Estimation question

If you are planning to kick off YouTube TV in SFO, how much bandwidth would be required?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Structured estimation: breaking network bandwidth into population, adoption, concurrency and stream-quality drivers, then sanity-checking the result.

Clarify whether you're sizing total network bandwidth into the city or bandwidth per household, and at what video quality (SD, HD, 4K).
Estimate the SF Bay Area population (about 4.7 million) and households (roughly 1.9 million).
Estimate launch adoption for a paid live-TV streaming product, say 8-10% of households in year one.
Estimate peak concurrent viewers using prime-time patterns, say 20-25% of subscribers watching during the 7-10pm window.
Apply an average bitrate per stream (about 5 Mbps for HD) to get total peak bandwidth, then convert to Gbps.
Sanity check the figure against known metro ISP backbone capacity to see if it's plausible.

States every input as a labeled assumption (e.g., 'I'll assume 8% adoption') rather than presenting guesses as facts.
Distinguishes peak bandwidth, what infrastructure must support, from average daily bandwidth, what a cost model would use.
Uses round numbers, roughly 1.9 million households, 10% adoption, 25% peak concurrency, 5 Mbps HD, landing near 240 Gbps peak.
Notes CDN edge caching shifts the real bottleneck from backbone bandwidth to last-mile bandwidth.

Skipping the clarifying question and estimating the wrong thing, like per-user speed instead of aggregate load.
Using average instead of peak concurrency, which understates what Google actually needs to provision.

Chapters of the AI PM course that teach what this question tests.