Diagnosing a 10% Rise in YouTube Playback Buffering Time

You are a product manager at YouTube, and the average time viewers spend waiting for videos to buffer has gone up by 10%. How would you find the cause, and what would you do about it? ```hint Is the average telling the truth? An average can rise because some viewers had a worse experience, because the mix of viewers, devices or videos changed, or because a few extreme sessions skewed it. Work out which before blaming the network. ``` ### Clarifying Questions - What does buffering time include: the wait before playback starts, stalls during playback, or both? Is it averaged per view, per hour watched or per viewer? - Did the 10% rise happen suddenly or build up over weeks, and is it measured against last week or the same period last year? - Is it visible on every platform (Android, iOS, web, TV apps) and in every region, or concentrated somewhere? - Have watch time or the share of viewers who give up before playback starts moved with it? ### What a Strong Answer Covers - Pinning down the metric's definition, its denominator and the timing of the rise before diagnosing - Ruling out logging changes, outlier sessions and shifts in the viewer or content mix - Segmenting by device and app version, network and internet provider, region, content type and delivery infrastructure - Ordered internal and external hypotheses, each with the check that would confirm it - A fix-and-measure plan with guardrails, especially for video quality ### Follow-up Questions - If the rise sits entirely with viewers on one internet provider in one country, what do you do next, and who do you involve? - An engineer proposes lowering the default starting resolution to cut buffering. How would you evaluate that? - How would you decide whether a 10% rise matters enough to take priority over other work?