{"resourceId":"boston-project-green-light-signal-lessons","versions":[{"version":"external-4bca3c55fbdede2b9cdb3e0443e0fd4e8799b6fa09800a3fbcd3104e3ee6c8cd","resource":{"id":"boston-project-green-light-signal-lessons","title":"Giving Boston the Green Light: lessons from AI signal-timing recommendations","organization":"Eno Center for Transportation (author: former Boston Chief of Streets)","sector":"City traffic signals","geography":"Boston, Massachusetts, US","publishedAt":"August 20, 2026","publicationDate":"2026-08-20","eventDate":null,"sourceName":"Eno Center for Transportation","sourceLabel":"Eno article","sourceUrl":"https://enotrans.org/article/giving-boston-the-green-light/","evidenceClass":"independent-reporting","outcomeClass":"mixed","topics":["operating-model","infrastructure","governance-procurement"],"finding":"Boston implemented Google Project Green Light recommendations at more than a quarter of its 900 signalized intersections, averaging a 13% delay reduction at modified intersections (27% at several Back Bay intersections), but reversed changes at about a third of locations for lack of improvement or negative impacts. The AI sometimes misread phasing at complex intersections, does no corridor or network analysis, surfaced equipment failures rather than new insights at some sites, and cannot resolve pedestrian-versus-vehicle value tradeoffs.","sledRelevance":"Why now: state DOTs such as Caltrans have adaptive AI signal control in proof of concept this month; Boston's account shows what advisory, human-reviewed signal AI does and does not deliver.","evidence":"First-person practitioner account with city-reported averages; no published methodology or independent verification of the 13% figure.","architectureImplications":"Advisory-only design with no integration into signal control let Boston deploy in weeks; built-in post-change reporting provided the validation loop.","governanceImplications":"Engineers approve every change; values-based tradeoffs such as crossing-time minimums and bike signals stay with people.","securityPrivacyImplications":"Relies on anonymized phone movement data; no direct control-system integration limits cyber exposure.","caveats":"Author is a former city official; impact measurement is vendor-provided; localized optimizations may not yield network benefit.","streamIds":["transportation"],"roles":{"sales":"Boston shows advisory signal AI producing modest average gains with frequent rollbacks, a credible and bounded story for city traffic engineers. The customer problem is too few staff to retime many signals. Lead with the human-review and rollback workflow, not the 27% best case. This evidence cannot support corridor or network benefits or safety outcomes. Position help with measurement and prioritization.","engineering":"The advisory design avoided controller integration, which is why deployment took weeks. Engineering checks: verify the AI's inferred phasing against actual timing plans, especially at complex intersections; use recommendations to find detector and plan faults; and add the corridor-level analysis the tool lacks. Independent before/after measurement should complement vendor-reported impact. Closed-loop control would need much more validation.","delivery":"Delivery is an engineering workflow: triage recommendations, approve or reject, implement, review impact within weeks and roll back when needed, as Boston did at about a third of sites. Owners are city signal engineers. Budget time for equipment repairs the tool surfaces. The acceptance criteria are proposals. Document pedestrian and bike constraints explicitly so optimizations do not erode them."},"roleDetails":{"sales":{"customerProblem":"Cities cannot retime hundreds of signals often enough with existing staff and tools.","stakeholders":["City signal engineering","Transportation department leadership","Pedestrian and bike program staff"],"discoveryQuestions":["How many intersections are retimed per year today?","Who reviews and approves timing changes?","How do you measure impact after a change?"],"valueHypothesis":"Advisory AI with fast feedback could raise retiming throughput with modest average delay reductions.","potentialEngagement":"A bounded measurement plan for advisory signal recommendations on a defined set of intersections.","unsupportedClaims":["Corridor or network throughput gains","Safety improvements","A guaranteed 13% delay reduction"]},"engineering":{"fit":"Fits cities with limited integration capacity; advisory only.","architectureIntegration":"Phone movement data, inferred phasing, a recommendations dashboard, manual implementation and post-change reporting.","prerequisites":["Accurate timing plan records","Working detectors","Engineer review capacity"],"constraints":["No corridor or network analysis","Pedestrian crossing minimums"],"security":"Low control-system exposure since there is no direct integration; confirm data-sharing terms for movement data.","proofOfValue":["Independent before/after delay measurement","Rollback rate tracking","Phasing-inference accuracy at complex intersections"]},"delivery":{"work":"Run a recommend, approve, implement, measure and roll-back cycle across selected intersections.","dependencies":["Signal maintenance crews","Vendor recommendations","Timing plan documentation"],"ownership":"City signal engineering.","skillsAdoption":"Engineers learn to evaluate AI-inferred phasing and impact reports.","governanceCheckpoints":["Engineer approval per change","Post-change review within weeks","Pedestrian standard check"],"acceptanceCriteria":["Proposed: documented delay change per modified intersection","Proposed: rollback decision recorded within an agreed window"],"risks":["Localized gains shifting queues elsewhere","Phasing inference errors","Over-reliance on vendor-reported impact"]}},"retrievedAt":"2026-10-29T22:06:00-05:00","enrichedAt":"2026-10-29T22:12:23-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Pedestrian standards (at least 20 seconds for a 45-foot crosswalk) constrain optimization; engineer review time is required.","procurementImplications":"Low-integration advisory tools are faster to adopt than closed-loop adaptive systems.","operatingModelImplications":"A short feedback loop (weeks) between change and measured impact, with rollback as a normal outcome.","sourceVerification":{"openedUrl":"https://enotrans.org/article/giving-boston-the-green-light/","referenceExcerpt":"changes were reversed at about a third of locations due to lack of improvement or negative impacts.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}