The Estimate Nobody Trusts
You are managing a data migration project moving 12 legacy systems into a single platform. Your lead architect provides a bottom-up estimate of 14 weeks for the migration work. A vendor's comparable project took 22 weeks. Your own historical data from three similar migrations shows 18, 20, and 26 weeks. The sponsor wants a single number for the business case by Friday and is leaning toward the 14-week figure because "it's the detailed one." Two of the legacy systems have never been inventoried, and the architect admits the estimate assumes clean data.
1. How do you construct the estimate you present, and how do you handle the sponsor's preference for 14 weeks?
Read the worked answer
The 14-week bottom-up estimate is not wrong - it is conditional. It is valid only if the two uninventoried systems are clean and the data is clean, which the architect has explicitly flagged as an assumption, not a finding. Presenting it as the project estimate converts an assumption into a commitment.
The defensible approach is a range with the basis stated:
- The bottom-up 14 weeks is the optimistic bound, contingent on the stated assumptions.
- The historical data (18, 20, 26) and the vendor benchmark (22) cluster around the high teens to low twenties - that is the realistic range, and the 26-week case is a real outcome from a comparable project, not a tail to discard.
- The two uninventoried systems are the dominant uncertainty; until they are assessed, the estimate should carry explicit contingency for them.
To the sponsor: explain that "detailed" and "accurate" are different properties. A detailed estimate built on unverified assumptions is precise but not reliable, and the historical record is the better predictor precisely because it captures the surprises the bottom-up estimate cannot see. Offer a single planning number (e.g., a P50 near 20 weeks) with the range and the conditions attached, and propose an early inventory spike on the two unknown systems to tighten the range before the business case is locked.
2. What does the spread between 14 and 26 weeks tell you about the project's risk profile, and what should you do about it now?
Read the worked answer
A spread of nearly 2x between the optimistic and pessimistic cases signals high estimation uncertainty concentrated in a few identifiable drivers, not general chaos. Here the drivers are concrete: two uninventoried systems and the clean-data assumption. That is good news - concentrated uncertainty can be attacked.
Actions now:
- Run a short inventory/assessment spike on the two unknown systems before committing to a number. This is the cheapest way to collapse the widest part of the range.
- Validate the clean-data assumption early with a profiling pass on a sample of each legacy system - data quality is the classic migration surprise, and it is testable.
- Carry explicit contingency proportional to the unresolved drivers rather than a flat percentage, and state what would trigger its use.
- Set a decision point - if the spike reveals significant remediation, the 26-week case becomes the planning basis, and the sponsor should know that now rather than at week 10.
The judgment: wide estimate ranges are a signal to investigate, not to pick the number you like. The sponsor's preference for 14 weeks is a preference for a lower business case, not evidence about the work.