Document Extraction from AzureAI
PreviousEnateAI for IDP - Document ExtractionNextNew Integration: Document Data Extraction from Infrrd
Last updated
Last updated
With version 23.4, Enate users now have access to a new Integration option: Document Extraction from AzureAI. This integration can be activated via the Enate Marketplace, then integrated into user's business flows (the same as you can EnateAI for IDP's Document Extraction) to help speed up the processing and extraction of data from Files attached to work items in Work Manager.
The Azure Document Extraction Integration automatically extracts the relevant data from the Files attached to incoming emails. Documents such as PDFs can be scanned and used both to start Cases in Enate and to form part of the ongoing process's activities.
When a Document Extraction Action runs for a Case, documents attached to the Case can be submitted to Azure for scanning, and processed JSON output files will be returned and automatically attached to the Case. The JSON files give you a structured breakout of data from within these documents, allowing for downstream processing by further external systems and technologies.
If at any point Azure is not confident enough of the results, based on a confidence threshold that you can set, Enate will instantly transfer the work to an agent in Work Manager to look over and verify, giving you that 'human in the loop' support.
When a Case is started in Enate by an incoming email with files attached, the agent can assign Tags to the individual files (or you can use one of the Document Classification integrations offered by Enate to have the system do this for you automatically). Once this is done, the case can move onto an AzureAI Document Data Extraction Action which has been set in the case flow.
The action will process all files that are tagged with the tags it has been configured to pick up. Once processed, if AzureAI is confident in its extraction results, the action will continue to the next point in the case flow, without the agent needing to intervene. A JSON output file of the extracted data (in a structured format) gets attached to the case, and the action will close automatically. Agents can still click to view the Action if they wish to, which will show the completed document extraction(s) and any output JSON files in the 'Files' tab.
If the confidence level returned by AI falls below a given threshold (which you can adjust in Builder if desired), Enate will place the Action into a state of 'To Do', ready for the next available agent to pick up. When the agent opens this Action they will be directed towards the file(s) that needs their review - these files will show a status of 'Requires Verification' in the main section of the Action screen.
To verify such files, the agent just needs to click on the 'Verify Now' button and scroll to the AzureAI Validation Station screen which is then displayed below, to review and amend contesnts. This can be displayed in-situe within the action screen, or in a popout screen to give more screen space.
On this validation screen the agent will be able to see a scanned copy of the file, which can be multiple pages, alongside three tabs showing extracted data.
The Extracted Data tab shows the agent key value pairs of the extracted data along with the confidence level that AzureAI has given them. The values can be adjusted when necessary and are saved once the agent clicks the update button for that value. Doing so will set the confidence value to 100% for that Key.
The Tables tab shows any repeating data that has been picked out as a table.
The Additional Data tab shows additional data that has been picked up from the document. AzureAI's document data extraction technology allows Agents to take this kind of data and actually promote it up to being a Key / Value pair that will be shown on the Extracted Data tab, allowing the Agent to not just adjust the proposed values of recognized keys but also adding further keys if they have not been picked up.
If the agent needs to leave the Validation Station screen at any time they can just click 'Save as Draft' to save their changes. Once an agent is happy with the data all they need to do to submit the updated data is to click 'Submit Validation'.
Once all files requiring verification have been verified by the agent, the action will automatically be marked as Resolved (and will then move to Closed).
How do you go about setting up AzureAI's document classification in your system? There are two main steps to follow:
Activate the integration in Enate Marketplace. This includes adding which 'models' of document type you wish to be available to the Integration (i.e. to tell it what different kinds of documents you want to be able to set the integration to analyse, such as 'Invoice' or 'Job Application')
Add Azure Data Extraction actions into your Case flow at required points, given some settings on what documents you want to be analysed, and which specific model you want the AI to use from the list of pre-defined models.
Before attempting to activate the Document Extraction from Azure integration in Enate, users must first ensure that they have an Azure account with the storage account and storage container fully set up. Information on Azure accounts and how to create one can be found here.
Once a user has their Azure account set up they should then go to Enate Marketplace, located in Builder, filter for AzureAI and then click to Activate the Document Extraction integration. This will cause a pop-up to appear which will need to be completed to successfully activate the integration.
Users will need to fill in the following information:
The URL (which is the 'Endpoint' URL found in a users Azure account)
The API Key (found in a users Azure Account)
The Azure Storage Account Name
The Azure Storage Account Key
The Connection String
The Storage Account Container Name
The Model ID
The Model Name
The Model API Key (same as the key for the API Key Box)
All of the information required above can be found in a user's Azure account apart from the Model ID which can be found from the complete list of Azure models here.
Users can add as many Model IDs as they want by simply clicking the 'Add More Models' button on the pop-up. This will create a new row per click in which the user needs to put the Model ID, Model Name and the API Key. To delete a Model ID a user needs only to click on the delete icon on the right hand side of the rows.
Once a user has filled out all of the required information they need to test the connection.
Once the connection has been successfully tested, users can click the final activate button.
Once you're activated the AzureAI data extraction integration, you can then add 'IDP Data Extraction' Actions into your desired Case flows in Builder. You can either add an existing one from the Actions list if one has already been created, or you can create a brand new one. To create an IDP Document Extraction Action in a Case, from the Action selection drop-down select to create a new Action.
Give the Action a name, add a description if you wish and for its type select 'IDP Data Extraction Action'. When you click 'OK, the Action will be created and added to the Case flow.
On the Action Info tab you will need to set when it's due and set an Allocation rule (i.e. where to route the Action if it needs to be manually reviewed by an Agent when the technology's confidence levels aren't high enough).
There's also general settings for the Action too, and ability to set a custom card, again only really for use in the event that someone needs to intervene and view the action in Work Manager - though remember that the Validation Station screen will automatically show in such circumstances.
Next, go to the 'IDP Document Extraction tab' for the Action to define the settings which specifically relate to the approval activities.
You'll need to fill in:
The Extraction Model - this is the ID of the model you want to use for that process. The Extraction Model options that can be chosen here are the Model Names specified in the Integration details (popup) screen in Marketplace.
The Input File Tag - the tag that a document must be tagged with in order for this Action to pick it up and perform data extraction on it. For example, setting this to 'Invoice' will ensure that only files tagged as 'Invoice' will be picked up from the Work Item. All other documents will be ignored by the Action.
The Output File Tag - the tag that the Action will assign to any analysed files once the document extraction process has completed. For example, you may want to set a value of 'Processed' for any documents which have been picked up.
Once you have filled in the above settings details, you can set the Case live and you'll now have automatic document data extraction working on your Case process.
