Dhruv’s approach
FRAMING Three sides here, not two. The consumer who wants food fast and cheap. The Dasher who will only log on if the trip is worth their time. The merchant who wants the order to go out. Delivery cost is what DoorDash pays the Dasher, and it is the lever that balances the other two. So the tension is structural. Pay Dashers less and supply dries up, deliveries run late and consumers churn. Pay more and margin per order disappears. Delivery cost is not a fixed cost, it is the price of keeping supply on the road. One week is a short window. Erosion does not move a number 50% in seven days. Something switched, so I am looking for a date, not a trend.
- VALIDATE THE NUMBER
- Mean or median? A 50% jump in an average can come from a handful of extreme trips. If the median is flat, this is outliers, not a cost shift.
- Numerator or denominator? Cost per order rises if costs rose, or if orders fell and roughly fixed incentive spend spread over fewer of them. Check both raw series before assuming.
- Did the definition change? A cost component newly folded in, such as insurance, benefits or support allocation.
- Accounting timing. A backfill, a delayed invoice batch or a correction posting in one week.
- Market mix and currency, if this is a global figure.
- Plot it daily. Does it step on one day, or ramp across the week? A step names the date to investigate.
-
DECOMPOSE THE COST Cost per order = pay per trip / orders per trip + waste. Pay per trip = base + time + distance + incentives such as peak pay and guarantees. Orders per trip = batching. Two orders on one trip halves the cost of that trip. Waste = cancellations, redeliveries and refunds, where we pay twice for one order. A 50% move is large enough that it is probably one lever, not all three drifting together. Three levers, three different problems. Find which one moved before anything else.
-
SEGMENT
- City or market. A national average can be one or two cities dragging everything.
- Hour and day. Cost spikes usually live in specific shifts, not across the clock.
- Vertical. Grocery and convenience trips are heavier and longer than restaurant ones.
- Distance band and trip duration.
- Batched against single trips.
- New against tenured Dashers, since onboarding incentives are expensive.
- Weather-affected markets, since storms cut Dasher supply and force incentives up.
- HYPOTHESES, ranked by likelihood x impact / cost to check H1 Outliers or a definition change. Confirm: median flat while the mean jumps, or a new component in the cost build-up. Kill: the whole distribution shifted. H2 Orders fell, costs did not. Confirm: order count down, total cost flat. Kill: total cost rose in step. H3 Supply shortfall met with peak pay. Highest prior among real causes. Confirm: Dashers online per order down, incentive spend up, concentrated in specific markets and hours, acceptance rates down. Kill: incentive spend flat. H4 Batching collapsed. A dispatch or algorithm change cut orders per trip. Confirm: orders per trip down, pay per trip flat. Kill: batching unchanged. H5 Distance or vertical mix shifted. Confirm: median distance and duration up, concentrated in grocery or an expanded radius. Kill: distance flat. H6 A minimum pay rule took effect. Confirm: the rise isolates to one or two cities and lines up with an effective date. Kill: the rise is national. H7 Waste. Confirm: cancellation and redelivery rates up, so we pay twice. Kill: those rates flat.
WHAT I WOULD HAND OVER An attribution, not one cause: how much of the 50% is measurement, how much is supply incentives, how much is batching or mix. H1 and H2 run first because both are cheap and either can close this in a day. Counter-metric if anyone moves to fix it: delivery times and Dasher acceptance rate. Cutting pay lowers cost per order immediately and buys it with late deliveries, which cost more later.