{"resourceId":"us-city-ai-procurement-interviews-2025","versions":[{"version":"external-8ee78cdad8ab9df6e215e75d7ab873875b3ac20f3974a9e1048b92daab3f12c9","resource":{"id":"us-city-ai-procurement-interviews-2025","title":"Municipal interview study finds oversight gaps across purchasing routes and vendor relationships","organization":"Nari Johnson and colleagues; Carnegie Mellon University and collaborators","sector":"Municipal technology procurement","geography":"Seven U.S. cities; sample weighted toward larger cities","publishedAt":"Inspected arXiv version 2 dated February 7, 2025","publicationDate":"2025-02-07","eventDate":null,"sourceName":"arXiv","sourceLabel":"Academic interview study, version 2; no reliance on placeholder DOI","sourceUrl":"https://arxiv.org/html/2411.04994v2","evidenceClass":"academic-research","outcomeClass":"cautionary","topics":["governance-procurement","data-security","operating-model","accessibility-workforce"],"finding":"Legacy purchasing routes shape AI oversight, while limited vendor cooperation and unclear evaluation responsibilities constrain municipal control.","sledRelevance":"Historical empirical context newly added to the archive. Directly relevant to city purchasing, but the study's 'small' category is under 200,000 residents and does not establish rural-town capacity.","evidence":"Researchers interviewed 19 employees across seven cities during December 2023–June 2024. They used thematic coding and document analysis. The study describes oversight gaps for low-cost and cooperative purchases and difficulties obtaining supplier evidence; it is not a causal outcome evaluation.","architectureImplications":"Interpretation: Inventory AI added through existing SaaS contracts as well as new purchases; map disablement, export and dependency boundaries.","governanceImplications":"Interpretation: Apply risk-based review to all acquisition routes and define who can reject an unsupported proposal.","securityPrivacyImplications":"Interpretation: Verify supplier data use, access and opt-out behavior against evidence for the contracted configuration.","caveats":"Snowball recruitment, city-employee perspective and larger-city bias limit generalization. Anonymity limits external verification. No vendor/resident interviews, quantified service gains or tested procurement reform effect. Findings describe an earlier policy and product environment.","streamIds":["local-government"],"roles":{"sales":"Interpretation: A CIO, purchasing lead, service owner and privacy officer may have incomplete visibility into AI arriving through different contracts. Ask which purchases bypass specialist review, who can request supplier evidence, and who pays for evaluation after launch. Offer a bounded inventory and responsibility assessment spanning a new purchase, a cooperative agreement and an existing SaaS service. The hypothesis is that clearer ownership exposes actionable gaps before scaling. Do not generalize seven cities into a national prevalence estimate or imply every vendor refuses cooperation. For a small locality, first establish available staff and shared-service options; the sample is not representative of resource-constrained rural governments.","engineering":"Interpretation: Fit is an assurance workflow across products rather than a single AI platform. Map feature discovery, contract boundaries, data paths and operational dependencies into a maintained register. Obtain version-specific evaluation evidence and test access where available; record evidence the supplier cannot provide. Validate an embedded feature's disablement and data export before treating an opt-out promise as operationally usable. A proposed proof of value should trace one application from acquisition route through approval, monitoring and exit, testing permissions and retention along the way. Evaluate cloud, on-premises and hybrid choices against local support capacity; the study supplies no performance comparison among those deployment models.","delivery":"Interpretation: Procurement and IT should jointly inventory acquisition routes, while each service owner accepts ongoing outcome review. Assign an evidence owner for supplier claims and a staff owner for incident response. Dependencies include contractual access, technical evaluation skills and enough protected time to run checks. Train teams to distinguish missing evidence from demonstrated failure, and include resident representatives where consequences justify it. Proposed acceptance: every sampled service has an owner, a review route regardless of price, a documented evidence gap decision, and an exercised escalation or exit path. No observed improvement is implied. Risks include adding review forms without decision authority or creating monitoring duties that nobody is funded to perform."},"retrievedAt":"2026-09-08T03:02:22Z","enrichedAt":"2026-09-08T03:05:44Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: Budget staff time to interpret evaluations, and add resident and accessibility perspectives absent from the interview sample.","procurementImplications":"Interpretation: Ask for test access, change notice and service-exit demonstrations before commitment where feasible.","operatingModelImplications":"Interpretation: Explicitly assign post-deployment evaluation and incident duties between service owner, IT and supplier.","updateExplanation":"Absent from the full archive, including alternate URL checks by arXiv identifier. Added as historical research, not newly published or updated evidence.","sourceVerification":{"openedUrl":"https://arxiv.org/html/2411.04994v2","referenceExcerpt":"Our reliance on snowball sampling skewed our sample of participating cities towards large cities","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}