The Two-Tier Notebook Architecture
How to organise your AI audit infrastructure so it scales from one engagement to your entire audit function.
Engagement Notebooks
One notebook per engagement. All sources uploaded progressively throughout fieldwork. Source selection (ticking) controls which documents are active for each query.
📄 Walkthrough transcript 📄 POS data
📄 Vendor invoice sample 📄 Management responses
Annual Risk Universe Notebook
One notebook for the full audit function. Fed by Briefing Docs exported from completed engagement notebooks. Use for cross-engagement synthesis and CAE reporting.
📄 Q1 F&B Audit Briefing Doc
📄 Q2 IT Audit Briefing Doc
📄 Annual risk register 📄 Board risk appetite
Source Selection = Audit Scope Control
In NotebookLM, you tick which sources are active before each query. Only ticked sources contribute to the response. This is your scope control. Document your source selections in your work papers: 'AI-assisted analysis conducted using [Source 1, Source 2, Source 3].' This forms part of your audit evidence trail.
Data Privacy FAQ
CAE Briefing Template
INTERNAL MEMO — AI AUDIT TOOLS BRIEFING
Proposed: NotebookLM Integration into Audit Workflow
Purpose: To brief the CAE on the proposed use of NotebookLM as a source-grounded AI research tool to improve audit efficiency and documentation quality.
What it does: NotebookLM allows auditors to upload audit documents and query them using natural language. Every response is cited back to the exact source passage.
Data governance: Documents remain within Google's infrastructure. Google does not use notebook content for AI training. Compatible with our existing data protection obligations.
Proposed use cases: Risk universe extraction, audit program generation, walkthrough gap analysis, cross-finding synthesis.
Estimated efficiency gain: [X] hours saved per engagement. Full methodology and data privacy FAQ available on request.