Assigning attributes such as Cost Centers, Subcategory and other to transactions follows different rules, patterns and practices in each company. Our Smart Automation utilizes AI capabilities to automatically pre-code the majority of your transactions for you.
Here is how we do it:
Learning from your own data:
Our system automatically reviews what accounting attributes you assigned to exported transactions.
Building an AI model for your organisation 🤖:
Once you have exported your first 150 transactions, we automatically train an AI model for your organisation based on your specific data - this model analyses how your expense accounts, VAT rates and others were assigned to transactions previously, based on which it predicts the correct accounting details
Prediction with AI 🤖:
For all future transactions our models then try to predict VAT, Subcategory and other dimensions. This prediction is probability based: if the AI system isn't confident in its prediction, it will not guess. Instead, it will leave the field empty, prompting you to fill it in manually. This approach ensures that the system only provides predictions when it is likely to be correct, minimizing errors.
Our data show that the predictions done by Smart Automation significantly outperform the accuracy of default accounting rules that you can set up. For this reason, if a transaction can be assigned an attribute using either a rule or Smart Automation, we will by default assign the attribute using Smart Automation. You can turn this behavior off on your settings page.
Self-learning with AI 🤖:
Our company’s AI model continuously improves through a feedback loop. If the system makes an incorrect prediction, you can correct it, and the AI will learn from this correction. This ensures the AI becomes more accurate over time.
Every time you export your transactions we build a new fresh model for you taking all your previous corrections into account.
When you delete a specific expense account or VAT rate we build a new model so we don’t predict it anymore.
When does this not work and what you can do about it
While our system is designed to be highly accurate, there are a few scenarios where it might not be able to predict correctly:
New or rarely used expense account or VAT rate:
Like all AI applications, we need data to make accurate predictions. If the value you want to predict does not appear in your data or is rare (less than 10 transactions) it’s unlikely that we would make an accurate prediction, hence we don't consider them.
Equal transactions are linked to different accounts:
When you have multiple transactions that look identical but are linked to different accounts the AI is not able to detect a pattern and is likely to pick the value you selected most often.
Lack of confidence in predictions:
If the system isn't confident enough in its prediction, it will leave the fields empty rather than risk an incorrect match. This lack of confidence typically arises from insufficient data, ensuring that your supplier accounts only include complete and correct data.
Note: For Smart Automation to make accurate predictions, historical data is crucial. The prediction models are (re)trained based on the data exported to your accounting system. Therefore, assigning an expense account or VAT rate without exporting the transaction will not enhance future predictions. Ensuring consistent and accurate data exportation is essential for the continuous improvement of the AI model.