Learn how to extract selected information from invoices#


  1. Register for free on

  2. Create a Project

  3. Invite Members to the Project

  4. Upload new Documents - Users normally start with 10 documents.

Define AI model structure#

Setup the following structure in your project to extract basic invoice information, vendor information and detailed line item information from invoices. Feel free to change the proposed scope of the Label Sets and Labels in this tutorial.

graph LR subgraph Users in Project User_1 --- Project User_2 --- Project User... --- Project end subgraph Category Project --- Invoice[German Invoice] end subgraph Label Set Invoice --- Info[Invoice] Invoice --- Item[Line Items*] Invoice --- Vendor end subgraph Label Info --- Date Info --- VAT Info --- Total Item --- Description Item --- Price Item --- Subtotal Item --- Pieces Vendor --- Entity[Entity Name] Vendor --- VATID[VAT ID] end

* Use the multi Label Set option when creating the Label Set.

Train the first version of your invoice AI model#

  1. Split your documents into Status: Training documents and Status: Test documents.

  2. Annotate all documents according to your defined AI model structure using the SmartView.

  3. Review our Improve Extraction AI checklist.

  4. Train extraction AI

  5. As soon you receive an e-mail the evaluation is available online. How to properly classify the evaluation and what the figures mean can be found in the Key Features under AI Model.

Retrain or reevaluate your Invoice AI#

After you have trained your Extraction AI, you can start retraining the AI to increase its accuracy. Each retraining process improves the accuracy of your model.

  1. Upload other documents than the ones you already uploaded

  2. Once the OCR process has been completed, Feedback required via the SmartView to every yellow marked Annotation.

  3. Now you have two options:

What happens if the accuracy of the model decreases after retraining?

You will get full access to the model and the evaluation. However, only the best model will be used via API. The accuracy for all models can be updated via admin action using the latest Test data.

Use AI model#

As soon you are confident about the results, you can use your AI via our Integrations & API. The API of the project is available since the project has been created.