{"resourceId":"maricopa-mari-assistant-usage-20260825","versions":[{"version":"external-31a27cf589bb822cfa1f70740238d723e290e430392b750cb0e20c0f66718869","resource":{"id":"maricopa-mari-assistant-usage-20260825","title":"Maricopa County’s Mari AI assistant logs 12,500+ early questions with taxes and pet adoption leading topics","organization":"Maricopa County, Arizona (via Route Fifty)","sector":"County resident services / digital CX","geography":"Maricopa County, Arizona, USA","publishedAt":"August 25, 2026","publicationDate":"2026-08-25","eventDate":null,"sourceName":"Route Fifty","sourceLabel":"Independent reporting / interview","sourceUrl":"https://www.route-fifty.com/artificial-intelligence/2026/08/early-findings-maricopa-countys-new-ai-assistant-reveal-resident-priorities/415629/","evidenceClass":"independent-reporting","outcomeClass":"emerging","topics":["operating-model","knowledge-work","accessibility-workforce","data-security"],"finding":"Route Fifty’s Aug. 25, 2026 interview with Maricopa County innovation leaders reports that “Mari,” a 24/7 AI assistant launched in June on the county site, had fielded more than 12,500 questions between June 8 and late August. Leading topics: property taxes/tax payments (~14%), pet adoption/availability (~8%), and county jobs (~3%). Mari pulls public data from 50+ departments (IBM stack) with Esri geocoding for location-aware answers; Animal Care and Control is treated as a flagship use to surface adoptable pets. Officials frame early analytics as an engagement signal, not yet a measured deflection or adoption outcome.","sledRelevance":"Why-now after Sept. 26 Wyandotte Treasury call-volume narrative: a large U.S. county publishing early topic-mix analytics for a resident AI assistant—showing what residents ask before claiming FTE savings—transferable to 311/CX pilots that need public usage dashboards.","evidence":"On-record quotes from Aaron Judy (chief of innovation and emerging technologies) and Kacie Baker (geographic information officer). Counts and topic shares are operator-reported preliminary data without published methodology or independent audit.","architectureImplications":"Public-web chatbot grounded on department public data; IBM AI stack plus Esri geocoding for spatial answers; multi-department content ownership.","governanceImplications":"Publish early usage topic mix and keep humans for complex cases; treat analytics as service-design input (e.g., animal welfare) not proof of autonomous resolution.","securityPrivacyImplications":"Limit corpus to publicly available department data; govern location queries and pet/personal preference logs under county privacy/retention rules.","caveats":"August publication predates this edition’s date; included for why-now linkage to ongoing county CX-agent scale decisions. No accuracy rate, apology rate, or call-deflection KPI published. Impact on shelter capacity described as uncertain.","streamIds":["local-government"],"roles":{"sales":"County CX and 311 leaders need resident AI that surfaces what people actually ask—not vanity launch metrics. Route Fifty’s Aug. 25 report on Maricopa County’s Mari assistant cites more than 12,500 early questions, with property taxes (~14%) and pet adoption (~8%) leading topics after a June launch. Frame discovery on topic mix, department content owners, and which questions stay human-only. Bound value to published usage analytics and service-design insights—not guaranteed call-center FTE cuts from early volume alone.","engineering":"Fit is a public-web, 24/7 county assistant grounded on public department content (IBM stack) with Esri geocoding for location-aware answers—not case adjudication. Prerequisites: multi-department public corpus, analytics on topic and failure modes, and privacy rules for chat/location logs. Constraints: preliminary metrics only; animal-welfare outcomes still uncertain; content drift across 50+ departments. Prove value by pairing topic dashboards with accuracy samples and escalation rates before expanding channels.","delivery":"Delivery work is operate-and-learn: keep Mari’s topic dashboard current, assign department content owners (starting with Animal Care and Control and Treasury-adjacent tax FAQs), and decide channel expansion only after accuracy review. Dependencies: IBM/Esri platform teams, MCACC SMEs, and privacy/records. Ownership: innovation/emerging tech with departments owning answers. Governance checkpoints: public-data-only grounding, stop criteria if accuracy slips, and no claims of shelter capacity wins until measured. Risks: celebrating 12,500 questions as success without quality metrics."},"roleDetails":{"sales":{"customerProblem":"Residents struggle to find county program info (taxes, pets, jobs) across fragmented department pages, creating after-hours and web dead-ends.","stakeholders":["Chief of innovation","GIS/data officer","Animal Care and Control","Treasury/tax","County board"],"discoveryQuestions":["Which topics dominate the first 90 days of assistant traffic?","Which departments own answer accuracy?","What accuracy/escalation KPIs accompany volume?","How are location and preference logs retained?"],"valueHypothesis":"Publishing early topic-mix analytics from a grounded county assistant can redirect content and engagement (e.g., pet adoption) before claiming staffing savings.","potentialEngagement":"Six-week CX-assistant analytics readiness: topic taxonomy, content-owner map, accuracy sample plan, and public dashboard draft.","unsupportedClaims":["Guaranteed call-center FTE reduction","That Mari improves shelter live-release rates","Autonomous resolution of tax or court disputes"]},"engineering":{"fit":"Strong for FAQ navigation and location-aware public info; weak for adjudicated cases or private data workflows.","architectureIntegration":"Website chatbot over public department corpora; IBM AI services; Esri geocoding; analytics warehouse for topic shares.","prerequisites":["Public corpus inventory","Department content owners","Privacy/retention for chats","Escalation path to humans"],"constraints":["50+ department content drift","Preliminary unaudited metrics","No published accuracy bar"],"security":"Public-data grounding only; minimize PII in prompts/logs; access logging; no training on resident private data without agreement.","proofOfValue":["Topic-mix stability vs baseline web search","Accuracy sample of answers","Escalation/handoff rate","Repeat-question rate"]},"delivery":{"work":"Run weekly topic and failure-mode reviews; refresh high-traffic FAQs; pilot pet-adoption deep links with MCACC.","dependencies":["IBM/Esri vendors","Department SMEs","Privacy/records counsel"],"ownership":"Innovation owns platform; departments own factual accuracy.","skillsAdoption":"Staff learn to interpret topic dashboards and correct wrong answers without over-trusting volume.","governanceCheckpoints":["Public-data-only attestation","Accuracy sample before new channels","Privacy review of location logs"],"acceptanceCriteria":["Published 90-day topic dashboard","Named owners for top-5 topics","Documented escalation SLA","No unresolved critical accuracy defects in sample"],"risks":["Volume vanity metrics","Content drift","Overclaiming animal-welfare impact"]}},"retrievedAt":"2026-09-27T22:21:13-05:00","enrichedAt":"2026-09-27T22:21:13-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"24/7 web access may help after-hours users; accessibility of the chat UI and non-web alternatives are not evaluated in the article.","procurementImplications":"Enterprise AI plus GIS stack (IBM/Esri) implies multi-vendor integration clauses, public-data use limits, and analytics deliverables in SOWs.","operatingModelImplications":"Shifts some FAQ load to AI while using topic analytics to steer department content and outreach (e.g., animal welfare) rather than only deflecting tickets.","sourceVerification":{"openedUrl":"https://www.route-fifty.com/artificial-intelligence/2026/08/early-findings-maricopa-countys-new-ai-assistant-reveal-resident-priorities/415629/","referenceExcerpt":"nearly 8% of more than 12,500 questions submitted to Mari were about pet adoption","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}