Agree the issues before the sources
The Disclosure Review Document works when the issues for disclosure are tight and the models are matched to them. It fails when a broad Model D is agreed across every issue because nobody had the data to argue for anything narrower.
You cannot argue for narrower without numbers, which is why the collection scoping exercise should come before the first draft goes over to the other side.
Sample first, promise second
A one day sampling exercise across the likely custodians tells you how many documents each keyword returns, how much of the mailbox is newsletters and how much is genuinely responsive. Those numbers turn a disclosure negotiation from an argument into a calculation.
It also protects you later. If the volumes move, you have a baseline that shows the change and supports a variation.
Chat is where the volume hides
Teams, Slack and WhatsApp now carry the decisions email used to carry. They also multiply document counts, because every message can become a document. Agree the unit of disclosure for chat early: a message, a day of a channel, or a thread.
Technology assisted review is not a shortcut past scoping
Predictive coding earns its place on large sets, and the courts have accepted it for years. It does not rescue a badly scoped collection, because a model trained on the wrong population simply prioritises the wrong documents faster.