{"resourceId":"kosdi-seoul-ai-119-callbot-wait-733-accuracy-20260331","versions":[{"version":"external-18b6be30910069ba304be7082741f2b99c5ff34e1443c4e315fbbe57f800396e","resource":{"id":"kosdi-seoul-ai-119-callbot-wait-733-accuracy-20260331","title":"KOSDI study: Seoul Emergency Operations Center AI 119 callbot cut average wait ~7.33% with >70% early urgency accuracy","organization":"Seoul Emergency Operations Center / Chung-Ang University","sector":"Fire / 119 emergency call-taking","geography":"Seoul, Republic of Korea","publishedAt":"2026-03-31","publicationDate":"2026-03-31","eventDate":null,"sourceName":"Journal of the Society of Disaster Information (KOSDI)","sourceLabel":"Peer-reviewed academic case study of Seoul AI callbot operations","sourceUrl":"https://cdn.apub.kr/journalsite/sites/kosdi/2026-022-01/N0620220131/N0620220131.pdf","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["knowledge-work","operating-model","infrastructure"],"finding":"Mun and Park analyze Seoul's AI-supported 119 call-taking: STT/NLP/urgency models assist human call-takers rather than replace them. Average wait fell about 7.33% (19.79s in 2024 to 18.34s in 2025). In an Aug. 13, 2025 110-minute rain surge, the callbot handled 2,150 of 2,898 calls; of 412 judged calls, urgency accuracy was 72.2% (emergency) and 73.4% (non-emergency), with 81% unjudged due to abandon or immediate human connect. Authors urge better training data, explainable AI and governance. Why now: the Nov. 5, 2026 Digital Times council audit of the same Seoul 119 AI program makes these ops metrics newly relevant for agencies deciding go-live gates.","sledRelevance":"Provides rare measured PSAP wait-time and surge-accuracy figures for fire/EMS call AI that U.S. centers can use as a benchmark—while today's Seoul audit shows why those figures must sit under independent governance.","evidence":"Open-access PDF DOI 10.15683/kosdi.2026.3.31.389; Tables 3–6 wait and surge accuracy; architecture described as assistive to existing exchange/CAD.","architectureImplications":"Prefer assistive AI analysis servers feeding call-taker UI/CAD with queue-threshold callbot engagement over unsupervised answer.","governanceImplications":"Even positive pilot metrics need institutional frameworks and explainability before scale—consistent with later Seoul oversight conflict.","securityPrivacyImplications":"Authors note de-identification and access control plus separation of training and ops data as design requirements.","caveats":"Single-city Korean 119 context; early-model ~70% urgency accuracy; large unjudged share in surge; not a randomized outcome trial; older than freshness window except for why-now linkage to Nov. 5 audit.","streamIds":["emergency-services"],"roles":{"sales":"Sales framing: Seoul's published AI callbot pilot reports about a 7.33% drop in average wait (19.79s in 2024 to 18.34s in 2025) and roughly 72% urgency-classification accuracy in a surge sample, with humans retaining final judgment. For U.S. fire/EMS PSAPs drowning in non-emergency volume (Seoul cites ~62% non-dispatch calls), the bounded conversation is whether similar assistive callbots can cut queue time without claiming clinical outcome gains. Pair this with the Nov. 5 Digital Times audit: measured ops gains do not replace an independent go-live scorecard. Do not sell guaranteed 7% wait cuts or 70%+ accuracy on U.S. accents.","engineering":"Engineering framing: the architecture is assistive—PSTN/exchange feeds an AI analysis server (STT, NLP, urgency model) that posts summaries and priority hints to call-takers and CAD, rather than replacing dispatch. Replicate with on-prem or CJIS-aware hosting, learning/ops data separation, and surge logic that only offers the callbot when queue depth exceeds a threshold (Seoul used waiting-call ≥4). Validate urgency accuracy on local protocols (EMD/MPDS), not Seoul's 119 taxonomy. International transferability is limited by language, street gazetteers and fire/EMS/rescue call mix.","delivery":"Delivery framing: a measured pilot should log wait time, AI-judged vs human priority, abandon rate and immediate-transfer share under surge, matching Seoul's tables, then hold expansion until accuracy and governance gates clear. Train call-takers to treat AI as decision support, and schedule model retraining after misroutes. Why now: the same program is under Seoul council scrutiny on Nov. 5, 2026, so U.S. buyers should demand the evaluation artifacts Seoul is debating before copying the architecture."},"roleDetails":{"sales":{"customerProblem":"Large 911/119 centers need to cut non-emergency load and wait time without replacing human call-takers on true emergencies.","stakeholders":["PSAP director","Fire/EMS medical director","Call-taker union or workforce leads","City IT/security","Elected public-safety oversight"],"discoveryQuestions":["What share of calls are non-dispatch, and what is current average speed of answer?","Will AI only engage above a queue threshold?","What urgency ontology maps to local EMD?","What independent evaluation will mirror Seoul's later audit questions?"],"valueHypothesis":"An assistive callbot that screens surplus queue traffic could modestly reduce wait time if local accuracy and governance match or beat Seoul's reported pilot—value is operational, not clinical outcome proven here.","potentialEngagement":"Pilot design and measurement plan for a regional PSAP comparing pre/post wait time and AI-vs-human priority agreement on a surge sample.","unsupportedClaims":["Guaranteed 7.33% wait reduction in any U.S. center","Clinical outcome improvement or lives saved","That >70% urgency accuracy is adequate for unsupervised answer"]},"engineering":{"fit":"Fits assistive STT/NLP urgency hinting integrated to existing call-taking and CAD screens; not a standalone AI call-answerer.","architectureIntegration":"Exchange → AI analysis server (STT/NLP/urgency) → call-taker UI + CAD notes; human issues dispatch. Separate training and production data stores.","prerequisites":["Labeled local call audio with priority ground truth","Low-latency STT for local accents and place names","CAD/API hooks for summary fields"],"constraints":["Seoul metrics from one center and one surge window","Callbot only entered when ≥4 waiting calls","Korean language and 119 multi-hazard mix"],"security":"Paper cites de-identification and access control; U.S. deployments need CJIS/HIPAA-aligned handling of medical 911 content.","proofOfValue":["Replicate wait-time A/B with queue-threshold gating","Agreement rate AI vs certified EMD on held-out calls","Surge drill measuring judged vs unjudged (abandon/transfer) share"]},"delivery":{"work":"Implement gated assistive pilot, instrument metrics from Seoul tables, and couple go-live to an external or council-visible evaluation.","dependencies":["Telephony vendor","CAD vendor","QA/training unit","Privacy counsel"],"ownership":"PSAP operations owns acceptance; medical director owns clinical protocol mapping; IT owns hosting.","skillsAdoption":"Call-takers practice using AI summaries without over-trust; QA scores AI-influenced calls separately.","governanceCheckpoints":["Pre-pilot protocol approval","Mid-pilot accuracy review","Pre-expansion independent evaluation"],"acceptanceCriteria":["Documented wait-time change with confidence bounds","Urgency agreement threshold met on sample","Human override and transfer paths tested under surge"],"risks":["Over-trust of ~70% early accuracy","Mis-localization of street names","Political conflict if ops metrics diverge from later audits"]}},"retrievedAt":"2026-11-06T04:02:10Z","enrichedAt":"2026-11-06T04:05:00Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Workload relief for call-takers is claimed; equity across accents/disabilities is not measured—must be tested before U.S. adoption.","procurementImplications":"Buyers should contract for metric transparency (wait, accuracy, unjudged share) matching these tables, plus independent evaluation rights.","operatingModelImplications":"Keep humans as final decision-makers; use AI for surge pre-screen and documentation assist.","peerReview":"peer-reviewed","sourceVerification":{"openedUrl":"https://cdn.apub.kr/journalsite/sites/kosdi/2026-022-01/N0620220131/N0620220131.pdf","referenceExcerpt":"the introduction of the AI-based system reduced the average call waiting time by approximately 7.33%.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}