You are asked how long a migration will take. You genuinely do not know within a factor of three. What do you say?

What the interviewer is testing

Whether you can be useful under genuine uncertainty rather than either refusing or inventing a number.

What not to do

Do not give a single number, which will be treated as a commitment and repeated without its caveats.

Do not refuse to estimate. The business must plan, and "I cannot say" transfers the problem back without helping. An architect who cannot support planning is not doing the job.

What to say

Give a range with the reasoning: "Between four and twelve months. The spread is driven by three unknowns: how many undocumented consumers depend on the current system, whether the data can be migrated incrementally or requires a single cutover, and whether the calculation logic can be extracted or must be reimplemented."

That is far more useful than a point estimate because it tells the listener what determines the answer, and it lets them decide how much to invest in reducing the uncertainty.

Propose work that narrows the range. "Three weeks of discovery — dependency analysis from network traffic and database access logs, plus a spike extracting one calculation — would let me give you a range of plus or minus 30%." That converts an unanswerable question into a small, funded piece of work with a defined output.

Name the decision the estimate supports. If they need to know whether it is this year or next, the range already answers it and further precision is not needed. Frequently the required precision is far lower than the question implies.

The commitment to offer

Incremental delivery with checkpoints, so the estimate is revised with evidence rather than defended. A migration structured as slices lets you report progress against reality every few weeks, which is worth more to a planner than any initial estimate.

What a strong answer adds

Explaining where the historical bias comes from: the unknown work is systematically the part nobody can see — undocumented behaviour, forgotten consumers, data quality. Naming that upfront makes the range credible and inoculates against the later accusation of sandbagging.

Common weak answers

A confident single number. "It depends" with no range and no path to narrowing it.