Assumption documents fail when they read like marketing. Auditors need a trail: what the model is for, who it scores, which conditions must hold, and what happens when those conditions break.
Start with a purpose sentence you can defend
Write one sentence that states the decision the model supports. If product language says “optimize growth,” translate it into the decision: approve, decline, price, or route. Ambiguous purpose statements are where most challenge sessions stall.
Define the population without hiding exclusions
List included segments and the reasons others were dropped. Thin-file exclusions, partner-channel filters, and geographic limits belong in the same table — not in a footnote nobody opens.
Name the assumptions that actually move risk
Prioritize assumptions about feature availability, label delay, and stability of the economic environment. Decorative assumptions (“users behave rationally”) waste review time.
Close with limitations, not apologies
A limitation statement should tell a reader what the model cannot claim. In our cohorts, teams that draft limitations early receive fewer late-stage findings because they have already framed the unknown.
Want structured practice? The documentation drills in Model Risk Audit Fundamentals walk through redacted packs with faculty critique.