Operate at the Lowest Level of Detail
Averages lie; the distribution is where trust is won or lost.
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 5
- Confidence
- 80%
DoorDash's core operating philosophy rejects average-based success metrics in favor of understanding the full distribution of customer outcomes, and it requires leaders to personally dig into hyper-specific, seemingly unscalable operational problems rather than optimizing dashboards from a distance. The mechanism is simple: an average delivery time can look great while a meaningful slice of customers have a terrible experience, and in categories like food delivery a single bad experience (cold food, a 53-minute wait) can permanently churn a customer. By going on-site, negotiating special infrastructure (a dedicated elevator, a dedicated restaurant serving station, dedicated parking), and setting thresholds based on actual customer tolerance rather than a race to zero, a company can deliver consistent quality even in its hardest, most complex locations.
Origin
Illustrated by Tony Xu's account of DoorDash's Cheesecake Factory delivery problem in San Francisco's Union Square, where the restaurant sits on the sixth floor of a mall with no dedicated parking.
Core principles
- 01Averages in operations are meaningless; the distribution of outcomes is what customers actually experience.
- 02Solve the hardest 1% of cases individually rather than only optimizing the median.
- 03Go do the job yourself (drive deliveries, work the line) before designing the system.
- 04Set service thresholds based on real customer tolerance, not an arbitrary race to be fastest.
How to run it
- 1
Reject the average
Stop reporting and celebrating average performance metrics (e.g., average delivery time) and instead track the full distribution, especially the worst-case tail.
Pro tip Ask 'what percent of customers had an experience worse than X?' instead of 'what was our average?'
- 2
Find the hardest venues
Identify the specific locations, restaurants, or customer segments where fulfillment is structurally hardest (e.g., a 6th-floor mall restaurant).
- 3
Go do the work yourself
Have leaders or early employees perform the frontline job (e.g., deliver pizza for Domino's) to learn the real operational constraints.
Pro tip This surfaces problems that no dashboard will show you.
- 4
Negotiate bespoke infrastructure
Where needed, negotiate special access or accommodations with venues (dedicated elevator, serving station, parking spots) to remove friction at the worst nodes.
Watch out This doesn't scale automatically — it requires manual, location-by-location effort.
- 5
Set thresholds from real behavior, not vanity
Use data on when customers actually stop caring (e.g., DoorDash found limited marginal benefit below a 42-minute ETA) to avoid over-investing in marginal speed gains.
In the wild
DoorDash needed to reliably deliver from a Cheesecake Factory on the 6th floor of a San Francisco mall with no dedicated parking, to customers 20 minutes away.
→ DoorDash negotiated a dedicated elevator shaft, a dedicated serving station at the restaurant, and dedicated Dasher parking spots in the garage to make the delivery reliable.
Rather than racing Uber Eats on raw speed, DoorDash analyzed conversion and retention data across delivery times.
→ They found no meaningful benefit to customer retention once deliveries were under 42 minutes, letting them redirect resources instead of over-optimizing speed.
Common mistakes
Chasing average speed over worst-case reliability
Racing to shave minutes off an already-acceptable average (as Uber Eats did) can waste resources versus eliminating the worst outlier experiences.
Treating hard venues as unsolvable
Writing off structurally difficult locations instead of solving them individually leaves permanent gaps in service quality.
Is it for you?
Best for
Operationally complex, multi-sided marketplace or logistics businesses where a single bad experience can churn a customer permanently.
Not ideal for
Low-stakes, high-frequency digital products where one bad session has little consequence.
From the transcript
“one of Doordash's uh core values, which usually core values are a bunch of baloney, but in this case, uh, I think actually makes sense…”
“Averages in our industry are meaningless. It's the distribution that matters. No consumers care if our average delivery time is fifth is 35 minutes if…”
“our analysis showed that there was a limited marginal benefit to customer conversion or retention rates under 42 minute ETAs.”
From the episode
DoorDash