Demand-responsive transport research shifts from operator cost to rider experience, review finds

Demand-responsive transport, the family of bus-like services that abandon fixed routes and timetables in favour of vehicles dispatched on request, has accumulated two decades of research without a single, quality-appraised map of how these systems are actually designed and judged. A new systematic review published in European Transport Research Review closes that gap. Only papers scoring at least nine of a possible twelve points survived, and two independent reviewers applied the criteria, reaching an inter-rater agreement of Cohen’s kappa of 0.92. The procedure followed the PRISMA 2020 reporting guidelines, and the full record-level decisions were released as supplementary material.

The review is organised around five research questions that together span the life cycle of an on-demand service: the criteria and methods used for design and evaluation, the preferences and barriers that shape social acceptance, the pricing strategies and service features on offer, the routing and dispatch algorithms that allocate vehicles, and the criteria used to choose vehicle types and fleet sizes. The authors deliberately treat demand-responsive transport as a flexible complement to conventional fixed-route public transport rather than a rival, reserving the term public transport for the bus, tram, metro and rail backbone of the network. Their scope is urban and low-density peri-urban settings, where most modelling evidence is concentrated; rural and regional studies were kept when they met the quality bar, including pilots in Amsterdam and Lolland, but purely rural systems remain under-represented in the corpus, a limitation the authors discuss explicitly.

The clearest pattern in the evidence is a shift in what designers optimise. Early formulations of the dial-a-ride problem were operator-oriented, minimising fleet cost and distance travelled. The retained literature instead converges on multi-objective frameworks that balance operator cost against user service quality, explicitly trading waiting time, route deviation and reliability against the expense of running vehicles. The authors argue that a one-dimensional, operator-only view fails in the real world, because riders face uncertain waits and long detours that a cost-minimising objective never sees. Evaluation metrics have moved in parallel: user-oriented indicators such as waiting time, in-vehicle time and total travel time are now reported as means and percentiles to capture adverse experiences, while satisfaction, acceptance and predictability of arrival times reflect a turn toward service stability rather than average performance alone.

Methodologically, the field splits along a scale line. At small problem sizes, mixed integer linear programming dominates, offering exact and explainable solutions to routing, scheduling and allocation formulations. As instances grow toward real operational scale, the problems become computationally intractable, the review notes, because dynamic dispatching is NP-hard. There, the literature turns to metaheuristic algorithms, rolling-horizon control and agent-based simulation to manage the complexity, often combined with data-driven demand forecasts. Exact mathematical programming still plays a structural role in longer-horizon design and planning questions such as network design and integration with fixed-route services, but in real-time operational control, exact formulations frequently yield infeasible solutions within the available computation time. Machine learning appears mainly as a supporting tool for estimating demand or model inputs rather than as the primary solving method.

On the question of what makes these systems succeed, the strongest recurring finding concerns positioning. Studies consistently show that demand-responsive services work best in a trunk-and-feeder structure, in which high-capacity metro, bus or bus rapid transit lines form the backbone and on-demand vehicles feed passengers to and from them, extending spatial and temporal accessibility. Where services are designed to compete directly with buses and metro, the reviewed evidence warns of reduced public transport demand, increased vehicle-kilometres travelled and lower social welfare. Integration goes beyond geography: payment systems, docking arrangements, coordinated scheduling, integrated ticketing and connection to mobility-as-a-service platforms recur as design requirements. The authors also identify a characteristic user base, including elderly riders, people with reduced mobility, residents of poorer neighbourhoods, tourists, night-time workers and off-peak commuters, for whom door-to-door access, fewer transfers and predictable timing matter more than minimising travel time.

The barriers to implementation fall into three clusters. Technical and operational constraints are the most frequent, led by the scarcity of high-quality route, network and demand data for calibration, a need the review calls a necessary condition for effective real-time control, alongside insufficient digital and physical infrastructure, dependence on failure-prone technology, missing kerbside space and, in electric and autonomous scenarios, inadequate charging infrastructure. Social barriers centre on acceptance: users must tolerate detours, delays, walking to pick-up points, transfers, ride sharing and sitting next to strangers, and low acceptance of pooling or advance reservation appears repeatedly as a deterrent. Policy and institutional barriers include rigid or absent regulatory frameworks, the risk of weakening fixed-route lines, and unresolved questions of price governance, fairness and how much delay is acceptable. The authors consolidate these into an enabler-barrier structure that lets a design or policy choice be checked against both the conditions it requires and the obstacles it faces.

Pricing emerges as more than a revenue line. In several models, fare enters directly as a component of a generalised cost or utility function, shaping route choice and seat allocation alongside travel time and deviation, which lets the operator use price to regulate demand flow and fleet utilisation, though the approach depends on rich behavioural data and careful calibration. Discount rules reward walking, accepting delays or sharing, with some schemes scaling the discount to detour length and occupancy; dynamic discounts outperform fixed ones but demand transparent implementation to avoid erratic fluctuations. Dynamic and surge pricing transmit scarcity signals during peaks, while fixed, public-transport-aligned fares trade flexibility for simplicity and comparability with existing tariffs. Willingness-to-pay analysis, subscription passes, tariff integration with fixed-route networks and attention to distributional fairness complete the taxonomy, and the review finds pricing is rarely fully integrated with operational control models.

On routing, the literature contrasts real-time dynamic insertion, which answers requests immediately without redesigning the network, against batch-based or pre-booking dispatch, which packages requests at intervals and generally achieves better route quality and fleet efficiency at the cost of temporal flexibility. Dynamic insertion is sensitive to the order in which requests arrive and degrades under congestion or heterogeneous demand, so the review argues it must be paired with congestion-aware, shortest-time routing fed by accurate real-time data. A more surprising gap is the walk to the vehicle. Most reviewed studies model the in-ride experience in detail while treating the pre-ride walk to the pick-up point as fixed or exogenous, even though in virtual-stop and semi-flexible designs, stop placement and users’ willingness to walk jointly determine vehicle positioning and achievable route quality. Making that pre-ride spatial behaviour endogenous is flagged as a tractable and consequential extension.

The review’s conclusions temper the enthusiasm that has surrounded on-demand transport since celebrated successes such as Mokumflex in the Netherlands and MOIA in Germany, and cautionary failures such as Kutsuplus in Finland, GoSutton in the United Kingdom and Bridj in the United States. Financial viability remains fragile without subsidy, since higher cost per passenger is the systems’ structural weakness, and studies reporting cost per passenger rarely couple it with the equity objectives that motivate many deployments, leaving the efficiency-equity trade-off under-examined. The authors call for responsive demand models that capture how service quality changes ridership, for modelling that anticipates autonomous and electric fleets with their charging-station and empty-repositioning problems, and for dynamic, fair pricing frameworks that cover costs without excluding low-income users. Targeted, zone-based deployment, they conclude, outperforms citywide many-to-many coverage.

Subject of Research: Systematic review of methods for designing and evaluating demand-responsive transport systems

Article Title: Designing and evaluating demand responsive transport systems: a systematic review

Article References: Razavian, S. B., Rodríguez, A., Cordera, R., & dell’Olio, L. (2026). Designing and evaluating demand responsive transport systems: a systematic review. European Transport Research Review, 18(1), Article 97. https://doi.org/10.1186/s12544-026-00858-6

Image Credits: AI Generated

DOI: 10.1186/s12544-026-00858-6

Keywords: demand-responsive transport, systematic review, public transport, routing algorithms, dynamic pricing, fleet sizing, agent-based simulation, social acceptance, mobility-as-a-service, urban mobility, NP-hard optimisation, transport equity