← Insights

ChatGPT vs. your books: what to try, what to avoid

By Fred Lundin6 min read
Financial operations

Your AI-ready financial operations roadmap.

From manual bookkeeping to a structured, AI-assisted system for optimized cash flow, faster closes, and automated collections — with human oversight.

Foundational setup
Data intake & categorization.
AI-assisted bookkeeping setup.

AI suggests categories based on vendor maps; humans review and correct until you hit 90%+ accuracy.

Centralized "bills@" intake.

A single "bills@" email eliminates missing receipts and fragmented approval trails.

Smart approval thresholds.

Set specific dollar limits for new vs. existing vendors — control without bottlenecks.

Accuracy baseline
50
Test batch
≥90%
Correct categorization
85%
Accuracy to go-live
Strategic management
Forecasting & performance.
13-week rolling cash forecast.

Run Base, Optimistic, and Conservative scenarios to predict cash cushions accurately.

The 10-day monthly close.

Accelerate reporting with an 18-step checklist and AI-drafted variance commentary for owners.

4-tier automated collections.

Transition Soft → Friendly → Urgent → Firm reminders based on custom past-due day counts.

Collections cadence
Soft
Day 1
Friendly
Day 15
Urgent
Day 30
Firm
Day 45

Every week a client asks some version of this: "Should I just be using ChatGPT for [thing]?"

The answer depends on what "thing" is. Here's how I sort it.

Green light: things ChatGPT is genuinely great at

  • Drafting client emails, quote follow-ups, and thank-you notes
  • Summarizing long PDFs (leases, insurance quotes, vendor agreements) you paste in
  • Explaining a concept before you go into a meeting
  • Writing job descriptions, SOPs, and training material
  • Reformatting messy text into structured lists

None of this needs your actual books. It needs your prompt and your judgment.

Yellow light: things it can do, but with a governance discussion first

  • Categorizing bank feed transactions
  • Summarizing revenue trends
  • Drafting month-end memos

You can do these. But if the underlying data is a mess (see: "The 7 places your customer data is hiding"), you'll get confidently wrong answers. And if you're pasting customer lists or financial detail into a public model, someone should have decided that was allowed.

Red light: never, without a proper setup

  • Pasting your full customer database into a public chat
  • Uploading tax returns or W-2s
  • Feeding it your payroll data
  • Anything you'd be uncomfortable seeing on the front page of the paper

Not because AI is scary. Because free tiers of public models don't always give you the data-handling guarantees your customers assume you have.

The real move

For the yellow-and-red list, the answer isn't "don't use AI." It's "use the right AI." That usually means a paid tier with an enterprise data-handling policy, or a local/self-hosted model. Which one depends on your business.

That's the conversation. Not "ChatGPT vs. no AI" — "which AI, for which task, with what guardrails."

Ready to make your data actually useful?

Book a free 30-minute assessment. We'll walk through your current data situation, spot the highest-leverage fixes, and give you a clear path forward — no commitment.

No commitment · No technical jargon · 30 minutes