DoorDash Marketplace Segmentation and Growth Loops

What's being tested

Interviewers are probing your ability to reason about a two-sided platform: how to slice the marketplace into meaningful segments, choose metrics that reveal health per segment, and design repeatable growth loops that scale supply and demand together. They want to see that you can translate qualitative hypotheses (e.g., “dense neighborhoods drive frequency”) into measurable experiments, prioritize tradeoffs using unit economics, and communicate a clear launch/measure/iterate plan. At DoorDash, this shows you can move beyond one-size-fits-all growth to durable, profitable growth across heterogeneous cities, restaurant types, and customer behaviors.

Core knowledge

Marketplace segmentation — common axes: geography (zip/neighborhood), customer lifecycle (new, active, dormant), order frequency, average order value (AOV), cuisine vertical, time-of-day, and supply characteristics (single-vs-multi-location restaurants).
Unit economics — know contribution margin per order = Price − Cost (delivery + variable platform cost); compute payback period = CAC / contribution_margin_per_period and basic LTV: $LTV \approx \sum_{t=0}^T \frac{m_t}{(1+r)^t}$ where m_t is contribution margin at t and r is discount.
Core metrics — track GMV, AOV, DAU/MAU, orders per active user, fulfillment_rate, wait_time, driver_utilization, and restaurant_activation_rate; measure them per segment, not only globally.
Growth loop primitives — acquisition → activation → frequency/retention → referral/virality; quantify conversion rates between stages and loop velocity (time between loops).
Network effects & density — thickness matters: below a density threshold, wait times and fulfillment degrade; model minimal orders-per-hour per sq-mile required to sustain target p99 wait-times.
Segmentation vs personalization tradeoff — coarse segments (city-level) are simpler to measure; micro-segmentation (user-persona + time) has higher signal but increases experiment weeds and operational complexity.
Experimentation & heterogeneity — run stratified A/B tests by segment; check interaction effects and power per stratum (smaller segments need larger samples or longer test windows).
Prioritization frameworks — use RICE (Reach, Impact, Confidence, Effort) while folding in payback_period and marginal margin; prioritize interventions that improve loop velocity or reduce friction in the tightest bottleneck.
Loop amplification & K-factor — estimate amplification as product of step conversion rates; prioritize points with high elasticities (small input → large loop output).
Supply-side levers — incentives, onboarding speed, scheduling flexibility, and capacity forecasting; understand operational constraints like driver shift patterns and restaurant prep windows.
Measurement unit & guardrails — pick the unit (order, customer-week, city-day) that best captures the loop; use holdout markets and Bonferroni-like adjustment when testing many segments to avoid false positives.
Failure modes — subsidies that boost short-term frequency but harm long-term margins; segmentation that creates product fragmentation or unacceptable maintenance cost.

Worked example — "Increase order frequency in mid-sized cities"

Frame: ask clarifying Qs in first 30 seconds—what is a mid-sized city definition, current orders_per_active_user, target timeline, and margin constraints. Organize answer into three pillars: (1) diagnose — segment users (by frequency, geography, AOV) and identify bottleneck (supply density, long ETA, limited cuisine selection); (2) design loops — if supply is the bottleneck, run a restaurant-onboarding loop + targeted promotions to drive initial demand; if demand is the bottleneck, focus on retention and habit loops (subscription, scheduled ordering, personalized promotions); (3) execute & measure — define primary metric (orders_per_active_user per city-week), secondary metrics (fulfillment_rate, contribution_margin), and an experiment plan (stratified A/B with city holdouts). Explicit tradeoff to flag: short-term promotions increase frequency but will raise CAC and may lower LTV/CAC unless retention improves; propose a capped promotion with a retention requirement (e.g., unlock discount only after 3 orders). Close by saying: if more time, build a simulation model to forecast payback under scenarios and run quick pilot in 2 representative cities before full rollout.

A second angle — "Grow restaurant onboarding in suburban markets to improve selection"

Same segmentation and loop thinking flips to the supply side: first define which restaurant types are missing (fast-casual vs family), quantify incremental GMV per onboarded partner using nearby demand maps, and build a restaurant growth loop: targeted outreach → fast technical onboarding → first-order guarantees → marketing that routes early demand. Measure restaurant activation rate, 7-day retention (restaurant still accepting orders), and effect on local order frequency. Prioritize onboarding for high-elasticity cuisines (where marginal increase in selection yields large lift in orders) and beware over-saturating a low-demand area which increases restaurant cancellations and driver idle time.

Common pitfalls

Pitfall: Averaging over the whole marketplace — Reporting a single national DAU or orders_per_user hides pockets of failure; always show segment-level metrics and the distribution (median, 10th/90th percentile).

Many candidates propose one global lever (e.g., more promotions) without diagnosing the bottleneck per segment. Interviewers expect you to map symptoms to specific loop frictions: acquisition, activation, frequency, or matching. The better answer ties specific experiments to the identified bottleneck.

Pitfall: Ignoring operational constraints — Suggesting to double demand without modeling driver capacity or restaurant prep windows will break fulfillment and worsen retention. Always call out operational limits and mitigation (surge, throttling, queued onboarding).

Pitfall: Optimizing vanity metrics — Proposing to maximize app_installs without connecting to orders or unit economics is a red flag; tie every growth lever back to contribution margin and sustainable LTV/CAC.

Connections

This area frequently pivots into marketplace economics (pricing, commissions, surge), experimentation design (powering stratified tests and sequential rollout), and customer segmentation analytics (cohorts and churn modeling) — be prepared to go deeper on any of these quickly.