My last four projects went over hours. I keep thinking it's bad luck. What if it isn't?

The situation

Priya runs a 4-person brand identity and web design agency. She estimates project hours before quoting, prices at a fixed fee based on that estimate, then routinely hits 15–25% over. She's been telling herself it's scope creep, one-off client behavior, or bad luck. But it keeps happening.

The analysis

This is the planning fallacy: estimators anchor on best-case sequences and underweight the tail of complications.

The What-If Calculator surfaces the structural issue. Priya estimates a typical brand identity project at 80 hours. But when she models her uncertainty — not a single number, but a distribution — the shape of her actual experience emerges. If her estimate were unbiased, overruns and underruns should be roughly equal in frequency. She looks back at her last 12 projects: 9 went over, 3 came in under. That's not bad luck. That's a systematic bias called the planning fallacy: estimators anchor on best-case sequences of events and don't adequately weight the tail of complications.

Modeling hours as a lognormal distribution (mean: 80, std dev: 18) — which is appropriate for tasks where time can't go negative and overruns compound — produces a median of 78 hours but a mean of 84. The 90th percentile is 103 hours. Her fixed-fee quote based on 80 hours means she's working for free above 80 hours on roughly 60% of projects.

The intervention is simple: add a 20% complexity buffer to all estimates and communicate it as a project investment range rather than a fixed fee. This shifts her from quoting '$8,000' to quoting '$8,000–$9,600 depending on revision cycles.' No client has declined based on this framing, and Priya's average project margin improved by 18 points in the first quarter she applied it.

The decision

Priya adopts a range-based quoting policy and saves the lognormal hours model to her distribution library, with Monitoring watching every project's actual hours against it. Two quarters later, Monitoring flags that the distribution has drifted — a new e-commerce service line she added runs consistently heavier than her original brand identity work, and the blended fit is quietly overstating her buffer. She splits it into two separate distributions, one per service line, and re-quotes accordingly instead of re-learning the same lesson from a fresh round of overruns.