Free · No Gate · Use It Today

The Safe AI Input
Framework

What to share with AI at work. What never to share. How to stay safe and compliant inside a real corporate finance role — built and tested over six months of live use.

↓ Download PDF Version Get Full Framework + Prompts →

Tested inside a listed pharmaceutical company under real GDPR constraints and real deadline pressure. Six months. Live month-end cycles. Real CAPEX reviews. Nothing theoretical.

The Four Rules — Read Before Every Session
✓
01
Always Anonymise First

Replace project names with codes — PROJECT-A, PROJECT-B. Convert absolute figures to percentages or index values. Remove any identifiable supplier, client, or colleague reference. Do this before you write a single word of your prompt. The input is the risk, not the output.

✗
03
Never Paste Live Company Data

No ERP exports. No live spreadsheet tabs. No named projects, clients, or suppliers — on any device, including your personal laptop. Not into any AI tool. Not even into a session marked "private". The data protection obligation does not end at the office door.

✓
02
Always Validate the Output

AI reasons with the data you give it. It has no visibility into the conversations, decisions, or politics behind the numbers. You do. Read every output critically. Correct what it cannot know. Your judgment is always the final layer — not optional, not delegable.

✗
04
Never Send Output Without Review

The variance commentary, the investment narrative, the board recommendation — these are yours. AI produces structure and drafts. You own the judgment call behind every number that leaves your function. Review everything. Sign off on nothing you have not personally validated.

The Anonymisation Checklist — Run Before Every Prompt
✓
Company name
Replace with sector descriptor — "Listed Pharma Co", "Global FMCG", "Manufacturing Group"
✓
Project names
Replace with codes — PROJECT-A, PROJECT-B, PROJECT-C. Never use real project names.
✓
Absolute figures
Convert to percentages, variances, or index values. Remove £/$/ signs where possible.
✓
Client and supplier names
Remove entirely or replace — VENDOR-1, CLIENT-1, SUPPLIER-A.
✓
People names
Use roles only — "Finance Director", "Project Manager", "CFO", "Line Manager".
✓
Cost centre codes
Replace with broad category — CAPEX, OPEX, SGA, R&D. Remove site or division identifiers.
✓
Currency amounts
Convert to % of budget, % vs prior year, or index where possible. If absolute needed, round significantly.
✓
System or tool names
Replace ERP, BI tool, or system names — "ERP System", "Reporting Tool". No SAP, Oracle, Workday references.
Three Things AI Will Never Know About Your Finance Role
1
The Political Context

Why a project is really delayed. What the CFO actually thinks. Which variances will be challenged and which will not. AI sees numbers — you see the room. No prompt can change that.

2
The Six Months of Conversations

Every re-phasing decision, every stakeholder alignment meeting, every assumption that was quietly revised. That context lives in your head. Not in the data. Not in the anonymised input.

3
What Is About to Change

The strategic decision not yet announced. The budget reallocation happening next quarter. AI works with what it is given. You work with what you know is coming. That asymmetry is your value.

How to Validate Before You Use

Read every assumption out loud. If it sounds wrong, it probably is. AI states assumptions confidently regardless of accuracy.

Check every forward-looking statement. AI cannot know what will happen. Any projection it makes is pattern-based, not informed.

Verify the variance narrative. Does each explanation match what you know actually happened? Cross-reference against your own notes.

Add what AI cannot know. The political context. The conversation that happened last week. The decision that is not in any system yet.

Remove anything that identifies the business. Even after anonymisation, re-read the output for any detail that could point back to your company.

Own the final version completely. If you would not be comfortable explaining every line without AI, rewrite those lines yourself first.

Members Access — Full Framework

This is the one-page version.
The full methodology goes deeper.

The free framework gives you the rules and the checklist. Members get the complete system — every grey area mapped, every edge case addressed, the compliance defence built in, and the audit trail that protects your career.

Decision Tree
Every scenario mapped — what to share, what to anonymise differently, what to never attempt. No grey areas left unresolved.
The Compliance Defence
Exactly what to say if your manager, your auditor, or your IT department asks how you used AI to produce your work.
Audit Trail System
The month-end context memory prompt. Log what AI found. Carry it forward each cycle. Stay protected and consistent.
335+ Safe Prompts
Every prompt built with this framework already applied. Safe input guide embedded in each one. Tested across real finance functions.
Get Full Access — Subscribe Free →