Skip to main content
All CollectionsTreasuryCategorization
How to Simplify Transaction Categorization with the Pre-Categorization Algorithm
How to Simplify Transaction Categorization with the Pre-Categorization Algorithm

Categorizing can be time-consuming. The Pre-Categorization Algorithm simplifies automatically suggests categories based on your past actions

Updated over a week ago

What Is the Pre-Categorization Algorithm?

The Pre-Categorization Algorithm is a smart tool that predicts the appropriate category for each transaction. It learns from the names of your previously categorized transactions, including those you've:

  • Manually categorized

  • Imported with a category already set

  • Categorized using a categorization rule

  • Validated from previous pre-categorizations

By learning from these actions, the algorithm continually improves its suggestions, saving you time and effort.

How Does It Work?

The algorithm operates by analyzing the labels of transactions you've categorized. It separates inflows and outflows to enhance accuracy. The learning occurs whenever you categorize a transaction, and the knowledge is applied during:

  • Bank Synchronization:

    • When new transactions are synced, the algorithm uses insights from the last 150 pre-categorized transactions to suggest categories for all synced transactions.

    • Knowledge is shared across your company and isn't specific to individual bank accounts.

  • Manual Categorization:

    • Each time you manually categorize a transaction, the algorithm learns from 7,000 transactions randomly selected from the last 70,000 categorized transactions.

    • This helps categorize the next 140,000 uncategorized transactions.

  • Transaction Import:

    • During imports, the algorithm learns from the last 300 categorized transactions.

    • It applies this learning to all uncategorized imported transactions.

Managing Pre-Categorized Transactions

When you encounter a pre-categorized transaction, you have three options:

  1. Validate the Suggestion

    • Go to the Bank section.

    • Review the suggested category on the transaction.

    • If it fits, select the validate symbol next to the category's name to confirm.

    • This action adds the transaction to the algorithm's memory, secures the category, and prevents further changes by the algorithm.

  2. Change the Category 💡

    • Click on the transaction.

    • Choose a different category from the list.

    • The algorithm will learn from this correction for future suggestions.

  3. Do Nothing

    • Leave the transaction as pre-categorized.

    • It remains in the suggested category in your cashflow plan.

    • ⚠️ The category might change if the algorithm finds a better match or if a rule applies (applies to the last 50 pre-categorized transactions by creation date).

Tips 💡

  • Create Categorization Rules for Ambiguous Transactions:

    • If the algorithm isn't accurately categorizing certain transactions due to ambiguous labels, numbers in names, or varying spellings, consider setting up categorization rules to improve accuracy.

  • Regularly Validate or Correct Suggestions:

    • By validating or correcting pre-categorized transactions, you enhance the algorithm's learning process, leading to better future suggestions.

  • Understand the Algorithm's Limitations:

    • The algorithm may struggle with:

      • Labels lacking relevant information (e.g., transaction details not in the label)

      • Labels containing irrelevant words (e.g., "SEPA", "cheque")

      • Names with numbers (e.g., suppliers or clients with numerical names)

      • Different spellings or abbreviations of the same name (e.g., "CPAM", "C.P.A.M.")

    • In such cases, setting up rules or manually categorizing multiple instances may be necessary.

FAQ ❓

Can I disable the Pre-Categorization Algorithm if it's not beneficial for my business?

Yes, you can disable the algorithm if it's not meeting your needs or causing incorrect categorizations.

What happens if I delete a transaction?

Deleting a transaction removes any learning the algorithm gained from it. If you reconcile an expected transaction with a paid one (which deletes the expected transaction), the algorithm will forget any associated knowledge.

Is the algorithm affected by typos or minor label differences?

The algorithm is sensitive to small and rare typos, which may affect its ability to suggest accurate categories. Creating categorization rules can help mitigate this issue.

Did this answer your question?