From this article you will learn:
- How does automation work in Rossum
- What do the various ticks on the validation screen mean
- What are the document automation levels
- What are field automation levels
- What are the automation blockers
How does automation work in Rossum
Ideally, all documents in the world should be processed automatically. Rossum tries to help with this by accurately capturing data from documents and creating an automation framework. As a result, it allows you to automate documents where Rossum is confident of the captured values.
When a document successfully passes the automation pipeline without any human intervention, you will see a green icon with a capital A next to the file name:
Thanks to this icon, you can quickly find out whether a document has been automated or not.
What do the various ticks on the validation screen mean
You have probably already noticed the green and grey ticks displayed on the validation screen. These are indicators of various automatic document validation checks. In this section, we will explain more what they mean and how they work.
Fields validated by AI
When you open a new document in Rossum, you will notice that some fields have a grey tick next to them. It means that the AI has already validated and confirmed these fields. Rossum’s AI Engine currently has several ways to determine whether a captured value is correct.
Automation of fields in Rossum depends on several different factors, also known as validation sources, which include:
- Confidence scores – these show how confident the AI is that the value it found is correct. Rossum marks values with sufficiently high confidence levels with a grey tick.
- Built-in checks – we perform Data Integrity checks based on values found on the document. You can find more information on built-in checks here.
- History-based checks – if you have already processed documents for a specific vendor, Rossum might validate the captured data against the exported data from the same vendor. You can find more information on history checks here.
- Custom validations – Rossum can be easily extended with a custom check. Learn more about it here.
To find out the field’s validation source, hover over the tick and you will see the information:
Fields validated by User
When you leave a field in the validation screen, a green field appears. In addition, if you hover over the check mark, the User key will also appear among the checks passed on the field.
Most of the time, Rossum users need to ensure that all values captured in the document are 100% correct, so it only makes sense to highlight these values for the User. Seeing the values validated by the User makes it easier to determine if all fields have been reviewed.
Fields that need validation
To zero in on the most problematic fields, check those that lack a tick next to the field label and are written in white.
If you have confidence in Rossum’s AI, you can proceed to the following fields by pressing ENTER.
Document Automation Levels
There are three different automation levels that users can choose. Once you select one of them, Rossum will apply it to all the documents in the queue.
[Please note that the document automation levels settings are available only for Admin accounts.]
We recommend Confident automation for most cases. The default confidence threshold is 97.5%. In practice, this will still mean that most (if not all) documents require manual checks.
You can change (lower or increase) the threshold. However, you should be aware that the lower the threshold, the higher the number of potential mistakes in the captured data.
We train on human input (corrections), not previous AI predictions. Otherwise, there would be a feedback loop. Since only manually checked documents are used for training, you should avoid this automation level if you have a dedicated AI engine.
This automation means that almost no documents will require manual checks, leaving very few documents for the AI to learn from.
This level of automation ensures that a user will check each document before it is moved to the Confirmed or Exported tab.
Field Automation Levels
If you select the Confident automation, documents will get automated when all the fields are automatically validated. You will see a grey tick next to them on Rossum’s validation screen when they are. The fields can get a grey tick if:
- The confidence score of the field is higher than the confidence threshold set in the Extraction schema editor for the given field.
- A data integrity check is fulfilled for the given field.
- The field’s value was seen in the history of the given vendor.
As automation of the fields by the confidence score is Rossum’s main automation component, you can easily set the confidence score threshold for all the fields on the given queue, as shown below.
Suppose you set the automation level in a queue to Confident or Always, and some documents do not get automated. In that case, the automation blockers stop the process.
On the validation screen, the blocker icon is a red “A.” If you have Admin or Manager rights, you can select the option to show automation blockers at the top of the validation screen.
Check some of the most common automation blockers:
When you hover the mouse cursor over the grey check mark or the red “A” (if you choose to see automation blockers), you will see the prediction’s confidence score for that specific value.
The field’s confidence score can be above or below the “Confidence score mean” that you can see on the Automation tab for that field (last column):
To lower the threshold for a specific field, go to the Automation tab and edit its threshold (under the Score threshold column in the table above).
Built-in checks failed
This automation blocker will appear on amounts, dates, or line items.
If the validation source is a passed built-in check, it will be referenced as Data integrity:
You can read more about built-in checks in this article.
Rossum will also stop the automation when there is an error. Some possible errors you may encounter:
- Value for a required field is missing,
- An extension is not working,
- There is a date or number formatting issue.
This automation blocker appears when the field is required, but there is no value predicted (even if there is rir_field_name for that field).
You will find more information about other automation blocker types in our API documentation.