{"resourceId":"prince-william-county-va-citibot-will-first-year-civic-marketplace-case-study-20261021","versions":[{"version":"external-9cfd32be8dd492f6bc3dfc9e25d0cfbb22508c1089f7086b64a672c3318fa9af","resource":{"id":"prince-william-county-va-citibot-will-first-year-civic-marketplace-case-study-20261021","title":"Civic Marketplace case study says Prince William County's Citibot assistant handled nearly 40,000 of about 90,000 PWC311 interactions in its first year","organization":"Prince William County, Virginia (PWC311); Citibot; Civic Marketplace","sector":"County 311 resident services","geography":"Prince William County, Virginia, United States","publishedAt":"21 October 2026 (Civic Marketplace Team)","publicationDate":"2026-10-21","eventDate":null,"sourceName":"Civic Marketplace","sourceLabel":"Vendor-marketplace case study (promotes cooperative purchasing of Citibot)","sourceUrl":"https://www.civicmarketplace.com/news/how-prince-william-county-handled-nearly-40-000-resident-conversations-with-ai-in-its-first-year","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["operating-model","governance-procurement","accessibility-workforce"],"finding":"A Civic Marketplace case study published 21 October 2026 says Prince William County's Citibot assistant, Will, deployed across PWC311 phone, text and web in May 2025, handled nearly 40,000 of roughly 90,000 resident interactions in its first year. It reports more than 95% completion of nearly 7,000 service requests and about 13-second live-agent answer times, with tax administration and real estate as leading topics. The piece promotes cooperative-contract access to Citibot and cites Denver's Sunny as a model, a week after Denver's council rejected Sunny's extension over accuracy and measurement concerns.","sledRelevance":"Vendor-side volume claims are being marketed through cooperative contracts at the same moment a peer council has rejected the same vendor's assistant for weak accuracy evidence, so buyers need independent measures.","evidence":"Vendor-marketplace figures: ~90,000 interactions, ~40,000 handled by Will, >95% of ~7,000 service requests completed, ~13 s live-agent answer. No accuracy, containment, satisfaction or cost data; no pre-deployment baseline.","architectureImplications":"Assistant integrated into existing 311 phone, text and web channels rather than a new app; service request completion implies integration with the county's 311 case system, though not described.","governanceImplications":"Cooperative contracts (AFI, TXShare) let agencies buy without a standalone solicitation, which shortens procurement but also removes a point where councils typically ask for accuracy and outcome evidence.","securityPrivacyImplications":"Not addressed in the source; tax and real-estate queries may involve personal data, so data handling terms need checking.","caveats":"Marketing content. The 95% completion rate covers all channels, not Will alone. Figures differ in framing from an archived county release reporting 64,000-plus interactions; neither source explains the difference.","streamIds":["local-government"],"roles":{"sales":"Framing: volume figures are not outcome evidence. This case study says Will handled nearly 40,000 of about 90,000 interactions, and it is sold through cooperative contracts that skip a separate solicitation. Buyers will compare it with Denver, where council rejected the same vendor's bot over accuracy doubts. Sellers should lead with measurement plans covering accuracy, escalation and satisfaction, and avoid promising staff savings or replicable results from a marketing piece.","engineering":"Framing: a multichannel assistant on top of an existing 311 operation. Engineers should establish how Will creates service requests in the 311 system, what knowledge sources it draws on for tax and real-estate answers, and how handoff to live agents works. The source gives no accuracy or containment metrics, so any proof of value needs instrumentation that logs answer correctness and escalation reasons by channel.","delivery":"Framing: running a 311 assistant is an ongoing content and quality operation. Delivery would include knowledge-base upkeep for seasonal topics such as tax bills, escalation routing, monthly quality sampling and reporting to the board. Interpretation: given Denver's experience, delivery teams should produce accuracy and satisfaction evidence from the start rather than relying on interaction counts at renewal time."},"roleDetails":{"sales":{"customerProblem":"High volumes of routine 311 questions on tax and real estate compete with complex cases for staff time.","stakeholders":["County 311 director","Finance and tax administration","Real estate assessments","County data and AI engineering","Procurement (cooperative contracts)","Board of County Supervisors"],"discoveryQuestions":["How will accuracy be measured and reported to the board?","What share of AI-handled conversations end without escalation?","Which cooperative contract terms cover data use and exit?","What happened to staffing and wait times before and after launch?"],"valueHypothesis":"Bounded: an assistant may absorb a substantial share of routine contacts, but the share that is resolved correctly is unknown from this source.","potentialEngagement":"An independent evaluation plan for a 311 assistant: accuracy sampling, escalation analysis and satisfaction survey before renewal.","unsupportedClaims":["Staff time or cost savings","Answer accuracy","That results replicate in other jurisdictions","Causal link between Will and 13-second answer times"]},"engineering":{"fit":"Counties with an established multichannel 311 operation and high-volume informational queries.","architectureIntegration":"Assistant embedded in existing phone, text and web channels with service request creation in the 311 case system (integration details not given).","prerequisites":["Maintained 311 knowledge base","Service request system integration","Live-agent handoff path","Conversation logging"],"constraints":["No accuracy or containment data","Cooperative-contract terms not reviewed","Integration method undisclosed"],"security":"Tax and property queries may involve personal data; logging, retention and vendor data-use terms must be verified.","proofOfValue":["Sample 200 AI conversations for answer correctness","Trace escalation reasons by channel","Compare service request quality from AI versus agents"]},"delivery":{"work":"Operate and improve a 311 assistant across channels, with knowledge upkeep and quality reporting.","dependencies":["311 knowledge base owners","Vendor configuration","Case system integration","Board reporting cycle"],"ownership":"County 311 (PWC311) with county data and AI engineering.","skillsAdoption":"311 agents shift to complex cases; staff need skills in conversation review and knowledge-base editing.","governanceCheckpoints":["Accuracy report before renewal","Data-use review of cooperative contract terms","Board briefing on outcomes beyond volume"],"acceptanceCriteria":["Proposed: at least 90% correct answers in sampled conversations","Proposed: escalation reasons logged for every handoff","Proposed: satisfaction measured per channel quarterly"],"risks":["Inaccurate answers on tax or assessment questions","Renewal decided on volume rather than outcomes","Vendor lock-in through cooperative contract"]}},"retrievedAt":"2026-10-22T22:15:13-05:00","enrichedAt":"2026-10-22T22:20:22-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Multichannel and 24/7 access may help residents outside office hours; live agents remain for complex cases. No language-access or staffing data given.","procurementImplications":"Promoted via AFI and TXShare cooperative contracts that avoid standalone solicitations; tiered monthly SaaS pricing not disclosed.","operatingModelImplications":"AI absorbs routine 311 contacts while agents handle complex needs; requires ongoing knowledge upkeep and quality review.","sourceVerification":{"openedUrl":"https://www.civicmarketplace.com/news/how-prince-william-county-handled-nearly-40-000-resident-conversations-with-ai-in-its-first-year","referenceExcerpt":"In its first year, Will handled nearly 40,000 of the County's approximately 90,000 total resident interactions.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}