{"resourceId":"mgo-ai-control-to-value-20261002","versions":[{"version":"external-909fae8efb1aa48dfbc6b8e8ef1cb06175ea8f0200ae0162681677cc514798e1","resource":{"id":"mgo-ai-control-to-value-20261002","title":"MGO: local governments should tie AI to service outcomes, controls, human review, and an AI Decision File","organization":"MGO (Matt Ely, MBA)","sector":"Public-sector advisory / IT risk and controls","geography":"United States (state and local government audience)","publishedAt":"October 2, 2026","publicationDate":"2026-10-02","eventDate":null,"sourceName":"MGO CPA","sourceLabel":"Independent professional guidance / Control to Value framework","sourceUrl":"https://www.mgocpa.com/perspective/ai-government-controls-public-trust/","evidenceClass":"independent-research","outcomeClass":"cautionary","topics":["governance-procurement","operating-model","data-security","accessibility-workforce"],"finding":"MGO’s Oct. 2, 2026 perspective argues state and local AI value depends on starting from a specific public-service outcome, designing use-case controls (approved sources, role access, citations, human review, unauthorized-action limits), and measuring end-to-end process results—including review and rework—not draft speed alone. It warns that capacity freed is not automatic budget savings and recommends an AI Decision File covering service objective, authorized use, evidence, public value, management decision, and next review, referencing NIST AI RMF and GAO AI Accountability Framework themes.","sledRelevance":"Actionable control-and-measurement playbook for U.S. local governments deploying resident or staff AI—directly counters launch-without-metrics patterns common in municipal chatbot and permitting pilots.","evidence":"Prescriptive methodology with concrete examples (policy assistant controls; resident draft-response review checks; two-minute draft vs 15-minute verify); cites NASCIO 2026 AI priority ranking, NIST AI RMF, and GAO AI Accountability Framework. Not an empirical evaluation of a named city deployment.","architectureImplications":"Architectures should enforce approved-source limits, role-based access, citation/verify paths, and hard blocks on unauthorized actions (send, approve, alter records) beyond drafting.","governanceImplications":"Requires accountable owners, predeclared acceptance criteria, retained test evidence, and continue/change/expand/stop decisions with next-review triggers—lightweight but mandatory before scaling.","securityPrivacyImplications":"Mandate IT/security/procurement/records review of provider data use/retention before staff enter sensitive data—including new AI features on familiar platforms.","caveats":"Advisory content from a firm that markets related services; no primary municipal dataset. Examples are illustrative. Budget-savings critique is conceptual, not audited.","streamIds":["local-government"],"roles":{"sales":"When cities pitch AI for faster drafts or 24/7 answers, use MGO’s Control to Value framing: sell a bounded use case with controls and measurement, not a tool alone. Discovery should surface baselines, review-time capacity, and what the AI is forbidden to do. Value hypothesis is credible only if end-to-end time (including verification) and resident correction paths improve—reject guaranteed savings narratives the article itself flags as unsupported.","engineering":"Fit is control design around a specific municipal use (policy assistant, draft correspondence, service-request triage). Integrate approved document stores, citation requirements, RBAC, and action authorization gates separate from generation. Prerequisites: data-use review with IT/security/procurement/records; test cases for edge conditions; reviewer SOPs. Proof of value must capture review minutes and rework, not model latency.","delivery":"Delivery work is standing up an AI Decision File, baselines, control tests, and a human-review operating rhythm before expansion. Ownership needs a service owner plus control/evidence owner. Acceptance criteria: documented authorized use, passing control tests, measured public-value indicators, and a dated next review—pause path if unreliable."},"roleDetails":{"sales":{"customerProblem":"Governments adopt AI for speed but lack controls and evidence that resident service actually improved after human review.","stakeholders":["City/county manager","CIO/CISO","Finance/auditor","Department service owners","Procurement"],"discoveryQuestions":["What specific service outcome and baseline metrics exist?","Who reviews AI output and how long does review take?","What actions is the AI forbidden to take automatically?","How will continue/expand/stop decisions be evidenced?"],"valueHypothesis":"Bounded AI uses with enforced controls and end-to-end measurement can improve capacity and service quality without claiming unverified budget savings.","potentialEngagement":"Control-to-Value workshop: select one use case, draft AI Decision File, define control tests and baselines, run a four-week measured pilot.","unsupportedClaims":["Guaranteed budget savings from AI capacity","That draft speed equals process efficiency","Empirical results for a named MGO-client city in this article"]},"engineering":{"fit":"Controls-first architecture for staff or resident AI assistants that draft or retrieve but do not auto-execute consequential actions.","architectureIntegration":"Approved-source retrieval, citation/verify UX, RBAC, separate authorization service for send/approve/record-change, audit logging for Decision File evidence.","prerequisites":["Inventory of permitted tools and data classes","Service owner and reviewer roles defined","Baseline time/quality metrics","Provider data-use/retention review"],"constraints":["Review capacity must be funded or claims fail","Familiar platforms’ new AI features still need review","Higher-consequence uses need stronger approval gates"],"security":"Block sensitive data entry into unapproved tools; verify provider retention/training terms; retain enough logs to explain significant AI-supported actions under records rules.","proofOfValue":["End-to-end time including review/rework vs baseline","Correction and repeat-contact rates","Control-test pass evidence (authZ, source limits)","Documented management continue/change/stop decision"]},"delivery":{"work":"Create AI Decision File, implement controls, train reviewers, run measured pilot, report public-value evidence, schedule reassessment.","dependencies":["IT/security/procurement/records participation","Department reviewer time","Baseline data availability","NIST/GAO-aligned control checklist"],"ownership":"Service department owns outcome; IT/risk owns control design; executive sponsor owns expand/stop decision.","skillsAdoption":"Reviewers need clear checklists and authority to correct/escalate; managers must not set productivity targets that skip review.","governanceCheckpoints":["Pre-pilot Decision File approval","Control-test gate before production","Post-pilot evidence review","Triggered reassessment on policy/model/source change"],"acceptanceCriteria":["Authorized use and forbidden actions documented","Human review steps exercised on sample outputs","Measured indicators include review time","Next review date and pause criteria set"],"risks":["Treating advisory guidance as empirical proof","Skipping review under speed pressure","Claiming savings without spending evidence"]}},"retrievedAt":"2026-10-02T22:22:20-05:00","enrichedAt":"2026-10-02T22:26:53-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Public-facing AI must offer plain-language role disclosure and an accessible human path; workforce impact includes funded review time, not only generator adoption.","procurementImplications":"Evaluate vendors on control evidence, data-use terms, and measurement support—not feature demos alone; include new AI add-ons on incumbent platforms in the same review.","operatingModelImplications":"Shifts AI from project launch to operated use case with Decision File, monitoring, and explicit stop/pause authority when unreliable.","sourceVerification":{"openedUrl":"https://www.mgocpa.com/perspective/ai-government-controls-public-trust/","referenceExcerpt":"A tool that produces a draft in two minutes may not save time if an employee needs 15 minutes to verify every answer.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}