How Rules work
This article aims to explain what Rules are and which kind of function they have in KAPTO Models. By the end of the article, users will fully understand the importance and specifics of Rules in KAPTO Models, allowing them to set them up in the most efficient manner for the optimal functioning of the KAPTO Models.
Rules in KAPTO models are one of the possible ways, in conjunction with post-evaluation extension, to route a document into a verification state to engage humans in the document process loop.
While post-evaluation extension, being an entirely external business logic, engaged with complete information both about the document and the executed process, can decide to route the document into a verification state for reasons well beyond a self-assessment of the information reckoned by the AI, Rules work on a much-restricted scenario, having the possibility to take the decision having only the information about the valorisation of entities and the self-assessed confidence of the AI around the extracted value. It means that Rule's primary filtering is based on simple self-assessment (entity discovered or not, self-assessed confidence on the data extracted passing or not passing a certain threshold). At the same time, post-evaluation extensions can perform a much more sophisticated job of creating a complete business logic encompassing the totality of information around the document process and interfacing it with external systems.
The Rules in KAPTO Models work in two different ways:
- UNDER THRESHOLD OPTION: Using a percentage value that determines when KAPTO should initiate a verification request based on the AI's confidence in the accuracy of its determination. Users can set a threshold as a percentage of X% KAPTO will only request Verification if it is less than that X% confident that it has identified the correct value for the entity. Let's make an example: The user sets the threshold to 80%, and KAPTO will only request Verification if it is less than 80% confident that it has identified the correct value for the entity.
By setting the Under Confidence Threshold for an entity, you can ensure that KAPTO will automatically request Verification if it is uncertain about the detected value. This mechanism can help catch potential errors or mistakes early on and prevent them from causing issues.
- NOT DETECTED OPTION: The user selects an "X" number of Entities during the Rules set up so that when the model does not detect the selected Entities, the involved documents are sent for Verification by the AI. For example, if the user selects "Seller" Entity, all the documents in which Seller Entity is not detected are sent in Verification by AI.
