Module 3

Sampling and coverage: what you can safely leave unread

A data room can hold ten thousand documents and the team has two weeks. Reading everything is impossible, and pretending to is dishonest. Sampling is the standard answer: read all of the documents that matter most, and read a well-chosen part of the rest to learn how much it is likely to hide. The word ‘sampling’ makes some lawyers uneasy, because they associate law with reading every word, and it is a skill that law schools do not teach. But the alternative to a sample is not a complete review; it is an unplanned and undocumented partial review, in which the documents that were read are those that were on top of the pile. A planned sample, with a stated method and a stated limit, is more honest and more informative. This module teaches the logic: how to divide the room into groups by risk, how to review the high-risk group in full, how to draw a random sample of the others, what a sample with no problems does and does not allow you to say (the ‘rule of three’ and the upper bound), what to do when a sample finds a problem and how to report coverage. It uses simple arithmetic and a Sampling Planner that works with invented groups of documents. The same method is used in the course Checking Legal AI Tools, where it is applied to testing a tool’s output.

General education on legal due diligence, using invented companies, contracts, people and figures. It states common practice in simplified form and does not describe any one country's law, which differs and changes; rules on disclosure, the effect of a buyer's knowledge, privilege, data protection, competition law and liability in particular vary. It is not legal advice, and it does not replace advice from a qualified lawyer on a real transaction.