Communicate forecast uncertainty with prediction intervals

WHAT THIS TESTS This evaluates statistical maturity and communication. The interviewer wants you to resist false precision, quantify uncertainty in a defensible way, and explain it to a non-technical stakeholder. Bonus signal: distinguishing a prediction interval from a confidence interval.

A GOOD ANSWER COVERS Start by reframing: a fifteen percent point forecast is the center of a distribution, not a guarantee. Quantify the spread using whatever the model supports. If it is a regression or time-series model, derive a prediction interval from the model's residual error and forecast variance, which widens further into the future. If inputs are uncertain, run a Monte Carlo simulation: place distributions on the drivers, like conversion rate and acquisition volume, sample many times, and read off the percentile range of outcomes. If data is thin, use scenario analysis with explicit conservative, base, and optimistic assumptions. Report the result as a range with a stated probability, for example a roughly eighty percent chance revenue lands between ten and twenty percent growth. Clarify that a prediction interval describes where a future observed value should fall, which is wider than a confidence interval that describes uncertainty about an estimated parameter. Finally, list the assumptions and the events that would push the actual outside the band.

COMMON WRONG ANSWERS Defending the single fifteen percent figure as if it were certain. Giving a range with no probability attached or no method behind it. Confusing a prediction interval with a confidence interval. Forgetting that intervals widen as the horizon lengthens. Hiding key assumptions, so the stakeholder cannot judge the risk.

LIKELY FOLLOW-UPS Why is a prediction interval wider than a confidence interval? Because it includes both parameter and irreducible individual variance. How do you choose the interval width? Tie it to a probability level and the stakeholder's risk tolerance. How do you update the forecast as actuals arrive?

ONE CONCRETE EXAMPLE A team simulates next quarter's revenue by sampling distributions for new signups, conversion, and churn ten thousand times. The median lands near fifteen percent growth, and the tenth-to-ninetieth percentile spans eight to twenty-two percent. They tell the stakeholder: our base case is fifteen percent, but there is roughly an eighty percent chance it falls between eight and twenty-two, driven mainly by acquisition uncertainty, so plans should be robust across that band.