DoorDash Three-Sided Marketplace Segmentation

What's being tested

Interviewers probe your ability to translate business goals into an actionable segmentation strategy for a three-sided marketplace (consumers, merchants, Dashers). They want to see product judgment: which segments matter, how to measure and prioritize impact, and how you’d target experiments or investments for measurable ROI. Expect clarifying questions, tradeoffs between short-term revenue vs. long-term supply health, and experiment/metric design to validate hypotheses.

Core knowledge

Know the three actors: consumer (demand), merchant (supply listing/capacity), and Dasher (fulfillment); segmentation must consider cross-side interactions and feedback loops, not siloed metrics.
Define business-focused metrics per side: `GMV`, `take rate`, `conversion`, `order frequency` for consumers; `acceptance rate`, `average ticket`, `menu completeness` for merchants; `utilization`, `earnings per hour`, `on-time rate` for Dashers.
Use orthogonal axes: value (LTV, `GMV` contribution), behavior (frequency, recency), cost-to-serve (distance, customized handling), and sensitivity (price/fee elasticity). Combine axes to form actionable cells.
RFM-style segmentation: Recency, Frequency, Monetary for consumers; for merchants/Dasher adapt RFM to `order_volume`, `fulfillment_latency`, and `downtime`. RFM scales to millions; use deciles/percentiles to keep cells interpretable.
Statistical segmentation methods: cohort analysis, decision-tree rules, and clustering (e.g., `k-means`, hierarchical) for exploratory grouping — but convert clusters into rule-based segments for productization and experimentation.
Experimentation per segment: stratified A/B tests with pre-specified primary metric and segment-level power calculations; use uplift vs. absolute effect when resources are limited. Compute sample size with standard formulas $n = \frac{(z_{1-\alpha/2}+z_{1-\beta})^2 ( \sigma_1^2 + \sigma_2^2)}{\Delta^2}$ .
Causal and attribution awareness: when segmenting for retention or supply operations, control for selection bias (e.g., high-frequency users differ systematically) and plan for randomized targeting or quasi-experimental methods.
Prioritization framework: estimate impact = (segment size) × (expected lift) × (value per unit); rank by ROI and strategic importance (e.g., supply-constrained geos get higher weight).
Operationalize segments: convert analytic buckets into rule-based definitions (e.g., "top 10% `GMV` consumers in SF" or "restaurants with <10% acceptance rate and >30 min prep"): necessary for targeting, dashboards, and guardrails.
Monitor leakage and health: instrument `p95` delivery latency, cancellation rate, and `DAU` churn per segment, and set alert thresholds tied to business SLAs so interventions remain timely and safe.

Worked example — "Design a segmentation strategy for DoorDash's three-sided marketplace"

First 30 seconds: clarify the objective (growth, margin, retention, or supply health), geography scope, time horizon, and available signals (order history, ETA logs, acceptance rates). Then state assumptions: e.g., we aim to increase sustainable orders by 10% over 6 months without raising take rate.

Organize the answer around three pillars: (1) define business-priority segments using value × cost-to-serve axes; (2) choose targeting levers and experiments per segment; (3) measurement plan and operationalization. For pillar (1) propose concrete segments: top-10% `GMV` users (high value), infrequent but high-lift users (reactivation candidates), supply-constrained restaurants (high demand, low acceptance), and low-utilization Dashers (target for incentives).

Flag an explicit tradeoff: targeting high-`GMV` users yields larger immediate revenue but smaller elastic lift; focusing on reactivation or supply fixes may have higher % lift but smaller absolute impact. For experiments, define primary metric (incremental weekly orders per user), stratify randomization by segment, and run power calculations to ensure detectability.

Close with next steps: if given more time, run exploratory clustering for non-obvious segments, prototype rule-based targeting in one city, and build a segment-dashboard to iterate.

A second angle — "Which segments should we prioritize to reduce ETA and improve on-time delivery?"

Here the objective shifts to operational KPIs and supply balance. Map segments by geographic density and `acceptance_rate` (merchant) and `idle_time` (Dasher). Prioritize interventions that unblock high-volume corridors: e.g., restaurants with high order queue but low prep staffing, or zip codes where Dashers under-serve at peak. Interventions differ: merchant-side may need prep-time SLAs and scheduling changes; Dasher-side may need routing incentives or dynamic pay. Measurement must include both direct ETA reduction and downstream effects: cancellations, merchant SLA compliance, and marginal cost per second of ETA improvement. Emphasize short experiments (time-windowed surge pricing or targeted scheduling) and careful spillover checks (improving one zone might worsen adjacent zones).

Common pitfalls

Pitfall: optimizing for segment lift without accounting for size.
Mistake: choosing a tiny segment with 100% lift but negligible absolute impact. Better: always multiply expected lift by segment size and per-unit value to compute realistic ROI.

Pitfall: using opaque clustering as the final product.
Mistake: presenting `k-means` clusters without rule-based definitions. Convert clusters into human-interpretable rules before proposing product changes to ensure implementation and experiment targeting.

Pitfall: ignoring cross-side effects and supply constraints.
Mistake: boosting consumer discounts in a supply-starved area increases cancellations and costs. Always model supply elasticity and include guardrails (caps, controls) in experiments.

Connections

Interviewers may pivot to adjacent topics like pricing/incentives (dynamic promos, surge), marketplace equilibrium (matching algorithms and capacity planning), or experimentation infrastructure (stratified randomization and guardrails). Be prepared to tie segmentation choices to these operational levers.