Design Delivery-Time Prediction with a Cold-Start Plan
Company: Instacart
Role: Machine Learning Engineer
Category: Machine Learning
Difficulty: medium
Interview Round: Technical Screen
Design a system that predicts when a grocery order will arrive. Explain the model, the training data, and how the system would start operating before enough historical examples are available.
### Clarifying Questions
At what point is the prediction shown: before checkout, after assignment, or throughout fulfillment? Is the output a point estimate, a time range, or both? Which timestamps and live operational signals exist? No latency target, error threshold, or available feature set is specified. State your chosen scope and distinguish desired inputs from confirmed data.
### What a Strong Answer Covers
Define a leakage-safe label and feature snapshot, establish a cold-start baseline, select an appropriate loss, evaluate error across operational slices, and explain online updates and monitoring.
### Follow-up Questions
- How would you handle a new store or region with little history?
- How would you evaluate an ETA that changes repeatedly during fulfillment?
Overview: Design grocery delivery-time prediction from a practical cold-start baseline to a monitored learned model. Cover timestamped labels, leakage prevention, quantile estimates, sparse stores, and changing ETAs.
Read the full Instacart Machine Learning Engineer interview experience this question came from