Claude for Small Business found 17 of 20 hidden P&L issues
Anthropic’s Claude for Small Business connected to Google Drive, Canva and Gmail to analyze a fictional seven-month P&L. In a test built with 20 planted problems, Claude found 17 in under six minutes and then created a Canva deck and email summary.
Anthropic launched Claude for Small Business this month, adding native connectors to business tools including QuickBooks, HubSpot, Canva and Google Workspace inside Claude Cowork. To see how well the system handled a real workflow, the source article tested it on a fictional financial review task: read a complex profit-and-loss report, summarize the findings, create slides in Canva and draft an email for coworkers.
The test used Google Sheets, Canva and Gmail. Before the run, the author built a seven-month P&L for a fictional software consultancy called Meridian Dev Studio and loaded it with nine tabs, twelve clients, twenty expense line items and 20 deliberately planted problems.
The hidden issues ranged from simple to forensic. Some were obvious once the numbers were examined, such as the company losing money every month for seven months and posting a cumulative net loss of $134,885. Others required more context, such as gross margin falling from 58% in November to 10.6% in March during the TerraLoop ramp, or January’s reported revenue of $112,080 including a $24,000 one-time late payment recovery from Client B.
The prompt asked Claude to read all nine tabs, write a plain-English executive summary, flag every anomaly or concern, identify clients or cost lines that needed immediate attention, note cross-tab inconsistencies and list questions a CFO would ask on first review. According to the article, Claude caught 17 of the 20 planted problems in less than six minutes.
It found all of the easy and medium-level items. From the harder group, it caught five of eight. The article says this included findings that one Claude account expected another Claude account to miss, which suggests the model handled more than just surface-level pattern matching.
But the misses mattered. The items Claude failed to spot included a ghost receivable behind a bad debt write-off, a churned client with no explanation in the notes, and a reimbursables discrepancy split across two tabs. The author also said Claude missed the most forensic signs in the sheet, including perfectly flat interest income of $180 every month for seven months, a bank fee drop in April that could indicate growing accounts receivable, and depreciation math on an equipment purchase that looked too clean to be real.
Those gaps are important because they are the sort of issues an accountant or finance lead would ask about immediately. The article argues that these patterns do not just require calculation; they require skepticism about why the numbers look unusually neat.
Claude also flagged five irregularities the author had not planted. According to the article, these included a commission plan based on bookings rather than gross profit, a designer cost increase with no explanation, conference spend peaking in the same month gross margin collapsed, an unflagged dependency on client referrals and even a typo in the Google Drive file name.
After the analysis, Claude built an 18-slide Canva deck and drafted an email summary with the deck attached. The author said the deck looked like a standard stock-image presentation and was not ready to send without editing, but it took only about three minutes to generate.
The Gmail step also picked up a small personalization detail. The account had the name Jessica but had been using Jess in recent emails, and Claude signed off using Jess.
The article concludes that Claude for Small Business saved a large amount of time on analysis and presentation work, but it still missed three important issues. The author says the tool is useful, but that a human still needs to review the output before relying on it.
