracked baselines, officers accepting drafts without real edits, and a pilot quietly becoming permanent."},"roleDetails":{"sales":{"customerProblem":"Officer vacancies and hours of report writing push leaders toward AI drafting, but they need evidence before committing department-wide.","stakeholders":["Police chief and patrol bureau commanders","Family Investigations and Assaults Unit supervisors","Councilmember sponsoring the pilot and the public safety subcommittee","County prosecutors who receive reports","Records management system administrators"],"discoveryQuestions":["What baseline report-writing time and overtime exist for the pilot units?","Is the planted-error check enforced in the software or by supervisors?","Are original AI drafts retained and disclosed to prosecutors and defense?","What threshold on the tracked metrics would justify expansion?"],"valueHypothesis":"A well-instrumented 90-day pilot can show whether drafting reduces report time and overtime in specific units, giving a defensible buy or stop decision.","potentialEngagement":"Design the pilot evaluation: baseline extraction, comparison units, metric dashboards and an end-of-pilot decision memo.","unsupportedClaims":["Guaranteed reductions in report time or overtime","Improved response times or more officers on patrol","Report accuracy improvements","Suitability for all incident types"]},"engineering":{"fit":"Fits departments already on Axon body cameras and evidence storage with an RMS that accepts drafted narratives.","architectureIntegration":"Body-camera audio is uploaded and transcribed in Axon, a draft is generated, edited and attested, then entered into the department's RMS; metrics come from RMS and payroll data.","prerequisites":["Axon body cameras and evidence platform","RMS integration or manual paste workflow","Payroll and CAD data for overtime and response-time metrics","Defined comparison units"],"constraints":["Drafts limited to recorded, transcribed audio","Sensitive family-violence cases","Pilot ends Dec. 31, 2026","Arizona public-records and disclosure rules for drafts"],"security":"Confirm access controls on transcripts and drafts, retention of AI drafts, audit trail of draft generation and edits, and disclosure labeling.","proofOfValue":["Compare report-writing time against pre-pilot baseline for the same units","Measure the share of drafts where the planted error was caught","Sample attested reports for factual errors against video","Track overtime changes against non-pilot precincts"]},"delivery":{"work":"Run the pilot: train officers, configure attestation and the planted-error check, collect metrics, and deliver an end-of-pilot evaluation.","dependencies":["Axon configuration access","RMS and payroll data extracts","Supervisor review capacity","Council subcommittee schedule"],"ownership":"Phoenix PD patrol and investigations leadership, reporting to the City Council's Public Safety and Justice Subcommittee.","skillsAdoption":"Officers need training to verify and correct drafts rather than accept them; supervisors need sampling routines.","governanceCheckpoints":["Written pilot policy before go-live","Mid-pilot metric review","End-of-pilot report before any expansion vote","Prosecutor consultation on draft disclosure"],"acceptanceCriteria":["Proposed: baseline and pilot report-time measured for every pilot unit","Proposed: at least 95% of sampled drafts show the planted error was caught","Proposed: no attested report found to contain uncorrected AI-introduced factual errors in audit sample","Proposed: evaluation memo delivered within 30 days of Dec. 31"],"risks":["Rubber-stamping despite attestation","Missing baseline makes results unprovable","Errors in sensitive investigations","Pilot extended without evaluation"]}},"retrievedAt":"2026-10-08T02:55:05Z","enrichedAt":"2026-10-08T02:59:04Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Framed as relief for an officer shortage; officers must learn to edit and verify drafts, and investigators in family and assault units will use it on sensitive cases.","procurementImplications":"A time-boxed pilot with named metrics positions Phoenix to make a buy decision on measured overtime and report-time data rather than vendor claims.","operatingModelImplications":"Results will be assessed after Dec. 31; the department may extend the evaluation or move toward department-wide use.","sourceVerification":{"openedUrl":"https://www.abc15.com/news/phoenix-police-launch-ai-pilot-program-using-body-worn-camera-audio-to-write-first-draft-police-reports","referenceExcerpt":"there actually is an error placed in the report that the officer must find in order for them to be able to sign off","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}