How would you probabilistically forecast 40 stories using throughput data?

WHAT THIS TESTS: Whether you understand that knowledge work has inherent variability and that historical throughput should be treated as a probability distribution, not an average rate. The interviewer wants to see if you can explain Monte Carlo or percentile-based forecasting and why a single-date estimate is misleading for complex work.

A GOOD ANSWER COVERS: First, data hygiene: gather 8 to 12 recent time periods of throughput, ideally from stable team conditions, and filter out anomalies like holiday weeks or sprints with major outages. Second, model selection: explain that you would use the historical throughput distribution to run a Monte Carlo simulation or simple percentile resampling, drawing random weekly or sprintly throughput values and summing them until you reach 40 stories, repeating this thousands of times. Third, presentation: produce an S-curve or cumulative probability chart showing delivery dates against percentiles, such as a 50th percentile date for internal planning, an 85th percentile for stakeholder commitment, and a 95th percentile for hard dependencies. Fourth, communication: explicitly tell the product manager that a single date implies false precision, while percentiles let the business choose its appetite for risk.

COMMON WRONG ANSWERS: Using average throughput multiplied by 40 stories to generate one deterministic date, even with a hidden buffer. Treating story count as perfectly deterministic rather than acknowledging that scope often shifts. Using velocity points instead of throughput counts when the question specifies similarly-sized stories and asks for throughput. Failing to mention sample size or data stability, which signals shallow familiarity with statistical forecasting.

LIKELY FOLLOW-UPS: How would you adjust the forecast if the team composition changes next quarter? What if the historical data shows a trend or seasonality rather than a stable distribution? How do you communicate to leadership that an 85th percentile date means there is still a 15 percent chance of missing it? Would you re-forecast during the quarter, and what trigger would cause you to do so?

ONE CONCRETE EXAMPLE: Suppose the team completed 3, 5, 4, 6, 5, 4, 7, and 5 stories over the last eight sprints. Instead of averaging five stories per sprint and promising eight sprints for 40 stories, you would use those eight values as a deck of cards, randomly draw sprint results with replacement, sum them until they reach or exceed 40, record how many sprints that took, and repeat ten thousand times. The results might show a 50th percentile of eight sprints, an 85th percentile of ten sprints, and a 95th percentile of eleven sprints. You then tell the product manager that the team is roughly 85 percent confident of finishing within ten sprints, but committing to eight would ignore historical variance.

Source: Focused Objective (focusedobjective.com)