Description
A validation station is a necessity if you want a simple, efficient and fluid interaction between automated digital tasks (by the RPA robot or other) and your employees. This tool is proposed by some RPA providers, including UiPath’s one presented in this article.
This interaction allows in particular to:
- Make a “4 eyes” control on critical points to be checked by a collaborator
- Give an execution validation to the robot
- Validate the extraction of data from a scanned document
- If necessary, complete the incomplete extraction
- Teaching the robot how to sort documents
Now let’s look at how a categorization and then an extraction validation works.
Document categorization
The bot (software robot) first fetches the documents to be processed, from a mailbox, a folder or stored in a third party application. Then it reads them, using OCR if necessary, to sort them according to the predefined method. Several methods exist, but the most common are by keywords (the importance of each keyword can be weighted) or by trained artificial intelligence (AI). If these two methods fail, or if the confidence threshold is not reached, the document is presented to a collaborator with a list of categories to choose from. The collaborator even has the possibility to indicate distinctive elements of the document to allow these two methods to learn and improve.
Extraction validation
As described in the article Document digitization, the extraction of data from a document is done automatically. However, not always successfully. The document may have missing data or a slightly different format, a coffee stain, etc. Before considering that nothing is going well and putting everything in error, the robot will call upon the grey matter of a collaborator to tell him what is missing. The person can actually indicate that the document is wrong. For certain sensitive cases, the validation of a collaborator will be systematically requested to ensure the accuracy of the data.
So how does this work? The collaborator is notified that a new task is assigned to him/her. He can then log on to the validation workstation. For UiPath, it is a web application, so he can connect with his favorite browser. After that, the collaborator can open the task and see this:
On the right, the person can see the concerned document and on the left, the data that is required, as well as the values extracted by the robot. Next to each extracted value is the extraction confidence rate. The collaborator can simply click on each extracted value and the image on the left is centered around the extracted value. It is even highlighted by a corresponding color. He can thus quickly and ergonomically check the extraction. He has nothing to do if everything is correct. But let’s consider the following cases:
- A value is missing but present in the document, it is possible to select it on the image to add it
- The value has been extracted, but it is wrong, we simply modify it in the field
- If the value is completely missing, just leave it blank
- The document is wrong, click on the error button and indicate the reason for the rejection
What is remarkable is the possibility of extracting entire tables and being able to have this simply validated by employees. It is also very useful for training AI. You can see on the image below an example of automatic extraction with 4 errors quickly detectable thanks to the highlighting.
The goal is to make the long, intensive and repetitive task as simple as possible. This by greatly increasing the efficiency of each employee. Indeed, he no longer needs to search and open each document. They are all organized and queued ready to be processed. He no longer has to type all the values, which greatly reduces the risk of typing errors. And finally, he doesn’t have to enter these values into the relevant applications (often more than one application!), nor file/archive the documents. This tool integrates with UiPath’s Document Understanding suite. You can see a complete article about it here.
For the curious, you can see below a short video showing how the validation workstation is used.
Could you take advantage of it? Are you ready to try?