NASA Requirements vs Cost Overrun: The 5-10% vs 20% Curve
One of the strongest pieces of evidence for "invest early to save late": NASA found that spending 5-10% of effort on requirements yields ~50% overrun, while spending 20% holds overrun to ~10%.
The curve in full
| Definition spend (Phase A/B) as % of total program cost | Cost overrun observed |
|---|---|
| under 5% | about 80% to over 200% |
| 5-10% | about 50% |
| 10-15% | about 30% |
| 15-20% | about 10% |
| 20% or more | close to 0% |
Approximate banding read off Werner Gruhl's scatter chart "Effect of Requirements Definition Investment on Program Costs" (Werner M. Gruhl, Chief, Cost & Economic Analysis Branch, NASA Headquarters). Gruhl plotted roughly 30 individual NASA programs, not averaged bands, and the y-axis is cost overrun only. The lowest-definition programs (under ~5% spent on Phase A/B) ran 100-200%+ over budget; only a handful of well-defined programs finished on budget.
Why the curve is so steep at the low end
The intuition is that requirements errors discovered late are vastly more expensive to fix than requirements errors discovered early. Boehm's 1981 cost-to-fix curve (a separate finding) showed defects cost about 5x to fix in design, 10x in build, 20x in test, and 100x in operations. NASA's requirements-vs-overrun curve is essentially Boehm's curve applied at the project level: starve the requirements phase and you bake in expensive rework.
For project managers, the practical reading is: budgets that allocate only 5-10% of effort to discovery and requirements are not lean. They are pre-committing to 30-50% cost overrun. The cheapest insurance against overrun is to fund the front of the project properly.
How to cite
The chart is widely reproduced in systems-engineering literature; see the discussion in the SEBoK "Economic Value of Systems Engineering" article and the NASA Systems Engineering Handbook. Gruhl's underlying per-program data was never published as an open dataset.