🔬 physics

Demand-agnostic assessment of on-demand pooled transit services

This paper introduces a demand-agnostic assessment framework that uses simulations and performance thresholds to identify optimal urban areas and hubs for deploying pooled on-demand transit services even when precise demand data is unavailable, as demonstrated in a case study in Krakow.

Original authors: Olha Shulika, Hanna Vasiutina, Michał Bujak, Farnoud Ghasemi, Rafał Kucharski

Published 2026-06-17

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Olha Shulika, Hanna Vasiutina, Michał Bujak, Farnoud Ghasemi, Rafał Kucharski

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a city planner trying to launch a new type of bus service. This isn't a regular bus that runs on a fixed schedule; it's a "smart bus" that only shows up when people ask for it, picks up neighbors along the way, and drives them to the nearest train or tram station. It's like a digital carpool that connects your home to the main transit network.

The big problem? You don't know who will actually use it yet. You can't ask people, "Will you ride this?" because the service doesn't exist yet. If you pick the wrong neighborhood, you might waste money on a bus that runs empty. If you pick the right one, you save the city money and help people get around easier.

This paper presents a crystal ball for city planners. It's a method to figure out where to launch this service before you even buy a single bus, even when you have zero data on how many people will sign up.

The "Seed Planting" Analogy

Think of the city as a garden with 12 different patches of soil (neighborhoods). You want to plant a special seed (the on-demand bus service) in the patch where it will grow the best.

Usually, farmers wait until they see the seeds sprout to know if the soil is good. But here, the researchers say, "Let's pretend we plant the seeds and see how the soil would react."

They use a concept called the "Demand Fraction." Imagine this as a dial you can turn.

Low setting (0.1%): Only 1 out of every 1,000 residents is interested in the service.
High setting (5%): 1 out of every 20 residents is interested.

The researchers run a computer simulation for each of the 12 neighborhoods. They turn the dial up slowly, asking: "At what point does this service start working well?"

The Three "Health Checks" (KPIs)

To decide if a neighborhood is a good candidate, the simulation checks three vital signs, similar to checking if a new restaurant is successful:

The "Gas Savings" Check (Mileage Reduction): Does sharing the ride actually save fuel? If everyone rides alone, it's wasteful. If they share, does the total distance driven drop by at least 10%?
The "Comfort" Check (Passenger Satisfaction): Is the ride still pleasant? If the bus has to take a huge detour to pick up a neighbor, the passenger might get annoyed. The simulation checks if the ride is still at least 2.5% "better" (or at least not worse) than taking a private car.
The "Crowded Bus" Check (Occupancy): Is the bus actually carrying more than one person? If the average number of people in the vehicle is less than 2, it's just a private taxi in disguise. The bus needs to carry at least 2 people on average to count as a "shared" success.

The Race to the Finish Line

The researchers ran this simulation for 12 different areas in Krakow, Poland. They asked: "Which neighborhood is the 'champion'?"

The champion isn't necessarily the one with the most people. It's the one that meets all three "Health Checks" with the lowest number of interested users.

The Winner: Area 9 (near the Mydlniki train station). This area was the "super soil." It met all the success criteria even when only a tiny fraction of residents (less than 1%) were interested. It was the most efficient place to start.
The Losers: Area 10 and Area 4. These areas were like rocky soil. Even when they turned the dial up to 5% interest, the service still didn't meet the efficiency or comfort goals. In some cases, these areas were just too close to existing train stations, so a feeder bus wasn't needed at all.

The "Growth Map"

Once they found the winner (Area 9), they created a roadmap for the future. They told the city:

The Spark: If you get just 0.025% of residents to use it, the first shared rides will happen.
The Growth Spurt: At 0.05%, the service starts to really work well.
The Steady State: Once you hit 0.1% to 0.7%, the service is running smoothly and efficiently.

The Bottom Line

This paper gives cities a decision-making toolkit. Instead of guessing or waiting for real-world data that might never come, they can use this method to say: "Don't put the bus in Neighborhood A; put it in Neighborhood B. It will start working with fewer people and save more money."

It's a way to take the guesswork out of launching new, shared transportation, ensuring that when the service finally arrives, it has the best possible chance of success right from day one.

Technical Summary: Demand-Agnostic Assessment of On-Demand Pooled Transit Services

Problem Statement
Municipalities seeking to reduce car dependency and improve urban liveability often face a critical barrier in deploying on-demand pooled transit feeder services: the absence of reliable demand data during the planning stage. Traditional service optimization relies on observed trip rates, modal splits, or route-level ridership, which are unavailable for entirely new services. This lack of data complicates the selection of optimal deployment areas and hub locations, particularly in low-density urban regions where demand is uncertain. Without precise demand figures, municipalities struggle to determine where a service will be operationally viable or how to set realistic performance expectations before launch.

Methodology
The authors propose a parameter-free, demand-agnostic assessment framework designed to evaluate the potential of pooled on-demand transit feeders without requiring prior calibration of demand-specific parameters. The methodology proceeds through the following steps:

Input Data: The approach utilizes OpenStreetMap (OSM) road networks, residential address point distributions (representing population), and candidate area-hub pairs provided by municipal planners.
Demand Generation: Instead of assuming fixed demand, the method simulates potential demand using a "fraction of demand" ( $\alpha$ ). This represents the probability that a resident will use the service. Demand is generated randomly across residential points within candidate areas, with travelers assumed to head toward the nearest public transport hub. The simulation tests $\alpha$ values ranging from 0.1% to 5% of the local population.
Matching Algorithm (ExMAS): The study employs the Exact Matching of Attractive Shared Rides (ExMAS) algorithm. This utility-based, offline algorithm matches travelers into pooled rides only if the shared trip offers higher utility than a solo ride-hailing trip, considering travel time, detours, waiting time, and discomfort penalties.
- Baseline: The comparison is made against a scenario where all potential users take individual ride-hailing trips.
- Assumptions: The feeder service is modeled as free at the point of use (no fare premium for pooling), meaning utility gains are driven solely by reduced travel costs (if applicable) or the trade-off between fare discounts and discomfort. The algorithm assumes a vehicle capacity of six passengers.
Performance Indicators (KPIs): Three key metrics are evaluated across varying demand fractions:
- Vehicle Hours Reduction ( $\Delta T_v$ ): The reduction in total vehicle hours compared to solo travel.
- Passenger Utility Gain ( $\Delta U_p$ ): The improvement in perceived traveler utility (comfort/convenience) relative to solo travel.
- Average Occupancy ( $O$ ): The average number of passengers per vehicle.
Threshold-Based Selection: The core innovation is the identification of the minimum demand fraction ( $\alpha^*$ ) required to meet specific operational thresholds:
- $\Delta T_v \geq 0.1$ (10% reduction in vehicle hours).
- $\Delta U_p \geq 0.025$ (2.5% improvement in utility).
- $O \geq 2$ (Average occupancy exceeds 2).
  Areas and hubs that meet these thresholds at the lowest demand fraction are ranked as the most promising candidates for deployment.
Benchmarking: For the top-ranked area-hub combination, the study establishes benchmarks for the "low-demand period," identifying the demand levels at which pooling emerges, potential grows, and consistent KPI achievement is reached.

Key Contributions

Pre-Deployment Evaluation Tool: The paper introduces a structured method for municipalities to select optimal areas for new on-demand feeder services in the absence of exact demand data.
Probabilistic Demand Simulation: By utilizing demand fractions rather than fixed trip matrices, the method allows for the assessment of service viability across a spectrum of potential adoption rates.
Operational Thresholds: The study defines specific, conservative thresholds for vehicle hours, utility, and occupancy that distinguish a viable pooled system from a collection of solo trips.
Hub Selection Logic: The methodology includes a mechanism to select the optimal transit hub for a given area by comparing KPI performance across multiple potential destinations.

Results (Krakow Case Study)
The methodology was applied to 12 pre-selected low-density areas in Krakow, Poland, paired with nearby public transport hubs (tram and train stations).

Ranking: Areas were ranked based on the minimum demand fraction required to meet all three KPI thresholds.
Top Candidate: Area 9, paired with the Kraków Mydlniki (PKP) train hub, was identified as the most promising candidate. It achieved all thresholds at the lowest demand fractions (e.g., $\alpha = 0.001$ for vehicle hours reduction, $\alpha = 0.005$ for utility, $\alpha = 0.007$ for occupancy).
Least Promising: Area 10 (high density but very close to a hub) and Area 4 ranked lowest, with Area 10 failing to meet occupancy and utility thresholds even at a 5% demand fraction. This suggests that proximity to existing high-frequency hubs reduces the need for feeder services.
Benchmarks for Area 9: For the top-ranked area, the study established that pooled rides emerge at a demand fraction of 0.025%, significant growth occurs at 0.05%, and stable operational capacity (meeting all thresholds) is reached between 0.1% and 0.7% demand.

Significance and Claims
The paper claims that this approach provides municipalities with a structured, data-driven tool to make pre-deployment decisions regarding on-demand pooled transit. By identifying areas that require the lowest level of resident interest to achieve operational efficiency, planners can prioritize locations with the highest probability of success. The authors emphasize that the identified minimum demand fractions represent operational viability thresholds, not predictions of actual ridership. They note that in contexts where adoption is low, additional measures (e.g., integration with ticketing, subsidies) may be required to reach these thresholds in practice. The study positions itself as a solution to the "chicken-and-egg" problem of launching new services without historical data, offering a way to assess potential based on spatial and demographic characteristics alone.

Limitations
The authors acknowledge that the ExMAS algorithm models an upper bound on pooling potential by not explicitly managing fleet constraints. The study assumes a single predetermined hub for all travelers and does not account for complex mode choices (walking, cycling) or the full journey (feeder + main transit). Future work is suggested to incorporate observed mode choice data, explicit fleet modeling, and broader journey assessments.