DoorDash · AI product · Mid · 45 to 60 minutes

How should DoorDash personalize its homepage?

DoorDash is a marketplace app that helps people discover local restaurants and other merchants and get deliveries. Answer this product management case and get it graded line by line against a weighted rubric.

Product strategy
AI product judgment
Systems thinking
Experimentation
Risk management

The situation

You are the PM for DoorDash's consumer homepage, the first screen many customers see before they place an order. Today the homepage selects from a few hundred predefined food rows, like salads or breakfast burritos, based on a customer's past orders and browsing, then ranks stores inside those rows. The system is dependable, but teams believe it is hitting a ceiling. Many users see broad or repetitive rows, niche tastes get flattened into generic categories, and homepage quality depends heavily on merchant tagging. Leadership wants a recommendation before the next planning cycle because repeat usage and discovery are becoming more important growth levers, especially in dense urban markets. You need to decide whether DoorDash should keep improving the current taxonomy and ranking system, or invest in a different way to define what appears on the homepage in the first place. The hard part is balancing customer relevance, merchant discovery, cost, latency, quality and safety risks, and the need to operate at DoorDash scale without relying on large amounts of manual curation.

What to produce

A 2 page product strategy memo.

Your answer must

Explain the core problem with the current homepage and separate issues in taxonomy, row selection, retrieval, and ranking.
Compare at least two credible options, including improving the current fixed row system and changing how homepage rows are defined, then recommend one.
Describe the proposed end to end product flow for how user signals become homepage rows, store candidates, and final ranked results.
Define success metrics and an evaluation plan that includes both online business metrics and offline quality metrics.
Explain how your chosen approach will handle quality and safety risks of user facing homepage content, with clear guardrails and escalation paths.
List the biggest assumptions, risks, and first tests you would run before a broader rollout.

Constraints

The homepage must keep current app open performance expectations, with no meaningful added latency at serving time.
The solution must work across millions of users and merchants, so manual creation or review can cover only a small fraction of content.
Merchant tagging quality is uneven and cannot be fully fixed in the near term.
Your MVP should be feasible within a normal planning cycle, without requiring a full rebuild of DoorDash's core marketplace systems.

Format

Use sections for problem, options, recommendation, product flow, metrics and experiment plan, and risks.

How DoorDash actually played it

Explains how DoorDash moved from ranking a fixed set of homepage rows to generating the rows themselves with LLMs, using a staged pipeline and measured results. The piece is useful for seeing where generation beats reordering in a personalization product.

Question the structure of a surface, not just the ranking inside it
Tie an AI pipeline to real business metrics before scaling it
State plainly what the system still cannot do

Built from How DoorDash Used LLMs to Rebuild Its Homepage by Sid Arora (JAPM).