An architect must estimate a large piece of work with genuine uncertainty. What makes an estimate useful rather than misleading?

What makes it useful

1. A range, with the assumptions stated. A single number is read as a commitment and is wrong. A range with its drivers named — "four to nine months, depending mainly on how much undocumented behaviour we find in the legacy system" — is honest and actionable.

2. The dominant uncertainty identified. Usually one or two unknowns account for most of the range. Naming them tells everyone what to investigate first, and frequently a week of investigation narrows the range more than any amount of estimation refinement.

3. Decomposition into knowable and unknowable parts. The parts resembling work you have done are estimable; the parts that do not are not, and saying so is more useful than averaging them together.

4. A cheap way to reduce the uncertainty. A spike, a prototype, a sample migration, a traffic shadow. For large uncertain work, buying information is usually better value than estimating harder.

5. What the estimate excludes, explicitly — the coordination, the migration, the decommissioning, the unknown consumers.

What makes it misleading

• A point estimate, which is treated as a commitment and defended rather than revised.
• Estimating the happy path, omitting the parts that historically consume the time: data migration, behavioural differences, coordination and decommissioning.
• Anchoring on a desired date and working backwards.
• Aggregating optimistic component estimates, which compounds rather than averages.
• Refusing to estimate, which forces someone with less information to do it.

The specific pattern for migrations

The copy is estimable; the verification and the long tail are not. Almost every migration estimate is wrong because it costs the bulk transfer and omits proving correctness, handling the last small percentage of awkward cases, and removing the old path — which together frequently exceed the main work.

The professional framing

An estimate is a statement about current knowledge, not a promise. Presenting it that way, with the range, the drivers and a plan for reducing them, is what makes it useful — and what makes it revisable without anyone feeling misled.