Amazon Video Reviews: Launch Now With a 3-Day Delay or Wait Six Months for ML?

Amazon is building a new feature that lets customers record a video review of a product and upload it, so the video shows on the product page for other shoppers. Video reviews increase sales, and timing matters: the sooner a video becomes visible, the more it helps the seller while shoppers are still interested in the product. The processing approach available today takes about 3 days on average before an uploaded video is visible to other customers. A machine-learning approach would bring that down to roughly 2 hours, but it needs another 6 months of development. As the product manager, would you launch the feature now without ML, or wait until the ML version is ready? Explain how you reach the decision and what you would do in either case. ```hint Count both timelines Lay out what each option delivers month by month over the first year, and ask what shipping early could take away from the faster version that arrives later. ``` ### Clarifying Questions - What makes up the 3-day turnaround (video processing, a review queue, staffing), and could it be shortened without ML? - Would launching now pull engineers off the ML work and push its 6-month date out? - Does a video lose value at the same rate for every product, or faster for new releases, trending and seasonal items than for products that sell steadily for years? - Can the feature launch to some categories, sellers or marketplaces first, with a cap on upload volume? ### What a Strong Answer Covers - A clear comparison of the cost of waiting (six months with no video reviews at all) against the costs of a slow launch (less value per video, operating cost, a weaker first impression, any delay to ML). - An explicit decision rule, ideally a simple model of the two options, rather than a gut call. - Middle paths: a staged rollout to categories where a 3-day lag matters least, non-ML speedups, and setting expectations for uploaders. - What an early launch buys besides sales, such as evidence of demand and data that helps the ML effort. - Success metrics, guardrails, and the conditions that would make you pause or reverse. ### Follow-up Questions - Suppose the version you can ship today already makes videos visible in about 2 hours, and ML would make it near-instant. Does your recommendation change? - Uploads far exceed your forecast and the backlog pushes turnaround well past 3 days. What do you do? - How would you measure the sales lift precisely enough to judge whether the ML investment is still worth finishing?