Estimation question

Estimate the audio minutes Deepgram must process to be profitable at $0.08/min.

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

Tests structured estimation: working backward from a per unit price to the volume needed to cover fixed and variable costs.

Identify the cost structure: variable cost per minute processed, like compute for STT, the LLM call, and TTS, plus fixed costs like infrastructure, engineering, and support.
Estimate variable cost per minute: assume a blended cost of 0.05 dollars per minute across the three components, leaving a gross margin of 0.03 dollars per minute at the 0.08 dollar price.
Estimate fixed costs: assume 20 million dollars a year in combined infrastructure, R&D, and operating costs that do not scale directly with volume.
Calculate breakeven volume: 20 million dollars divided by 0.03 dollars per minute gross margin is roughly 667 million minutes a year needed just to cover fixed costs.
Convert to a more intuitive unit: 667 million minutes is about 11 million hours a year, or roughly 30,000 hours of audio processed every single day.
State clearly that the cost per minute, fixed cost base, and margin figures are assumptions you would replace with Deepgram's real cost structure if available.

Builds the estimate from a clear cost structure, variable cost per minute against fixed costs, rather than guessing a breakeven volume directly.
Shows the full calculation chain from gross margin per minute to total minutes needed to cover fixed costs.
Converts the raw number into a more intuitive daily or hourly figure, making the scale of the answer easier to sanity check.
Flags every input, the 0.05 dollar cost, the 20 million dollar fixed cost base, as an assumption clearly open to correction.

Skipping the cost structure entirely and guessing a breakeven number with no supporting math.
Not converting the final number into an intuitive unit, leaving the scale hard to evaluate.
Treating assumed inputs as known facts rather than clearly labeled estimates.

How would this breakeven change if fixed costs were significantly higher due to R&D investment?
What would happen to the breakeven point if compute costs dropped by half?
How would you validate the assumed 0.05 dollar cost per minute?