Why Transit Network Redesigns Fail Without Good Demand Data
A full transit network redesign — restructuring routes, adjusting frequency, realigning service patterns — is one of the highest-stakes analytical projects a transit agency undertakes. The decisions made during a redesign typically remain in place for 10–15 years before being revisited. They affect where workers can commute, where residents can access healthcare and education, and how equitably the system distributes mobility across income and demographic groups. Getting the analysis right matters enormously.
Most network redesigns fail not because planners make bad decisions, but because they make reasonable decisions based on incomplete information. The classic failure mode: a redesign planned primarily from travel demand model outputs and systemwide ridership statistics that misses spatially concentrated demand in lower-density corridors, underestimates the impact of proposed changes on transit-dependent populations, or optimizes for current peak patterns that have shifted since the modeling data was collected.
The four-phase methodology described here reflects how Mobvynt approaches network redesign analysis — not as a proprietary process, but as a structured approach to ensuring that the demand surface data actually drives the design decisions rather than simply validating them after the fact.
Phase 1: Data Ingestion and Demand Surface Construction
The first phase is technical and, done right, largely invisible to the planning process. Its output is a clean, validated demand surface that the rest of the analysis depends on.
The data inputs for a typical agency-scale redesign analysis:
- GTFS static schedule feed: Current network — routes, stops, service patterns. This is the reference geometry for the analysis. Quality check: validate that all route shapes are present, stop coordinates are accurate to within a reasonable tolerance, and service calendar entries are current.
- GTFS-RT historical archive: 12–24 months of archived VehiclePositions and TripUpdates feeds, from which we reconstruct actual service patterns (not scheduled service) and derive stop-level delay accumulation. Quality check: coverage rate — what percentage of scheduled trips have matching GTFS-RT records? Archives below 70% coverage require supplemental analysis.
- APC data: Boarding and alighting counts by stop, by trip, for the analysis period. This is the primary ridership signal. Quality check: validate APC sensor coverage across the fleet; identify trips where APC records are missing or flagged as low-confidence.
- GBFS bike-share data: Archived GBFS station status and, where available, trip records. Quality check: data freshness and coverage — are all stations in the service area represented? Are there gaps in the archive?
- ACS demographic data: 5-year estimates at the block group level, for Title VI and equity analysis in Phase 3. Quality check: confirm that the ACS vintage is current (within 2 years of the analysis period) and that block group coverage aligns with the study area.
Once ingested and validated, these inputs are processed into an H3 resolution 8 demand surface: a spatial grid where each cell carries time-series attributes representing ridership demand, service frequency, demographic context, and multimodal accessibility. The demand surface is the analytical canvas for the rest of the redesign process.
Phase 2: Existing Conditions Analysis and Gap Identification
Phase 2 answers three specific questions about the current network:
Where is demand concentrated relative to current service? This is the core spatial analysis — comparing demand density (boardings per H3 cell per time window) against service supply (routes, headways, stop coverage per cell). The cells and corridors where demand density is high and service supply is low are the primary candidates for service expansion. The cells where service supply is high and demand density is low are the candidates for service reduction or restructuring.
What does the OD (origin-destination) structure of demand look like? Stop-level boarding and alighting data, combined with GTFS trip sequences, allows reconstruction of implied OD matrices — which stops are generating outbound trips that arrive at which destinations. This is less precise than a traditional household travel survey OD matrix, but it's derived from actual observed ridership rather than a survey conducted at a point in time. The OD picture reveals whether current routes are aligned with actual trip patterns or whether riders are making complex transfers to reach destinations that could be served more directly.
Where are the first/last mile gaps? Using the GBFS analysis methodology described in our first/last mile post: identify the H3 cells with high transit demand but poor pedestrian access scores and limited micro-mobility connectivity. These are the candidates for first/last mile intervention — either route alignment changes, new stop placement, or infrastructure investment to improve transit access.
The Phase 2 output is a set of spatial findings — specific corridors and H3 cell clusters — with documented analytical support. This is the foundation for the design alternatives in Phase 3.
Phase 3: Service Design Alternatives and Scenario Testing
Phase 3 is where the planning process engages most directly with the demand surface. The process involves generating alternative network scenarios — proposed route configurations, headway schedules, stop placement decisions — and evaluating each scenario against the demand surface.
The specific evaluation metrics for each scenario:
- Network coverage: What percentage of the population falls within a defined walkshed of a route stop? What percentage of the population in low-income and minority census areas falls within coverage? How do these numbers compare to the current network?
- Service frequency distribution: What is the weighted-average headway experienced by the served population? How is that headway distributed across demographic groups? Does the proposed network allocate frequency improvements equitably?
- Projected ridership impact: Using ridership elasticity estimates derived from the demand surface and empirical ridership-frequency relationships, what is the projected ridership change under each scenario? (This is an estimate with uncertainty, not a precise forecast, and should be presented as a range.)
- Title VI disparate impact test: As described in our Title VI article: overlay demographic data with service change impacts and compute the distributional metrics required by the agency's Title VI program.
For a mid-size agency redesign, we typically run 3–5 alternative scenarios plus a no-build baseline. The scenario comparison is the analytical core of the public engagement process — it gives community members and elected officials a data-supported picture of the trade-offs between different design choices.
Phase 4: GTFS Scenario Export and Implementation Handoff
The final phase converts the preferred alternative into operational outputs that the agency can actually use. For a transit agency, the primary output is a GTFS feed for the proposed network — route shapes, stop locations, service calendars, trip sequences, and headway schedules in the standard GTFS format that the agency's CAD/AVL system, trip planning applications, and NTD reporting pipeline can consume.
A few implementation handoff considerations that are specific to demand-surface-based redesigns:
Monitoring plan: The demand surface analysis establishes clear baseline metrics for the current network. The implementation plan should specify how those metrics will be tracked post-redesign — which APC data queries, which ridership elasticity indicators, which equity metrics — and at what intervals. This turns the redesign analysis into a continuous improvement cycle rather than a one-time event.
Staged implementation support: Large network redesigns often implement changes in stages rather than all at once. The demand surface can be updated after each stage to confirm that ridership is responding as projected and to identify any unexpected patterns that warrant adjustment before the next stage.
Title VI documentation package: The FTA Title VI requirement for major service changes includes documentation that needs to be retained by the agency. The Phase 3 analysis produces this documentation as a by-product: the disparate impact analysis, the demographic overlay methodology, the scenario comparison. Packaging this into a Title VI compliance file during Phase 4 — rather than reconstructing it later — reduces the documentation burden significantly.
The four-phase process described here takes roughly 4–6 months for a typical mid-size agency redesign, from data ingestion through GTFS export. Agencies that have their GTFS and APC data well-maintained and accessible can move faster; agencies with significant data quality remediation needs in Phase 1 will take longer. The most common timeline risk is Phase 1: data quality issues that weren't apparent from the outside take time to resolve, and that time compresses the planning phases downstream.
A redesign methodology is only as useful as the demand data that feeds it. Investing in data quality before the analysis starts is almost always faster than discovering data problems after the scenarios are already drafted.