Architecture review checklist
Work out where your documents go at every step of an enterprise AI deployment, who holds them there, how long they stay and what deletion removes.
Updated 21 Aug 2026
Definition
A useful data-flow review names the data, processor, location, purpose, retention, access and owner at every stage.
List every place a copy of your content comes to rest, from the original file to the logs and backups.
For each class, record service/operator, location, encryption, identity, network path, retention, deletion and whether data can enter training or human support workflows.
Test onboarding, role change, revoked source access, departed users, connector deletion and organisation separation through retrieval and generated output.
The successful request flow is the easy diagram.
Draw managed, customer-cloud, self-hosted and any local-model variant separately. Do not reuse one diagram when the processors or responsibilities differ.
Assign an owner and disposition to every unknown: provide evidence, change configuration, accept risk, add a control or reject the design.
Pick a real piece of work, agree what a good answer looks like, then go through the results together.