One consistent talent bar and rubric across all analytics
Shared metrics and methodologies prevent duplicate work
Team culture and brand that attracts top data talent
De facto embedded in pods but report centrally
Growth paths across functions, not trapped in silos
Why embedded failsLoses consistency of talent, metrics become fragmented, people get stuck with nowhere to grow, you rebuild the same models on six teams.
The Philosophy
Analytics as a Seat at the Table
Own outcomes, not just queries
Bring POV and ideas, not just answers
Answer "So what?" not just "Why?"
Balance exploratory work with reactive asks
The DoorDash hackathon
Deep dive on referral revealed fraud hiding in the data. Bimodal distribution: great customers vs. code-sharing discount-chasers. Recommendations: better fraud checks, caps on referrals.
The lesson
Data averages can be incredibly misleading. Distribution matters. Understanding why requires stepping outside the data.
"Analytics is a business impact driving function and not purely a service function. Not just answering the why, but answering the, 'What do we do now that we know this?'"
Cultural DNA
Extreme Ownership: The DoorDash Way
Boston launch, 2014: Whole team (4 people) waking at 5 AM to hand out promo codes, handing out cupcakes, even the sales guy who got commission—everyone just wanted to win
Outage night: Entire 20-person company jumps on support, answers phones, processes refunds. Jessica goes dashing to preserve courier capacity
WeDash program: 4x/year, every employee goes out dashing or does support. Build empathy with consumers, dashers, and merchants; catch bugs; live the product
Today: Expect this same ownership from every team member—if a data scientist needs to call customers to understand context, that's the job
Example: affordability feature launch
Feature shipped but didn't work as expected. Data showed segments, but why was the question. Team made phone calls. Data scientists included. That's extreme ownership—go find the truth, not just the numbers.
The hire
Look for people who blur boundaries. Product work, ops work, engineering work—people who do what's needed, not what's in their title.
Metrics
The Short-Term Leading Indicator
Retention is a terrible KPI—impossible to move meaningfully short-term
Find the input that drives the long-term output
Test whether short-term metrics actually predict the outcome you care about
Keep metrics simple—intuitive, not composite coefficients nobody understands
A 0.1 increase in a composite score means nothing; make it defensible
"Ultimately, you want to find a short-term metric you can measure that drives a long-term output. Retention is almost impossible to drive in a meaningful way in the short term."
The metric playbookSimple > Perfect. If people understand it, have intuition around it, and talk about it across the company, it'll drive better outcomes than a mathematically pure metric nobody gets.
Contrarian
Data Team Myths
✗Embed analytics in each functionINSTEAD →✓ Central org with pods. You get consistency, growth paths, shared methodologies, and team culture—without losing proximity.
✗Data teams are service functionsINSTEAD →✓ Data teams are business partners. Own outcomes. Bring ideas. Answer "so what?" not just "why?"
✗Stay in your lane (data scientist = data only)INSTEAD →✓ Extreme ownership means calling customers, doing PM work, understanding qualitative context. Role boundaries are not helpful.
✗Optimize retention as your primary KPIINSTEAD →✓ Find the input that drives retention (short-term, testable). Retention itself is a lagging indicator that's impossible to move intentionally.