Under the Hood: Rossum Product Improvement News
August 2019

Every month, we share the newest and coolest features of Rossum’s document processing application, the underlying platform and the AI engine. Scroll down to see all the key features introduced in August 2019!

You can see a comprehensive list of the change log from July and all other updates from 2019.



You can reset your password

If you have forgotten your password, you can now easily reset it with the web login form. An authentication email is sent to confirm your identity. Password strength is validated interactively during the password reset.

Password reset

Receipts are automatically detected

Our AI can autodetect receipts thanks to the new invoice_type class “receipt”, assigned to receipt documents.

You can extract data from initial document pages

We are introducing a new effective_page_count queue parameter. In some cases, the invoice is quite long, due to various additional attachments. To simplify the data extraction, the data can be set to always extract from the first page or first two pages. The queue configuration now has an additional rir_params field which may include an effective_page_count parameter limiting the pages.

Improved accuracy of extracted documents

We have tweaked the table separator (slicing) algorithm which produces more accurate boundaries. The postprocessing component did not optimally interpret the neural network predictions before. This squeezes out a couple extra percent for the accuracy!


AI prefills fields with candidates

If a schema field is configured in the “multivalue” mode, it is now prefilled by all candidates identified from the AI core engine. The AI core engine can identify multiple candidates for a particular field value and normally, only the top-scoring value is used.
Let’s see the example of ‘ order_id ‘ field. The AI core engine may not lock onto the appropriate choice when multiple legitimate candidates are available. However, all candidates may be captured in the multivalue, and an extension may choose the correct one automatically (e.g. using regular expressions or database matching).

An experimental “status change” webhook

An external webhook can be called every time the status of an annotation changes. The first webhook ties to the “changed annotation_status” queue event action and is called every time an annotation status changes.

We fixed the order of AI fallbacks applied

AI fields associated with a datapoint were applied out of order.   
– The ‘ rir_field_name ‘ attribute of a schema is supposed to list AI fields prefilled in the datapoint, in order of priority.
– E.g. ‘“rir_field_names”: [“amount_due”, “amount_total”] ‘ should mean that “Amount Due” is prefilled, preferably with “Amount Total” as the fallback if “Amount Due” is not found.
– This logic was mistakenly omitted, causing the fields to be applied in random order rather than by priority.

Elisctl is upgraded

The elisctl tool got an update – now you can bulk update user-queue permissions with the new ‘ elisctl user_assignment ‘.

Important notice

As of Oct 1, we plan to modify the values pre-extracted in the currency field from lower-case to upper-case. E.g. “EUR” will be returned instead of “eur”. We recommend that extensions and downstream integrations treat currency values as case-insensitive.

User Experience Updates


The app is faster

Interactive work in the app could get sluggish in long documents full of tables. Working with big tables means the app has to juggle many more data points (even if they are not all visible at that moment). We have improved the speed for these cases by numerous optimizations.

You can work with tables easier

We have improved our magic grid tool. When drawing a new magic grid, column separators are placed intelligently on the detected text. Furthermore, it is now possible to fully edit cells that are not bound to an on-page box.

Pressing “Enter” when choosing column type does not exit magic grid mode anymore.

Magic Grid - Line items extraction, tabular data
The magic grid tool in action

Navigation and terminology are more intuitive

In August, we have improved the way users can get around the app. There is a new navigation system with a unified top-screen bar which allows easy navigation across different views.
The dashboard sidebar now contains to-review document counts for workspaces and queues.
Users without Rossum directly hooked to their ERP system found the word “export” (and tab “exported”) confusing. We have therefore renamed the document finalization action from “Export” to “Confirm”. The API remains unchanged for now, the annotation action is still “export” and status remains “exported”. In the long term, we plan to introduce an additional status “confirmed” in the API, aimed at setups without a /save connector hook.

User interface updates

Dashboard sidebar now contains to-review document counts

Do you want to keep in touch and get our change log announcement? Register for our API announcement group.

We would love to hear feedback on the updates introduced in August. Send your comments to rossum@rossum.ai.

Try all those new features for yourself. Sign up for a trial!

Ready to get started?

Make a quantum leap in your document processing approach. Boost accuracy and effectiveness with an AI-powered data capture solution for all documents.