Jessica Lachs on Building DoorDash’s Data Team, Centralized Analytics, and Metrics That Drive Real Outcomes

Source: Lenny’s Podcast Speaker: Jessica Lachs Link: Episode

Jessica Lachs is Vice President of Analytics and Data Science at DoorDash, where she has built one of the largest data organisations in consumer tech over more than a decade. She joined as the first GM, launching Boston in 2014, and taught herself SQL and Python out of necessity. This conversation covers organisational design, metrics philosophy, and the operational culture that produces analytical insight rather than dashboard maintenance.

Key ideas

• Analytics as a business impact function, not a service function. The failure mode is a data team that answers questions and builds dashboards on demand. The model Lachs built has analysts as thought partners with a seat at the table, proactively identifying opportunities and holding a point of view on what decisions the company should make.
• Centralised analytics org outperforms embedded teams. Embedding analysts in business units delivers control and camaraderie but fragments talent bars, prevents cross-pollination, and duplicates methodology. A central team with pods mapped to partner teams captures most embedding benefits while preserving career paths, consistent metrics, and a shared culture.
• Short-term proxy metrics for long-term outcomes. Goaling a team on retention is ineffective — retention is slow to move and hard to attribute. Find the short-term input metrics that predict long-term retention and run experiments against those. Then verify empirically that moving the input actually shifts the outcome.
• Simple metrics over composite scores. DoorDash built a composite merchant health score and found it meaningless in practice: no one could interpret whether 0.35 was good, and no one knew which lever to pull. Three simple metrics — merchant photo coverage, accurate hours, first-week order rate — performed better than the composite they replaced.
• Measure fail states, not just averages. ‘Never Delivered’ orders are rare but catastrophic: they cause churn and cost far more than a successful delivery. Tracking only averages masks tail failures. DoorDash has a team whose sole goal is to eradicate Never Delivered by understanding root causes (human error, fraud, logistics) and building preventive systems.

Topics covered

• Centralized vs. embedded analytics: the case for a centre of excellence
• Pod structure: central reporting with partner-team alignment
• Analytics as business impact driver, not a service function
• Curiosity as the unteachable data hiring criterion
• Case interview design: embedding hidden signals in stated problems
• Short-term proxy metrics and the theory of change
• Simple metrics over composite scores: the merchant health score case
• Never Delivered as a fail-state metric
• Common currency (GOV and volume) to compare cross-team investments
• Hackathon discovery: the referral fraud bimodal distribution case
• Extreme ownership culture: data scientists making customer calls
• Managing global analytics across DoorDash and Volt
• AI tool: Ask Data AI — democratising SQL access inside DoorDash