A Blueprint for Data Excellence: Lessons from DoorDash's Analytics Organization

Jessica Lachs reveals the contrarian strategies that built one of tech's most respected data teams and transformed DoorDash into a data-driven powerhouse.

Key Takeaways

• Centralized data teams outperform embedded models through consistent talent bars, growth opportunities, and shared methodologies across business units
• Analytics should drive business impact, not just answer questions - teams need seats at decision-making tables alongside product and engineering
• Hire for curiosity over credentials - the best data people pull threads and investigate anomalies without being asked to do so
• Short-term proxy metrics that drive long-term outcomes work better than trying to optimize directly for retention or other lagging indicators
• Simple metrics beat composite scores - if people can't understand or intuitive grasp a metric, it won't drive meaningful behavior change
• Focus on edge cases and failure states like "never delivered" orders that cause disproportionate churn and cost despite low frequency
• Extreme ownership culture means data scientists should call customers directly when quantitative analysis hits limits rather than staying in their lane
• Cross-functional experience creates stronger data teams - importing talent from operations, marketing, and engineering brings diverse problem-solving approaches

Timeline Overview

• 00:00–04:59 — Jessica's background: Introduction to Jessica Lachs, VP of Analytics and Data Science at DoorDash, who built one of tech's largest and most respected data teams over 10 years
• 04:59–10:52 — Centralized vs. embedded analytics teams: Jessica's contrarian view on why centralized data organizations outperform embedded models despite business leaders preferring embedded teams
• 10:52–15:10 — The benefits of a centralized analytics team: Detailed explanation of advantages including consistent talent bars, growth opportunities, shared methodologies, and strong team culture
• 15:10–20:45 — Balancing proactive and reactive work: Strategies for carving out time for exploratory deep dives while handling constant incoming requests from business partners
• 20:45–24:20 — Advice on how to push back effectively: Framework for prioritizing work through shared goals and making trade-offs transparent to business partners rather than suffering in silence
• 24:20–28:57 — Hiring for curiosity and problem solving: What to look for when building data teams, focusing on curiosity and self-motivation over pure technical skills, plus interview techniques
• 28:57–34:40 — Coming from a non-traditional background: Jessica's journey from art portfolio to self-taught data science, learning SQL and Python out of necessity at early DoorDash
• 34:40–40:39 — The early days and culture at DoorDash: Stories from DoorDash's startup phase including handing out promo codes in Boston winters and company-wide customer support during outages
• 40:39–44:39 — Encouraging cross-functional roles: How extreme ownership culture leads data scientists to call customers directly and work outside traditional role boundaries
• 44:39–46:30 — Defining effective metrics: Principles for choosing good metrics including finding short-term proxies for long-term outcomes and avoiding retention as a direct goal
• 46:30–55:28 — Simplifying metrics for better outcomes: Why simple metrics outperform composite scores, with examples of failed "merchant health scores" and successful decomposition into understandable components
• 55:28–01:00:12 — Focusing on edge cases and fail states: The importance of setting goals around disaster scenarios like "never delivered" orders that cause disproportionate business impact
• 01:00:12–01:02:31 — Managing a global data organization: Insights on running international data teams and how problems are more similar across countries than different
• 01:02:31–01:05:25 — Leveraging AI for productivity: How DoorDash uses AI tools like "Ask Data AI" to empower non-technical users and reduce bandwidth demands on the analytics team
• 01:05:25–01:08:40 — Building diverse and skilled data teams: Strategy of importing talent from different functions and company stages to create complementary skills and diverse perspectives
• 01:08:40–End — Lightning round: Book recommendations, favorite products, life motto about sleep solving problems, and career influences including strong women leaders

The Case for Centralized Data Organizations

Jessica champions a contrarian approach: centralized analytics teams that earn seats at decision-making tables rather than serving as embedded support functions. Despite business leaders preferring dedicated resources, centralized models deliver superior outcomes.

• Analytics becomes a business impact driving function with shared goals and aligned incentives alongside product, engineering, and operations. Teams answer "what do we do now?" rather than just providing dashboards or responding to support tickets.
• Consistent talent bars across all data hires prevent the quality variations that emerge when different business units apply different standards. Centralized teams use the same evaluation rubric and technical requirements, resulting in stronger overall capabilities.
• Growth opportunities multiply when data professionals can move between domains rather than being trapped as the most senior analyst in a single business unit. Career progression includes lateral moves across marketing, product, and operations analytics plus promotion paths across multiple areas.
• Methodology consistency prevents different teams defining core metrics like "sales" differently. One churn prediction model serves six teams rather than recreating it everywhere, with shared definitions improving through collective input.
• Economies of scale emerge as common problems surface across business units, enabling automation, improved processes, and early identification of scaling challenges before they become critical.

The solution uses pod structures mapping to business units while maintaining centralized reporting. Teams feel embedded locally while preserving centralized benefits.

Hiring for Curiosity Over Credentials

Building exceptional data teams requires prioritizing natural investigative instincts over traditional qualifications. Curiosity cannot be taught, making it the most valuable screening criteria.

• The best candidates pull threads when something seems odd, investigating anomalies without direction rather than stopping at minimum answers. This self-motivated problem-solving distinguishes top performers from technically competent but passive analysts.
• Interview techniques include presenting cases with intentional inconsistencies to see whether candidates notice and investigate discrepancies. Real DoorDash business problems work better than abstract cases because they reveal how people approach ambiguous situations with incomplete information.
• Response to being wrong reveals crucial adaptability skills. Data work involves constant iteration based on unexpected results, so candidates must demonstrate grace under correction and willingness to pivot when proven incorrect.
• Decision-making under uncertainty becomes critical since data professionals frequently recommend actions without perfect information. Push candidates to take definitive positions rather than hedging to test comfort with judgment calls.

Technical skills remain table stakes through coding exercises, but curiosity and problem-solving instincts create the differentiation between good and exceptional team members.

Building Extreme Ownership Culture

DoorDash's data team operates under extreme ownership where members solve problems completely regardless of role boundaries. This philosophy drives exceptional impact and professional development.

• Data scientists focus on figuring out what's happening rather than limiting themselves to analytical tasks. When quantitative analysis hits limits, they pick up phones and call customers directly rather than declaring work complete or passing responsibility elsewhere.
• Cross-functional work becomes normal, with data professionals engaging in product management, engineering, and operations when necessary to unblock decisions. This keeps work interesting while ensuring efficient problem resolution.
• The culture traces to startup days when founders took out garbage because it needed doing. This willingness to do whatever it takes to win, regardless of title, became embedded DNA that continues influencing operations at scale.
• Real examples include data scientists making customer calls to understand why affordability initiatives failed, gathering qualitative insights that quantitative analysis couldn't provide. The team takes direct action rather than accepting analytical limitations.

Leadership expectations reinforce outcome ownership over role adherence: "Yes you are a data scientist but your goal is to figure out what's happening, and if that means calling customers, that's what you do."

Mastering Metrics: Simplicity Over Sophistication

Jessica's metric design emphasizes simple, actionable measures that drive long-term outcomes through short-term behavioral changes. Common mistakes involve complex composites that teams can't understand or influence.

• Use proxy metrics for long-term outcomes rather than optimizing lagging indicators directly. "Retention is a terrible thing to goal on" because it's nearly impossible to drive quickly, whereas retention inputs can be measured and influenced short-term.
• Simple metrics outperform sophisticated composites because teams need intuitive understanding to drive improvement. A "merchant health score" of 35 provides no actionable direction, while separate goals for photo coverage and accurate hours give clear improvement targets.
• Common currency translation enables cross-functional decisions by quantifying all levers in shared terms like gross order value. Understanding that reducing price by $1 generates equivalent volume to cutting delivery time by 1 minute allows rational resource allocation across marketing and logistics.
• Edge case focus prevents averaging fallacies where rare but devastating events get overlooked. "Never delivered" orders occur infrequently but cause disproportionate churn and cost, requiring dedicated goals despite minimal average impact.
• Avoid metric rotation inefficiency where teams constantly switch optimization targets, losing expertise in specific domains. Teams become proficient at improving cancellation rates only to be reassigned to response rates, creating unnecessary learning curves.

The key insight: metrics must be understood intuitively by improvement teams, even if this sacrifices analytical perfection for practical effectiveness.

The Value of Non-Traditional Backgrounds

Jessica's journey from art portfolio to leading tech's most respected data organization demonstrates how diverse experiences create competitive advantages. Her approach challenges conventional hiring practices.

• Self-taught necessity drove her data science transition when early DoorDash needed goal-setting and performance measurement help. Learning SQL and Python while solving real business problems developed practical skills through application rather than theoretical study.
• First principles thinking emerges naturally from non-traditional backgrounds because formal training hasn't established rigid constraints. Focusing on immediate problem-solving often produces more practical solutions than academically-driven approaches.
• Business impact grounding becomes an advantage when leaders come from finance, operations, or other business-focused backgrounds. Pragmatic perspective ensures technical capabilities stay focused on measurable outcomes rather than analytical sophistication for its own sake.
• Hiring complementary skills allows non-traditional leaders to recruit PhD-level technical expertise while keeping everyone focused on business impact. The combination often outperforms teams composed entirely of traditional data science profiles.

DoorDash imports more talent from other functions than it exports, creating diverse teams where finance backgrounds teach cash flow modeling while consulting backgrounds contribute presentation skills and statistics PhDs provide technical depth.

Scaling Global Operations Effectively

Managing international data teams requires adapting to complexity layers like currencies, languages, and regulations while recognizing that fundamental patterns remain consistent across cultures.

• Data scientists, consumers, and business problems prove more alike across countries than different despite surface variations. This consistency enables pattern recognition where solutions developed in one market transfer effectively to new regions with minor adaptations.
• Regulatory complexity adds legitimate challenges through different privacy requirements, but these represent engineering obstacles rather than analytical methodology changes. Teams adapt processes while maintaining core approaches.
• The "answer key" advantage emerges when expanding to new markets because teams can apply lessons from previous launches, creating strong hypotheses about outcomes and challenges before encountering them.

International expansion primarily adds operational complexity rather than requiring fundamental changes to analytical methodologies or team structures.

Jessica Lachs's approach reveals that exceptional analytics teams emerge from contrarian structural decisions, non-traditional hiring, and relentless business impact focus. Her philosophy prioritizes curiosity over credentials, simplicity over complexity, and extreme ownership over role boundaries. Most importantly, she demonstrates that effective data leaders often come from unexpected backgrounds and world-class teams require importing diverse perspectives rather than hiring exclusively from traditional analytics pipelines.

Practical Implications

• Structure data teams centrally with pod alignment to business units for methodological consistency while preserving embedded partnership benefits
• Hire for curiosity and self-motivation over credentials - look for candidates who investigate anomalies naturally
• Create shared goals between analytics and business teams to enable effective prioritization rather than suffering with competing requests
• Choose simple proxy metrics teams can understand and influence quickly rather than complex composites or lagging indicators
• Set goals around edge cases and failure states that cause disproportionate impact despite low frequency
• Encourage cross-functional problem-solving where data scientists call customers when quantitative analysis hits limits
• Import talent from operations, marketing, and engineering to build diverse skills and fresh perspectives
• Focus on business impact over analytical sophistication - pragmatic solutions beat technically perfect approaches
• Use real business cases in interviews to assess uncertainty handling rather than abstract exercises
• Create AI-powered tools to empower non-technical users while preserving analytical bandwidth for high-impact work
• Expect extreme ownership where team members solve problems completely rather than passing responsibility
• Quantify business levers in common currency for rational resource allocation across functions