How AI Categorization Works

How AI Categorization Works

Propio uses AI to automatically suggest categories for your transactions. The AI learns from your calibration choices and improves over time. Suggestions come with confidence levels (high, medium, low) so you can review accordingly.

Overview

Propio's AI categorization engine analyzes your client's transactions and automatically suggests the most appropriate category from the Chart of Accounts. The AI uses transaction details — including the description, amount, vendor, and historical patterns — to make its suggestions. Each suggestion includes a confidence level so you know how certain the AI is about its recommendation. Over time, as you accept, reject, and calibrate suggestions, the AI learns your preferences and becomes more accurate for each client.

Steps

1. Understanding AI Categorization

When new transactions are imported into Propio — whether from a connected bank account or a bank statement upload — the AI categorization engine automatically processes them. Within moments, the AI assigns a suggested category to each transaction based on available data. No manual action is required to trigger this process; it happens automatically in the background.

2. Confidence Levels Explained

Every AI suggestion comes with a confidence level that indicates how certain the AI is about its recommendation:

• High confidence — The AI is very confident in this suggestion, typically because it has seen similar transactions categorized consistently in the past. These suggestions are usually correct and can often be accepted without further review.
• Medium confidence — The AI has a reasonable suggestion but isn't fully certain. These transactions are worth a quick review to verify the category is correct.
• Low confidence — The AI is uncertain about the best category. These transactions require your attention and manual review to assign the correct category.

3. How Suggestions Appear on Transactions

On the Transactions page, AI-suggested categories are displayed alongside each transaction. You'll see a visual indicator showing the confidence level of each suggestion. Transactions with AI suggestions that haven't been reviewed yet are clearly marked so you can prioritize your review accordingly — start with low-confidence suggestions that need the most attention, or quickly approve high-confidence batches.

4. Accepting or Rejecting Suggestions

For each transaction with an AI suggestion, you can either accept the suggested category (confirming it's correct) or reject it and assign a different category manually. When you accept a suggestion, the transaction is categorized and the AI registers your confirmation as positive feedback. When you reject a suggestion and choose a different category, the AI learns from that correction for future transactions.

5. How Calibration Improves Accuracy

Calibration is the process of training the AI to understand your client's specific categorization patterns. When you first set up a client, Propio guides you through calibrating a set of sample transactions. Each calibration choice teaches the AI about your preferences for that client. The more transactions you calibrate and categorize, the more accurate the AI becomes. Over time, you'll notice the proportion of high-confidence suggestions increasing as the AI learns your patterns.

6. Use Check Images for Better Suggestions

For check transactions, the AI can analyze check images to extract additional context such as the payee name, memo, and other details. This additional information helps the AI make more informed category suggestions. Make sure check images are available in your imported bank data for the best possible AI performance on check transactions.