{"resourceId":"arlington-nonemergency-ai-diversion-2025","versions":[{"version":"external-19b1c8be472418e0ceaf2e7c840922462c46e60e07e887514dae5994a638fabf","resource":{"id":"arlington-nonemergency-ai-diversion-2025","title":"Arlington describes non-emergency call diversion with opt-out and reporting-portal dependencies","organization":"Arlington County Public Safety Communications and Emergency Management; Timothy Kane","sector":"Emergency services","geography":"Arlington County, Virginia, United States","publishedAt":"Undated PDF; includes operational data through February 15, 2025","publicationDate":null,"eventDate":null,"sourceName":"Use of AI for Non-Emergency Call Diversion in PSAPs, hosted by National 911 Program","sourceLabel":"Local government operator case report","sourceUrl":"https://www.911.gov/assets/Use-of-AI-for-Non-Emergency-Call-Diversion-in-PSAPs.pdf","evidenceClass":"government-evaluation","outcomeClass":"emerging","topics":["knowledge-work","infrastructure","data-security","governance-procurement","accessibility-workforce","operating-model"],"finding":"Arlington reports non-emergency workload reductions after call routing and related process changes.","sledRelevance":"Interpretation: Storm-damage and fire-alarm administrative workloads make this relevant to emergency communications; towing is ancillary, with no Public Safety cross-tag.","evidence":"Operator describes approximately 30 days of utterance collection, AWS routing and SMS links, with human opt-out. It reports 22,167 fewer administrative calls comparing the first six months of 2022 and 2024. No controlled comparison, misrouting rate or confirmed emergency-response improvement.","architectureImplications":"Interpretation: Keep non-emergency routing separate from emergency handling; monitor portal availability and completed handoffs rather than SMS sends alone.","governanceImplications":"Interpretation: Define unsafe routing categories and escalation tests with dispatch leadership. Audit the complete caller journey.","securityPrivacyImplications":"Interpretation: Control retention of utterances and phone numbers, restrict routing changes and verify destination links.","caveats":"Operator report with concurrent towing outsourcing; reduction cannot be attributed solely to AI. Visual dashboard pages did not yield inspectable images through tools. Future multilingual bot capabilities are not evaluated results.","streamIds":["emergency-services"],"roles":{"sales":"Interpretation — Ask the communications director, emergency manager and service-portal owners which storm and administrative calls consume staff time. Determine whether callers can already complete those requests online and how exceptions reach a person. Offer a limited review of one non-emergency category with an accessible fallback. The value hypothesis is fewer avoidable call-taker interactions while preserving resolution quality. Arlington's historical comparison is a useful lead, not an independently measured return. Do not promise faster emergency response, annualized savings or equal benefits for another county. Qualify portal readiness, language needs and operating support before proposing a routing service.","engineering":"Interpretation — Build a test route that recognizes a narrow intent and offers a verified destination while retaining immediate human escalation. Prerequisites include approved call examples, an operational reporting portal, telephony ownership and an agreed emergency exception policy. Test ambiguous storm descriptions, declined SMS, broken links, speech errors and cloud outages. Log transfer completion without excessive personal-data retention. Compare correct resolution and staff handling time with existing routing, not just the number of workflow interactions. Keep future knowledge-base agents outside the initial proof unless separately evaluated. The report lacks misrouting evidence, so adversarial and accessibility tests must precede live expansion.","delivery":"Interpretation — Dispatch operations should own a staged pilot with portal teams and telephony IT. Baseline the selected category, review its utterances, configure exceptions and train call-takers to recover failed diversions. Dependencies include maintained public pages and support for callers who cannot use digital links. Approve each route before launch and review complaints, transfers and material service changes. Proposed acceptance includes independently checked routing accuracy, successful human fallback, measured completion and no unresolved critical misroutes in the agreed test set. These are proposed criteria. Track repeat calls and abandoned journeys so apparent call reduction does not conceal unmet needs."},"retrievedAt":"2026-09-09T03:02:22Z","enrichedAt":"2026-09-09T03:04:34Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: Provide a usable human route for callers unable to receive SMS or use online portals; test languages and speech differences.","procurementImplications":"Interpretation: Price cloud usage, portal maintenance and monitoring; require routing export, failure support and evidence of successful resolution.","operatingModelImplications":"Interpretation: Dispatch operations owns routing, service departments own destination portals and IT owns telephony continuity. Knowledge-base bot development remains prospective in this source.","updateExplanation":"New source after full-archive URL review. Historical case fills non-emergency storm-call operations coverage; no new deployment announcement or exact publication date inferred.","sourceVerification":{"openedUrl":"https://www.911.gov/assets/Use-of-AI-for-Non-Emergency-Call-Diversion-in-PSAPs.pdf","referenceExcerpt":"they are then put into the non-emergency queue to be answered by a call-taker.","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}