How to train large number of invoices

  • Last Post 23 April 2019
pranai posted this 27 August 2018


I am new to ABBYY FlexiCapture so need some help. 

If I have a huge collection of invoices, some of which are similar and some different, how should I go about training these invoices? Should I create training batches for each of the invoice categories (there are over 1000 such categories) or should the operator train them on the fly?

I wish to proceed with pre-defined Invoice document definition. 

Order By: Standard | Newest | Votes
Vladimir Dimitrijevic posted this 28 August 2018


Please follow this link to get to know ABBYY training:

Here is the example of possible training problem I faced on projects earlier. For one field that is recognized incorrectly, user will once delete value in text box and type value manually and leave incorrect rectangle. But second time user will delete incorrect rectangle and create correct one for that particular field. This means that you would have 2 rectangles on different places and training could not be performed correctly. 

If you have a lot of invoices, I recommend you to create training batches for each vendor, and learn one user how to train invoices correctly: to create rectangles for each field. Use approximately 10 docs a batch, but you will actually train around 3-5, other can serve as test docs later.
Maybe it will sound like too much of work, but you will avoid problem when operators are training on the fly, because they can train it incorrectly and complain that fields are still not recognized even after training was performed. 

Best regards,

  • Liked by
  • Tobor
Abbyy_user0707 posted this 29 August 2018

 What is the use of "Use for Training" in verification station? Will it reflect the trained documents in project setup station or it will store the trained parameters internally?

How we get to know the status of the documents trained or not?


Vladimir Dimitrijevic posted this 29 August 2018

Hello Abbyy_user0707,

Use for Training and Use for Testing are described in help files:

It will reflect of course, don't look at Verification Station and Project Setup Station as solely independent software. It will be more clear for you in future if you see them as a tools of FlexiCapture.

 Best regards,


  • Liked by
  • Abbyy_user0707
  • Tobor
Abbyy_user0707 posted this 30 August 2018

I would like to setup all batches to "Use for Training" default in verification station. Is this possible? or Do I have to do that manually "Use for Training" for all batches.


Also, can we train table? I have loaded and trained same documents more than 5-6 times. Tool able to identify the fields, but not the table. Why?

Ekaterina posted this 20 September 2018


1) Yes, you should send batches for training manually

2) Tables are not trained

diskoboy posted this 21 March 2019

Why is tables not trained?


What is your best recommendation regarding extracting data from tables (sometimes on 1 page - sometimes on +10 pages from the same vendor)


Currently I'm working on a project to extract information on account statements, but it's more or less the same as invoices. For me it seems that the training works on the tables, or am I just lucky?

Earlier today I've upgraded to 12.2 release 7...

Ekaterina posted this 23 April 2019

We recommend you to connect your regional support, provide them your images and describe the scenario in more details.