An architect is asked how long a migration will take and has genuinely no idea. What is the responsible answer?

The responsible answer

Not a number, and not a refusal. Both are unhelpful: a fabricated number will be treated as a commitment, and a refusal leaves the person who asked unable to plan.

The responsible answer has three parts:

1. A range with a stated confidence — "somewhere between three and nine months, and I would not be surprised by twelve" — which communicates the uncertainty honestly rather than hiding it in a single figure.
2. What drives the range, named specifically: the number of undocumented consumers, whether the data converts cleanly, whether the old system's expert is still available, whether the coexistence period is weeks or a year.
3. What would narrow it, and how much that costs — "two weeks of investigation on the data and the consumers would move this to a three-month range."

Why the third part matters most

It converts an estimation problem into a decision. The person asking can choose to spend two weeks buying certainty or to plan against the wide range, and both are legitimate — but only if they know the option exists.

An estimate offered without the option to improve it is a guess presented as a finding.

The decomposition that helps

Estimate the parts you understand and name the parts you do not. A migration's mechanical work is frequently estimable within 30%; the unknowns are concentrated in data quality, undocumented consumers, and the coexistence period.

Saying "the build is six weeks and the unknowns are the schedule" is more useful than any single number, because it directs investigation at the right place.

The specific pattern in migrations

The last 2% takes most of the effort, because it is the anomalies: records created before a constraint existed, records fixed by hand, encodings nobody documented, and the consumer discovered after cutover.

An estimate that does not allow for the tail will be wrong by a large multiple, and stating that explicitly is what makes the range defensible rather than pessimistic.