How DOKKA Learns
DOKKA combines two core technologies:
Artificial Intelligence (AI) – which allows the system to learn from the actions of accountants and improve over time
Optical Character Recognition (OCR) – a technology that identifies characters in a document based on the pixel patterns within the document
Thanks to this combination, DOKKA can learn where information is located in a document and automatically extract it in future uploads.
However, it’s important to know that not all fields can be learned, and there’s a correct way to teach the system.
How Does DOKKA Work?
1. Smart Learning from Users – AI
DOKKA observes the actions performed by users (such as dragging text into fields), and learns where to extract each type of data next time.
2. Recognizing Text from Pixels – OCR
Each document uploaded to DOKKA is broken into millions of tiny pixels (invisible to the naked eye).
The system detects which characters are located in which parts of the document, which enables:
Dragging text from the document into the financial fields (instead of manual typing)
Learning where information is located so it can be extracted automatically in the future
What Is DOKKA’s Learning Based On?
DOKKA learns based on the vendor and the layout of the document.
Every invoice uploaded into the system receives a unique Layout ID.
If DOKKA recognizes the same vendor and the same layout, it will apply what it has learned in the past.
When Does Learning Actually Happen?
Learning in DOKKA happens only when you click the “Publish” button.
Publishing doesn’t just create a new invoice in your ERP — it also saves the way you completed the fields as learning data for the system.
But there’s something important to understand:
The moment DOKKA extracts the information from the document into the fields — is the moment the document is uploaded to the system.
You can actually see this in action — when you upload a document, DOKKA performs a brief "thinking" process.
At that stage, it does a one-time analysis of the file, detects the vendor and layout, and fills the fields based on what it already knows.
What Does That Mean?
If you upload two invoices from the same vendor with the same layout, but only teach the system after both have been uploaded —
The second invoice will not reflect what the system learned, because DOKKA does not re-analyze documents that are already in the system.
What Should You Do in This Case?
To apply the learning to a document that was already uploaded:
Download the document to your computer
Delete it from DOKKA
Re-upload it — DOKKA will now recognize the vendor and layout, and apply the new learning automatically
How to Teach DOKKA Correctly
You must drag the text from the document into the desired field (do not type manually)
Only dragging allows DOKKA to know where the data came from
Which Fields Are Not Learned?
Some fields are intentionally not set to auto-fill, such as:
Part numbers
Product descriptions
Invoice Details
Custom fields pulled specifically from your ERP system
Additionally, fields like Account Number are not learned via dragging.
In such cases, DOKKA will automatically use the same value used in the last invoice for that vendor.