Smart Cities Mobility Data: What Agencies Actually Need

Smart-cities mobility procurement has matured past "buy a feed and explore." The 2026 brief is specific: census-block-group OD flows, dwell anchored to land-use POI, equity overlays against ACS and environmental-justice layers, and consent-scoped privacy that survives public-records review. Smart cities and urban planning solutions frame GSDSI products; this piece is what agencies actually operationalize. Pair with FTC location enforcement for buyers and what privacy-safe means for location.

Origin-Destination Flows Are the Foundation

Most procured product: OD at block-group grain. Transit planners pressure-test route realignments; MPOs validate travel-demand models; active-transportation teams find mode-shift opportunities. Minimum spec: block-group origin and destination, time-of-day distribution, k-anonymity floor (typically ≥10 devices per published cell). Global Mobility publishes OD with documented floors. Triangulate employment commuting with Census LEHD. LEHD for work trips, mobility OD for non-work.

Procurement should reject tract-only feeds when block-group is required: you cannot disaggregate downward.

Dwell Patterns Anchored to a Land-Use POI Layer

Dwell without POI join is an engineering project for the agency. POI & Geofencing should return "dwell at Park X, Library Y" matching GIS tables. Parks-and-rec tests activation programs; public health reads civic-center usage; public works rationalizes hours against observed demand. USDOT BTS frames national context; agency work layers mobility on local GIS.

Equity Overlays Are a Procurement Requirement

Capital plans and service changes need equity screens: mobility intersected with American Community Survey and state environmental-justice layers. Delivery must support overlay at block-group with consistent suppression so sparse equity cohorts do not vanish. Document join methodology for auditable equity outputs. Private-sector analog: CRE origin-destination for site selection.

Privacy Scoping Is the Binding Constraint

Public-records laws make agency-held documentation disclosable. Vendors must ship consent provenance and aggregation enforcement so agencies publish analysis, not raw traces. FTC 2024 location orders set vendor expectations agencies now encode in RFPs. Sensitive-place exclusions must be pipeline-enforced, not contract-only.

The Procurement Brief That Actually Works

Briefs that survive legal, IT, and capital review specify:

• Block-group OD, k≥10, documented suppression below floor.
• Dwell joinable to agency land-use GIS without custom geocoding.
• Equity-overlay structure for ACS and EJ layers.
• Consent chain documentable in public-records responses.
• Parquet/CSV on secure transfer. API-only still blocks many agencies.

Score vendors with RFP scoring matrix governance pillar weighted ≥25% for public-sector buys.

A Mobility Panel Is Not the Population, and Planning Decisions Assume It Is

Agency work carries a burden commercial analysis does not: decisions allocate public money and affect residents who never opted into anything. That makes the gap between panel and population a governance question rather than a technical footnote. Smartphone-derived panels under-represent the very groups transport planning most needs to see, including older residents, lower-income households, and anyone whose device or data plan limits background reporting.

Handle it explicitly rather than through weighting alone. State which populations are under-observed and by roughly how much, weight against an independent source such as census or survey data, and publish the weighting method alongside the results so it can be challenged. Where a corridor or neighborhood is too thinly observed to support a conclusion, say so rather than reporting a weak estimate at the same confidence as a strong one. An agency that documents its blind spots can defend its analysis; one that presents a panel as the population cannot.

Procure for Reproducibility, Because Someone Will Ask

Public decisions get revisited, in consultation, in appeal, and sometimes years later when a scheme is evaluated. That makes reproducibility a procurement requirement rather than a nicety. If the analysis behind a decision cannot be re-run and re-explained after staff have changed and the vendor has updated its methodology, the agency has bought a slide rather than evidence.

Require versioned deliveries with a documented methodology per version, the right to retain the underlying aggregates for as long as the decision may be questioned, and clarity on what happens to that data when the contract ends. Ask specifically whether a re-run next year against the same period will return the same numbers, since a vendor that continuously reprocesses history will not, and knowing that in advance is very different from discovering it during a challenge.

Validate Against Counts the Agency Already Owns

Most agencies already hold ground truth and rarely use it to test a new data source: permanent count stations, ticketing and fare records, parking occupancy, and periodic manual counts. These are imperfect and they are independent, which is exactly what a validation needs. Comparing a mobility panel against them turns an unfamiliar dataset into one with a known relationship to something the agency already trusts.

Run the comparison at several granularities, because a panel often tracks well at corridor level and poorly at a single intersection, and knowing where that boundary sits defines the level at which the data may be used. Track the ratio between panel and count over time rather than a one-off correlation, since a stable ratio supports confident use even when the panel under-counts, while a drifting ratio signals a change in the panel that would otherwise be read as a change in travel behavior.

Validating the Panel Against Counts the Agency Already Owns

Before a mobility read reaches a capital plan, calibrate it against ground truth the agency already holds. Permanent count stations, transit boarding counts from AVL and APC systems, bike and pedestrian counters, parking transactions, and permit records all sit inside the building. Pick corridors where you hold counts and compare shape before level, meaning the time-of-day curve and the day-of-week pattern. Level agreement can be manufactured with an expansion factor. Shape agreement cannot, which makes shape the honest first test.

Then establish who is doing the expanding. A vendor publishing expanded volumes should document the factor by geography and by period, because panel penetration is not uniform across a city. One citywide factor imports bias into precisely the block groups the equity screen exists to protect, and it does so invisibly, since the output still arrives looking like counts. Where penetration is thin, prefer indices and shares over absolute volumes, and say so in the methodology note rather than in a footnote nobody reads.

Before-and-after studies run longer than a procurement cycle. A bus lane, a road diet, or a new station is evaluated across years, and a vendor that improves its methodology mid-contract hands you two series that cannot be compared. Negotiate notice of methodology changes and the right to a restated back series computed under the current method, then re-run the "before" window after any change. Without that, part of the effect you present is the vendor's release notes.

Write the validation memo alongside the finding, not after someone challenges it. Agency analysis is disclosable, it gets read by advocates and by the council, and a stated limitations section is what separates a finding that survives a hearing from one that gets withdrawn. Name the corridors used for calibration, name the ones where the panel disagreed, and say why you still trust the read. Keep the polygon versions the analysis used as well, since POI accuracy drift will otherwise redefine "visits to the library" between two reporting years without anyone deciding to.