Free Preview: Chapter 4

If there's one task that consumes more management time than any other, it's reading. Reports, proposals, contracts, consultation papers, board packs, regulatory filings — the modern manager is buried in documents. Most of them are longer than they need to be. Most of them could be summarised in two pages. And most managers spend their evenings and weekends catching up on the ones they couldn't get to during the day.

AI changes this equation entirely. Not by replacing your judgement — you still need to make the decisions — but by doing the heavy lifting of extraction and summarisation, so that your reading time is spent on what actually matters.

What AI Can Do With Documents (And What It Can't)

Modern AI assistants can process uploaded documents — PDFs, Word files, Excel spreadsheets, PowerPoint decks, even images of text. They can:

• Summarise a 200-page document in seconds, at whatever length you specify
• Extract specific information — "pull out every mention of budget figures in this report"
• Compare two or more documents to find differences, contradictions, or gaps
• Answer questions about a document's contents without you having to read it first
• Translate dense technical or legal language into plain English
• Identify missing information — "what should this report cover that it doesn't?"
• Restructure content — "take this board paper and turn it into a 10-slide presentation outline"

What AI cannot reliably do:

• Guarantee 100% factual accuracy. AI sometimes hallucinates — it invents plausible-sounding but incorrect information. Always verify critical facts against the source document.
• Exercise professional judgement. AI can tell you what a document says. It cannot tell you what decision to make based on it.
• Handle very long documents perfectly. While context windows are generous, the AI's attention is not uniform across the entire document. Information in the middle of a very long document may receive less "focus."
• Read scanned documents with perfect accuracy. If a PDF is a scan of a physical document, OCR quality varies. Test with a page before committing to a full document.

The Document Analysis Workflow

Here's a systematic approach that works for virtually any document:

Step 1: Upload and Summarise

Upload the document and ask for a structured summary. This gives you layered understanding — you can read the one-paragraph version to decide if the document is even relevant, then drill deeper as needed.

Step 2: Ask Targeted Questions

Now that you have the landscape, zoom in on what matters to you. Ask about financial implications, required actions, risks, or assumptions.

Step 3: Request Plain-English Translation

If the document is legal, regulatory, or technical, ask the AI to rewrite key obligations in plain English while preserving legal substance.

Step 4: Generate Next Steps

Ask the AI to draft an action plan based on the document — what needs to happen, who should do it, and the recommended timeline.

Worked Example: Analysing an FCA Consultation Paper

Let's walk through a realistic scenario. You're a compliance director at a mid-sized financial services firm. The FCA has just published a 180-page consultation paper on proposed changes to consumer duty requirements. You need to brief the board by end of week.

In under 30 minutes — mostly spent reading and verifying the AI's output — you've produced a board-ready paper. Without AI, this would be a two-day task.

Comparing Documents

One of AI's most powerful and underused capabilities: uploading two or more documents and asking it to compare them. Use it for contract review, policy alignment checks, and competitor analysis.

Privacy: What You Should Never Upload

This is critical. When you upload a document to an AI tool, you're sending it to that company's servers. On free and consumer plans, the AI provider may use your uploaded documents to train their models. Never upload documents containing personal data, commercially sensitive information, confidential client data, or legally privileged material on these plans.

Business and Enterprise plans offer contractual data protection. Copilot (within Microsoft 365) keeps data within your existing tenant.

The "Don't Upload" Checklist:

• Personally identifiable information (names, addresses, NI numbers, etc.)
• Commercially sensitive financial data
• Client-confidential documents
• Legally privileged communications
• Trade secrets or intellectual property
• Anything covered by an NDA

A practical rule: If you wouldn't email the document to a stranger, don't upload it to a free AI tool.