Back to Blog
Case Study February 2026 5 min read

I Asked My AI to Double-Check My Bookkeeper. Here Is What It Found.

My bookkeeper has 12 years of experience and confirmed the books were ready for BAS. I asked my AI Employee to take a second look. It found $5,936 in wrong categories, $142 in dodgy GST claims, and 38 missing receipts.

20
Wrong categories
$142
Incorrect GST
38
Missing receipts
1
Question asked

What Happened

BAS was due. The accountant had sent a third reminder. My bookkeeper confirmed everything was clean and ready for lodgement.

I opened a chat with my Agentive AI Employee and typed:

"Can you double check all code assignment and GST allocation? Double check my bookkeeper."

Within minutes, the AI had pulled every transaction from Xero, cross-referenced each one against the chart of accounts and GST rules, and come back with a list of problems.

Problem 1: $5,936 in Wrong Categories

Multiple transactions were sitting in a "Suspense" account, which means they had never been properly classified. Others were in generic categories instead of the correct specific ones.

Examples of What Was Wrong

Vendor Amount Was In Should Be
AWS$600SuspenseCloud Infrastructure
Google Ads, LinkedIn$141SuspenseAdvertising
Lovable, OpenRouter$203SubscriptionsAI Subscriptions
Lawpath$4,442ConsultingLegal Expenses

20 transactions. $5,936.79 total.

Why does this matter? Because when AWS charges are filed under "Suspense" and Lawpath under "Consulting", your P&L tells you nothing useful. You cannot see where the money is actually going.

Problem 2: $142 in Incorrect GST Claims

This one could have cost me with the ATO. The bookkeeper had claimed GST credits on US vendors that do not charge GST. Anthropic (Claude.ai), X Corp (Twitter), Meetup: all US companies, all incorrectly marked as "GST on Expenses".

Vendor Country Wrong GST Claimed
Claude.ai (x4)USA~$112
X Corp (Twitter)USA~$11
MeetupUSA~$9
LinkedIn (x2)USA~$0.18

~$142 per quarter. ~$568 per year. Enough to trigger an ATO review over time.

The kicker? The bookkeeper had correctly marked Anthropic as GST-Free, but then marked Claude.ai (same company, different billing name) as GST on Expenses. The AI caught this because it understood they are the same vendor.

Problem 3: 38 Missing Receipts

38 bank transactions over five months had no invoice or receipt attached. Under Australian tax law, you need records to support your BAS claims.

The AI searched my email, found matching invoices and receipts, generated PDFs from email confirmations where no formal invoice existed, and attached all 38 to Xero. Automatically. Done.

The Part No One Talks About: Email Access

To find those 38 receipts, the AI searched through my personal and work email accounts. It matched invoices to bank transactions by vendor name, amount, and date.

Think about that for a second. Would you give your bookkeeper access to your email? Your inbox has client conversations, contracts, personal messages, sensitive business information. No chance.

Would you give a virtual assistant access? Also no.

But I could give it to my AI Employee, because it runs on my own infrastructure. Hosted in Australia on AWS. Behind enterprise-grade firewalls. None of my data is used for model training. None of it is sold to third parties. The AI works within my own secure environment, reading only what it needs, when it needs it.

That is the unfair advantage. The AI can see everything: Xero, email, invoices, vendor communications. It connects dots across systems that a human would never be given access to. And it does it without any of the trust, privacy, or security risks.

Why I Could Trust the AI With Email Access

Hosted in Australia on AWS, subject to Australian privacy law
My infrastructure, behind AWS enterprise firewalls
No data training: my data never trains AI models
No third-party sharing: nothing is sold or shared

Why This Matters for You

My bookkeeper has 12 years of experience and handles multiple clients. These are not unusual mistakes. They are exactly the kind of errors that happen when someone is juggling many clients under deadline pressure:

  • Card transactions auto-imported to a holding account and never reclassified
  • US vendors mistakenly treated as Australian for GST
  • New SaaS tools that do not match existing categories
  • Receipts scattered across email accounts that no one has time to search

This is not about replacing the bookkeeper. I still need one. But now I can verify the work myself with a single question. I do not need to understand debits, credits, or GST rules. The AI handles that.

What the AI Did in One Session

Audited every transaction against the chart of accounts
Found 20 miscategorised transactions ($5,936)
Caught $142 in wrong GST claims on US vendors
Searched email for 38 missing receipts
Generated PDFs and attached all 38 to Xero
Created a correction spreadsheet for the bookkeeper

Want an AI That Double-Checks Your Books?

Connects to Xero. Reads your emails. Catches what humans miss. Deploy in 24 hours.

Read more about our AI BookkeeperAll finance use cases