{"resourceId":"bexley-ai-webchat-first-month-86-6-percent-scrutiny-20261014","versions":[{"version":"external-eb372ed7dab0b98b8c548920da730c7d73c38b8fe764ef7b09393d55dc66899a","resource":{"id":"bexley-ai-webchat-first-month-86-6-percent-scrutiny-20261014","title":"London Borough of Bexley reports its AI webchat gave an exact answer or the correct page for 86.6% of first-month questions, while scrutiny members urge caution on public-facing AI","organization":"London Borough of Bexley","sector":"Council customer contact and digital services","geography":"Bexley, London, United Kingdom","publishedAt":"Public reports pack for 14 October 2026 committee (reported by London Local 15 October 2026)","publicationDate":"2026-10-14","eventDate":"2026-10-14","sourceName":"Bexley Finance and Corporate Services Overview and Scrutiny Committee","sourceLabel":"Council committee report pack","sourceUrl":"https://democracy.bexley.gov.uk/documents/g30966/Public%20reports%20pack%2014th-Oct-2026%2019.30%20Finance%20Corporate%20Services%20Overview%20and%20Scrutiny%20Committ.pdf?T=10","evidenceClass":"government-evaluation","outcomeClass":"mixed","topics":["knowledge-work","operating-model","governance-procurement"],"finding":"The Future Bexley Corporate Core report to the 14 October scrutiny committee says the council's AI-driven webchat went live on 1 July and that in its first month 86.6% of questions resulted in an exact, complete answer or the person being directed to the correct page. It also reports a new Copilot AI agent that summarises key points in Stage 1 or 2 complaints under the new complaints code. Both sit within a move from 8x8 telephony to a Microsoft Teams-based contact centre with a delivery partner. In the same pack, member feedback from a July assurance workshop shows strong appetite for internal AI such as Magic Notes but caution on public-facing webchat, citing concern about residents getting trapped in automated loops and a need to review existing AI policies.","sledRelevance":"One of few council chatbots reporting a defined first-month answer-quality measure, paired with elected members' explicit risk split between internal and resident-facing AI, which mirrors debates in U.S. cities such as Denver.","evidence":"Full 1.2 MB public reports pack PDF opened via curl and pdftotext; the Corporate Core report section and member workshop feedback were read. London Local (15 Oct) used to locate it. The measurement method behind 86.6% is not described.","architectureImplications":"Webchat and AI agents are bundled into a Teams-based contact centre migration, so AI capability depends on the telephony platform change and delivery partner.","governanceImplications":"Scrutiny members want an AI policy review before further public-facing AI; a DDaT Board sets priorities and is centralising application management.","securityPrivacyImplications":"The complaint-summarisation agent handles personal complaint data; the report does not describe controls or DPIA status.","caveats":"Single-month, self-reported metric with no stated sample size, method or independent check; the success definition includes redirecting to a page, which is weaker than resolution. No later-month data in the pack.","streamIds":["local-government"],"roles":{"sales":"Sales framing: Bexley gives a concrete benchmark, 86.6% first-month answer-or-redirect, and a clear political boundary between internal and public AI. Lead discovery on how the metric is measured and what members need to see before expanding webchat. A bounded engagement is an independent answer-quality and escalation audit of the webchat. The evidence cannot support claims of call deflection or savings, which the report does not quantify. Interpretation: assurance on resident experience is the gating need.","engineering":"Engineering framing: Fit is a knowledge-base webchat embedded in a Teams-based contact centre with handoff to advisers. Integration needs logging of each answer with outcome labels so the success rate can be audited. Prerequisites are a curated content set and the telephony migration. Constraints include complaint-code rules for the summarisation agent and UK GDPR for complaint data. Proof of value should sample transcripts, separate exact answers from redirects and test loop-exit paths. Interpretation: measure resolution, not just redirection.","delivery":"Delivery framing: Next steps are phase two transactions, continued monitoring and the AI policy review members asked for. Ownership sits with the Customer Experience and Corporate Core programme under Finance and Corporate Services. Advisers need clear handoff procedures. Acceptance criteria should report monthly answer quality with method and loop-exit rates. Risks include residents trapped in automated loops and reliance on a single self-reported metric. Interpretation: publish the measurement method with each update."},"roleDetails":{"sales":{"customerProblem":"Council needs to show resident-facing AI is reliable enough to satisfy cautious scrutiny members.","stakeholders":["Senior Responsible Officer, Customer Experience and Corporate Core","Director of Finance and Corporate Services","Overview and Scrutiny Committee members","DDaT Board","Customer service advisers"],"discoveryQuestions":["How is the 86.6% measured and sampled?","What share of interactions hand off to an adviser?","What would members need before phase two transactions?","How is the complaint agent assured?"],"valueHypothesis":"Audited answer quality could unlock expansion of webchat; savings are not reported.","potentialEngagement":"Independent transcript-sample audit of webchat answer quality and escalation paths.","unsupportedClaims":["Call volume reduction","Cost savings","That 86.6% persists beyond month one"]},"engineering":{"fit":"Council webchat on curated content within a Teams-based contact centre.","architectureIntegration":"Webchat with adviser handoff in Teams contact centre; Copilot agent for complaint summaries.","prerequisites":["Telephony migration to Teams","Curated knowledge content","Interaction logging with outcome labels"],"constraints":["Complaint-code compliance for summaries","UK GDPR for complaint data","Members' caution on public AI"],"security":"Complaint data handled by an AI agent; controls not described in the source.","proofOfValue":["Transcript sample separating exact answers from redirects","Loop-exit and handoff success test","Accuracy check of complaint summaries against originals"]},"delivery":{"work":"Monitor webchat, build phase-two transactions, review AI policy and assure the complaint agent.","dependencies":["Telephony delivery partner","Content owners","Complaints team","Scrutiny committee"],"ownership":"Customer Experience and Corporate Core programme, Finance and Corporate Services.","skillsAdoption":"Advisers and complaint handlers need procedures for AI handoff and summary checking.","governanceCheckpoints":["AI policy review requested by members","Phase-two go-live approval","Regular scrutiny updates"],"acceptanceCriteria":["Proposed: monthly answer-quality metric with published method","Proposed: loop-exit to adviser available in every flow","Proposed: sampled complaint summaries verified"],"risks":["Residents trapped in automated loops","Over-reliance on one self-reported metric","Dependency on telephony migration"]}},"retrievedAt":"2026-10-17T22:16:05-05:00","enrichedAt":"2026-10-17T22:40:00-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Members flagged resident frustration with automated loops; the report positions AI tools as support for staff complaint handling.","procurementImplications":"AI capability arrives through a telephony platform change and appointed delivery partner rather than a standalone AI procurement.","operatingModelImplications":"Internal AI gets a low-control, high-flexibility stance from members while public-facing AI gets higher control, implying tiered governance.","sourceVerification":{"openedUrl":"https://democracy.bexley.gov.uk/documents/g30966/Public%20reports%20pack%2014th-Oct-2026%2019.30%20Finance%20Corporate%20Services%20Overview%20and%20Scrutiny%20Committ.pdf?T=10","referenceExcerpt":"In the first month, 86.6% of questions resulted in the exact, complete answer or the individual being directed to the correct page.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}