Classify expenses by category with Numerous.ai in Sheets

Classifying hundreds of expense lines by hand is the kind of work that takes a whole afternoon and still leaves inconsistencies. Numerous.ai adds a function to Google Sheets and Excel that sends each payment reference to a model and returns a category, so a period's expenses are sorted in about 20 minutes. The quality depends almost entirely on one thing you write before starting: a closed list of categories.

Prepare the export

Export the period's expenses with the payment reference and the amount. Remove account numbers and names of individuals from the column you will classify. Every cell's text goes to the service, and a payment reference like "salary advance to [name]" should become "salary advance" before it does. Keep the amount in a separate column; the model only needs the text.

Write the closed list

List your management accounting categories with a short description of each: "Rent: office and warehouse lease payments", "Logistics: carriers, couriers, fuel for delivery vehicles", "IT: software subscriptions, hosting, hardware under the capitalization threshold". Descriptions resolve the borderline cases, such as whether a delivery charge on an office supplies order is logistics or office costs.

Without the list, the model invents categories as it goes and names the same one three ways: "Transport", "Delivery" and "Logistics services" all appear in the same column, and the totals by category are meaningless.

The prompt inside the function

Role: you are an accountant assigning expenses to management accounting categories. Context: categories: {{list of categories with descriptions}}. Task: for the payment reference below, return exactly one category from the list. If you are unsure, return "other". Format: only the category name, no explanation. Payment reference: {{cell text}}

Put the list in a single cell and reference it from the formula, so every row uses exactly the same wording. "Only the category name" keeps the output clean enough to total with a pivot table.

Run a small batch first

Fill the function down a small batch of rows, look at the results and adjust the descriptions where the model was confused. Then fill the rest. Filling down thousands of rows uses the monthly limit in minutes, so a trial run protects your quota as well as your results.

Check and freeze

Filter the rows marked "other" and assign them by hand; they are usually the unusual payments that deserved a human look anyway. Then pick 20 random rows and check them. If more than one is wrong, refine the category descriptions and rerun those rows.

Finally, save the results as values. AI functions recalculate and may change a category when the sheet updates, and a report built on a live function can shift after you have sent it.

Keep the category list and the prompt cell in the workbook for next month. Using the same list every period is what makes the totals comparable over time; a list edited each month turns a trend into noise.

What to do

1. Export expenses with payment reference and amount, removing account numbers and names of individuals.
2. Write a closed list of categories with short descriptions.
3. Put the list in a cell and run the function on a small batch.
4. Adjust descriptions where the model was confused, then fill the column.
5. Assign rows marked "other" by hand.
6. Check 20 random rows for correctness.
7. Save the result as values and total amounts by category.

When it is done

Every row has a category from the closed list, the sample of 20 rows has at most one error, and category totals equal the overall total.

Pricing

Numerous.ai offers a trial number of requests after installing the add-on. Paid pricing depends on the number of requests; check the vendor's site. See Numerous.ai pricing, the auditor's version of this task and Numerous.ai vs SheetAI.