Showing results for 
Search instead for 
Did you mean: 

How to enable ML model auto training

Hi Team,

We have a Batch Type with ML enabled. We have trained multiple documents, but under the Capture Models panel, it keeps showing N/A in Last Trained section. How can we set Decipher to automatically keep training the ML model as the user verifies the documents?

Basically, we want to have the ML model keep learning as each invoice is verified in the Data Verification tab by user. How to achieve this, please advise.



If I was of assistance, please vote for it to be the "Best Answer".

Thanks & Regards,
Tejaskumar Darji -
Technical Lead


Hi Tejaskumar,

You generally don't want the Capture model to update after every document you process, for a few reasons:

  1. It uses a lot of CPU to create/update a model and will slow down your document processing,
  2. A single document won't massively improve the model's accuracy,
  3. The natural of the Capture model's ML engine is that every document being processed will be influenced by the training of every other document, so different document formats will impact one another's extraction. This is a great way to extract data from various document formats, but only if the mode has been trained on a substantial number of documents. The general recommendation is at least 500 documents and 5 - 10 samples of each format, depending on the complexity.

The native rules-based ML model (always on) is updated after each document and can provide a high degree of accuracy for most supported documents. The Capture ML is designed to add an extra layer of intelligence, improving an already well trained model. If your document confidence isn't good without the Capture model, it won't substantially improve with it.

There's additional info in the Best Practice guide, see the Machine Learning section at the bottom.

Also check out the Training Models overview.


Ben Lyons
Senior Product Specialist - Decipher
SS&C Blue Prism
UK based
Ben Lyons Senior Product Specialist - Decipher SS&C Blue Prism UK based