AI Data

Updated over a week ago

•

4 min read

Overview

 

AI Data action is used to collect data from a document using optical character recognition (OCR) in combination with Willow360 artificial intelligence. It can read the amount on the invoice, check if a contract has been signed, read the name from the passport, and so many more powerful things. The data read from the document is available for use in future actions, even for deciding which path through the workflow the document will follow based on its content (by using it in conditions).

 

Rules

 

  • This action automatically reads the data from the document's content
  • The user can add data fields by entering:
    • Data Name - which is how this data field will be referenced in future actions
    • AI Query - the question that AI will try to answer
    • Data Type - the expected format of the answer (Text, Number, Email, Date or Yes/No)
  • There is no limit to the number of data fields in one AI Data action
  • AI queries can be tested during action configuration by clicking the test arrow to the right of the AI Query textbox. The Willow360 AI will retrieve answers from the testing document's content
    • The testing document should typically be something that will be used in the live system, and the questions asked should be refined until the user is satisfied with the AI answers
      • Before testing, the user needs to upload the testing document. 
      • In case the testing document is not uploaded before clicking on the testing button, the prompt will inform the user, and the browse dialog will open
      • Testing document size is limited to 5MB. The number of its pages is limited to 40.
      • The testing document can be previewed by clicking its name. The preview opens in another browser tab.
      • Once uploaded and saved, it is stored in the action configuration for future use
    • Clicking the test arrow will run the query on the testing document
      • The answer will be displayed below the question in a chat-like way
      • Each data field is tested separately (separately only from the user's perspective; in the background, Willow optimises AI usage and tries to ask all the questions at the same time and use cached answers if the question isn't changed)
        • Note that a data name change, AI query change, data type change or new testing document will invalidate the cached answers and require the query to be executed again
      • In case the query couldn't be answered, the answer will show “(not found)”
      • In case an error occurred while testing, it will be displayed in the answer field in red color
      • If the returned result is of the wrong type (e.g. having letters while the number type has been specified), the result text will be displayed in red with a warning icon next to it
      • For Yes/No data type, the answer will be Yes/No, translated into the user (personal) language
    • The Test All button will test all the queries at the same time
  • Example templates can be used to prefill data names and ai queries for a certain purpose. When the user applies a template, it replaces all current ai fields. The idea is to cover most use cases in different areas. Willow offers those templates:
    • Invoice, Receipt, ID Document, Contract, Document Classification, Document Smart Summary
  • AI Power Settings can be picked between Regular and Pro mode
    • This setting will influence the power of the AI engine that provides answers
    • It also alters the cost of the action. Regular AI costs 1 credit, and Pro costs 2
      • As soon as the user switches between these two options, they will see the live update of the total workflow cost
  • User(s) who fill in the missing information can also be configured
    • By default, it is the user who added the files
    • Information suppliers may be specific users of the organisation or be decided dynamically from the form data (user type)
  • Always manually check data will prompt the user to check the data before submitting, even if no data is missing
  • Reading the metadata in the AI Data action workflow
    • Willow will try to automatically read all the required data configured in the action configuration
    • If all the information is there (which will typically be the case), the action automatically continues
    • If any data field is missing, the user(s) configured in Who supplies the missing information will be notified so they can fill in the missing data in the same way as they fill in the Supply Information form.
    • The data fields that are successfully read from the document will be pre-filled in the form
    • Form input type will depend on the data type
      • Date and Yes/No type will be displayed in the current user's language
    • All AI metadata data fields are validated against their type (specified in the action configuration). If the user tries to submit invalid data, we will show a validation error.
    • If the Always manually check data option is checked, the user will be notified to verify the form, even if all the info is there
    • When a file finishes the AI Data action, for sure, all the data specified in the action configuration will be there, automatically or manually inserted
    • When the action is done, AI metadata details will be added to the File History (the same way as Supply Information action details)
      • Date, Number and Yes/No data field details will be shown in user language in the file history, but when using it as metadata it will be translated into organisational language 
  • Using the AI Data
    • Metadata inputs have the AI Data option. All the data fields specified in the AI Data action configuration will be listed.
    • Only the actions after the AI Data action can use its data.
    • If you want more info on the actions where AI data can be used in metadata inputs, please check the Metadata Concept chapter (under AI Data bullet).

 

 

Related Articles