Problem
A ridesharing marketplace has dozens of operational metrics: cancellations, ETA, fulfillment, driver utilization, earnings, incentive spend. Each answers a different version of "do we have enough drivers?", the metrics are noisy and have different units, and a dashboard of them does not say whether to change pricing, dispatch or incentives. The result is mixed messages and no rule for choosing a lever.
Example from the article: "Enough drivers" can mean enough in the right places for good service, enough system-wide if any pickup distance is accepted, or enough to keep driver earnings attractive.
Solution in brief
Measure how far the driver distribution is from the rider-request distribution, using an asymmetric optimal transport problem that moves only drivers. Its solution gives two signed indices (a demand-centric one and a supply-centric one) on the same scale. Their position on a 2-D plot says whether to add volume or realign supply with demand.
Questions this case answers
- How to measure the balance of a two-sided market over space and time with one unit-free index that says which lever to pull
- How to separate a market-wide shortage or surplus of one side from a misplacement of that side
- How to tell whether a two-sided marketplace has too few drivers, too many drivers, or drivers in the wrong places?
- How to replace a dashboard of noisy rider, driver and platform metrics with one diagnostic that points to a lever?
Tools and methods
- SD-GEM (supply-demand graph-based equilibrium metrics) — two-sided indices computed from the transport solution.
- Asymmetric optimal transport (linear program) — moves driver supply between cells at a cost to match request demand.
- Dual values (marginal values) — per-cell value of one more driver or request, used as a map of where to act.
- H3, geohash or S2 cells — spatial grid for counting riders and drivers.
Google or a one-time email code. No password.