Start project estimates with reference class forecasting from 3-5 comparable past projects — base rates beat inside-view planning

Why This Is a Rule

Kahneman and Tversky's distinction between the inside view (planning based on the specific details of this project) and the outside view (planning based on what happened with similar projects) is one of the most important findings for estimation accuracy. The inside view generates optimistic estimates because it focuses on how things should go; the outside view generates realistic estimates because it focuses on how things actually went.

Reference class forecasting operationalizes the outside view: before estimating this project from its specifics, find 3-5 comparable past projects and use their actual outcomes as the starting point. "How long did our last three website redesigns take?" provides a much better anchor than "How long should this website redesign take based on the feature list?" The past projects include all the complications, delays, scope changes, and unexpected problems that the inside view systematically omits.

The 3-5 project minimum is necessary because fewer than 3 doesn't establish a reliable pattern (one outlier can dominate), while more than 5 provides diminishing returns and may include projects that are less comparable. The reference class should be similar in scope, complexity, team, and domain — close enough that the base rate is relevant, not so broad that it dilutes the signal.

When This Fires

• When beginning the estimation process for any multi-day or multi-week project
• When inside-view estimates feel precise but ungrounded in experience
• When projects in your area consistently overrun despite detailed bottom-up planning
• Complements Apply your personal estimation ratio as a multiplier — if actual/estimated = 1.8, budget at 1.8x your initial estimate (personal estimation multiplier) with the project-level outside-view anchor

Common Failure Mode

The "this time is different" objection: "Past projects took 6 months, but this one has a simpler scope / better team / clearer requirements, so it will only take 3." Every project feels different from the inside. The reference class exists precisely because inside-view reasoning always finds reasons for optimism. Start with the base rate; then adjust for genuinely unique factors — but the adjustment should be modest, not halving the estimate.

The Protocol

(1) Before estimating the new project, identify 3-5 comparable past projects. Comparable means: similar scope, similar complexity, similar team size, similar domain. (2) Record their actual outcomes: how long did they take? What was estimated vs. actual? (3) Use the average actual duration as your anchor estimate — the starting point before any project-specific adjustments. (4) Now consider project-specific factors that genuinely differ from the reference class: simpler scope, more experienced team, tighter constraints. Adjust the anchor modestly (±20% maximum for each factor). (5) The final estimate should be between the reference-class average and your inside-view estimate, weighted toward the reference class. If the two diverge dramatically, trust the reference class — your inside view is probably doing exactly what the planning fallacy predicts.