Diagnose and recover a sudden 35% drop in Medium's active-user percentage

You are the product manager for Medium. For the past three weeks you have been tracking the percentage of active users, which has been moving up and down. Then one morning the figure shows a drop of 35%. Explain how you would work out what caused the drop and how you would get the metric back to where it was. ### Clarifying Questions - How is the percentage of active users defined: active users divided by registered accounts, daily actives as a share of monthly actives, or something else? Is the 35% a relative fall or 35 percentage points? - Is the morning figure a complete day of data, or a partial reading from a dashboard that refreshes during the day? - How large were the swings over the past three weeks, and did they go in both directions? - Has the drop already been cut by platform, country or user type, or is that still to be done? ### Part 1 — Find the cause Lay out how you would investigate the drop, in the order you would check things. For each hypothesis, say what evidence would confirm it or rule it out. ```hint Question the number first Before hunting for a product or market cause, decide what you would check to be sure the drop reflects what users actually did. ``` #### What This Part Should Cover - Validation of the metric and the data behind it, including both halves of the percentage - A segmentation plan and what each cut would point to - Internal and external hypotheses ranked by how well they fit a sudden overnight fall - Whether the earlier three weeks of fluctuation are linked to the drop ### Part 2 — Restore the metric Once you have a likely cause, explain how you would bring active users back and keep the problem from recurring, and how the plan differs depending on what the cause turned out to be. ```hint Count who was lost during the drop Fixing the cause stops further loss, but think about the people who could not use Medium, or stopped coming, while the problem was live. ``` #### What This Part Should Cover - A response matched to the type of cause, with clear owners - Winning back users affected while the problem lasted - A clear test for when the metric counts as restored - Monitoring that would catch a similar drop sooner ### What a Strong Answer Covers - A structured diagnosis checked in a sensible order, rather than a list of guesses - Use of the scenario's specifics: the size and suddenness of the drop and the earlier weeks of fluctuation - Urgency proportionate to a 35% fall, balanced against confirming the cause before acting - Communication with engineering, data and leadership while the investigation runs ### Follow-up Questions - If every segment fell by roughly the same 35%, what would that suggest, and where would you look next? - Suppose the loss is concentrated among visitors arriving from search engines: what can Medium control and what can it not? - How would you set up alerting so that a drop like this reaches someone within the hour instead of the next morning?