Interpretation of a fax message: Intent recognition and data extraction


By loading the video, you agree to Vimeo's privacy policy.
Learn more

Load video

In case of bad video quality, manually change the resolution to 720p or higher on the video menu bar (settings icon)

Extracting the right data from multiple options

In this utility case we can show the power of our extraction framework - a lady declares that her attended person has a different bank account number as indicated and askes to change it. In this case the aim was to automate the process. Therfore several data points (eg IBAN, effective date, account holder,...). Our AI detects the name entities, it understands the concept of custodianship and selects the right name for further processing. We also are able to interpret relative dates eg. "yesterday" or "with immediate effect".

One model for all channels

Having set up the model for eg incomming faxes you could immediately use the same model also for your email inbox. Our solution is based on one knowledgebase and serves all channels with the same model.

Transcript of the analyzed fax (OCR)

In advance via fax: 0221 799658904
Edited by:
Sturmer, Frederike
Admin.- / Customer-Nr:
Date: 01. September 2016
Custodianship Susanne Wagner
Attention: Please consider new bank account!!!
Dear sirs,
I am the custodian of Susanne Wagner. She receives a credit advice 
over EUR 142,67 as stated in the letter from 26.08.2016.

Unfortunately the indicated bank account is not the one of my attended person Susanne Wagner.
The correct bank details are:
Susanne Wagner
IBAN: DE16 1604 0000 0571 4442 00

Proposed Solutions

DEEPSEARCH automatically detects the proper solutions that apply to the situation. It weights the cases with probabilities and ranks them from highest to lowest. In the video shown the intent is properly identified:

Notification – Change Bank Account

Further the defined entities are identified and extracted:


2. BIC

3. Account holder name

4. Transaction type: refund

5. Date