Raise at-checkout delivery ETA accuracy from 88% to 95% for a warehouse delivery service

You are a product manager at a company that delivers orders from its own warehouses to customers' homes. As soon as a customer places an order in the consumer app, the app shows an estimated delivery time. The head of operations is worried about how often that estimate turns out to be right: over the last three months, the organization-wide accuracy of the estimate shown at order time was 88%, and they want it at 95% within the next three months. What you know about the operation: - The estimate appears in the consumer app right after checkout. - Drivers carry out deliveries through a dedicated driver app, and their location is known whenever they are actively working in it. - Every milestone of an order is recorded and stored, from the moment it is processed and joins the warehouse queue until the driver completes the handoff to the customer. - Drivers are independent contractors who collect orders at the warehouse and choose which days and shifts they work. - Nearly all orders travel 10 miles or less and arrive within 30 minutes. Answer in the two parts below. ### Clarifying Questions - How is an estimate judged accurate: delivered by the promised time, within some window around it, and does an early arrival count as a miss? - Is the estimate a single time or a range, and does the metric score only the estimate shown at checkout or also later updates? - Is the 88% roughly uniform, or do some warehouses, times of day or order types miss far more often than others? - Beyond estimation software, are operational changes in scope, such as warehouse staffing, driver incentives or the delivery promise the business offers? ### Part 1 — Partnering with the head of operations and choosing where to invest How would you work with the head of operations on this goal, and how would you decide where your team's resources should go? ```hint Break the 88% apart Every order's total time is a chain of recorded milestones. Ask which links in that chain, and which slices of orders, account for most of the misses before you pick anything to build. ``` #### What This Part Should Cover - How you align with operations on the goal, the accuracy definition and the way the two teams will work together - How the milestone data is used to locate where and why estimates miss - How candidate investments are sized and ranked against the three-month deadline ### Part 2 — Features that make the estimate more accurate Which specific software or technology features would make the delivery time shown to customers at order placement more accurate? Explain which source of error each one targets. ```hint Follow the variance The estimate is only as good as its shakiest stage prediction. For each stage of an order, ask what live signal the system could observe but probably does not use yet. ``` #### What This Part Should Cover - Features tied to specific stages: the warehouse queue and picking, driver availability and pickup, travel, and the handoff at the door - Improvements to the data that feeds the estimate, not only to the model - Which features can ship within three months and which come later ### What a Strong Answer Covers - A precise definition of accuracy agreed with operations, including how early and late arrivals are treated - A diagnosis grounded in the per-milestone data rather than a jump to solutions - Features that use the stated assets (milestone logs, driver location, the two apps) and deal with the independent-driver supply model - A success metric plus guardrails that stop the team from reaching 95% by simply padding every estimate - A realistic sequence of work for the three-month target ### Follow-up Questions - Padding every estimate with extra minutes would probably push accuracy past the target. Why is that a bad idea, and how would you prevent it? - Drivers choose their own shifts. How would the estimate stay accurate on a day when far fewer drivers log on than expected? - After the first month, accuracy has improved at most warehouses, but one is stuck well below target. How would you proceed?