Classify Invoices using AI

Hello, and welcome to this demonstration of Oracle Fusion Cloud Sustainability. In this session, we will explore a new AI capability that has been added to invoice classification. This new feature is introduced in Release 25B.

Oracle Sustainability can now automatically classify invoices that are not covered by predefined classification rules. Previously, when invoice distributions could not be classified using existing rules, Oracle Sustainability generated activities based on the default activity type specified in a profile option. These activities required manual review and classification before being posted to the sustainability ledger.

Depending on your invoice data, it can become increasingly challenging to write invoice classification rules that can cover all invoices in your system. But efficiencies can only be achieved if a maximum number of invoices are classified without human intervention. To address this, Oracle Sustainability now offers optional AI driven capabilities to classify invoice distributions that do not match existing rules. This enhancement improves classification coverage without the need to create and maintain complex rules.

AI powered classification leverages machine learning and generative AI to classify invoices. To train the machine learning model, we use historical activity data generated from activities and post it to the sustainability ledger. When sufficient data is unavailable, we use generative AI for this prediction.

When running the generate sustainability activities from payables invoices schedule process, we first evaluate exclusion rules and exclude matching invoice distributions. We then evaluate classification rules. And any matched invoice distributions will result in activities with the activity type and activity type attributes specified by the rules.

The unmatched distributions will now go through AI classification. If any of these invoices can be classified with high confidence, the corresponding activities will be generated using the classification predicted by AI. The rest of the invoice distributions will use the default activity type for invoice classification.

Once the activities are created, they can be reviewed on the sustainability activities page. Before proceeding with the demo, here are some key considerations. The effectiveness of AI based classification depends on the number of possible classification categories. The number of emission factor mapping combinations plays a critical role. Load only those emission factor mappings that you intend to use.

The second aspect is data quality. AI predictions rely on historical patterns and meaningful textual data from invoices. Then the third is about incremental learning. AI learns from posted activities in the sustainability lecture. If incorrect classifications exist in historical data, the model may learn incorrect patterns affecting prediction accuracy. While only high confidence predictions are applied automatically, AI generated classifications should still be reviewed before posting these activities to the ledger.

Now let's see AI driven invoice classification in action. I am now in the Sustainability Activities page. To filter and see all AI classified activities, all I need to do is to open the filter panel. And in the Classification section, I need to toggle the classified by AI, switch, and then click here to see the results.

Here are all the activities that were classified by AI. Now you can drill into a specific activity and you can also see that this activity was classified by AI based on this token that is visible here. If an activity is incorrectly classified, you can take one of the following actions.

If the activity type is correct but attributes are incorrect, edit the activity and correct the activity type attributes. If the activity type is incorrect, create a rule to classify similar distributions correctly in the future. Then delete the incorrectly classified activity. The next scheduled run will generate a new correctly classified activity.

This concludes our demo. In summary, we have explored how AI enhances invoice classification in Oracle Fusion Cloud Sustainability, reducing the need for complex rules and manual intervention. Over time, AI improves accuracy by learning continuously from the data. Thank you for joining this session.