RE: Creating automation that reads email

[RahulSurana]

Hello Everyone,

I am trying to solve a business problem where we currently receive 10-15K emails a month from our customers with work requests. The details of the work that needs to be completed is typically described in the body of the email and many times the data is unstructured (Please see attached)

Once the email is received, we have human workforce which then manually reads the email and looks for key words to determine the type of request then create a saleforce ticket for the support team to work on.

As you can see the process is very manual with intense volume and more than often we end up missing our sla's as sometimes it can take days before a human can get to the email and pass if along to the support team.

Has anyone tried to solve similar problem? If so, did you use any ocr technology? or just in-built blue prism functionality? Any guidance and assistance would be truly appreciated.

Thanks,
Rahul Surana 

------------------------------
Rahul Surana
Director, RPA
CDK Global
------------------------------

[MelissaSuarez_G]

We worked on a similar automation where the users used to send emails to a department to process refunds. The content of the email was unstructured. So we decided to reengineer the process and create a form that generates an standardized message to the automation, this way the bot knows exactly where to find the keywords that are needed to complete the process.

If your company has a website, you can create a form for your customers to use when submitting requests. For example, you can have a drop-down of predefined request types, and a free field to enter the details.

A fancy approach would be to leverage one of the AI tools to automatically detect the intent on the unstructured messages. However, implementing this option is normally more expensive and very time consuming.

------------------------------
Melissa G.
------------------------------

[RobinToll]

I 100% agree with Melissa's comments here - the key to this is making the data structured, and there are multiple ways to do it. OCR I don't think would come in to the picture - after all, generally e-mails are text already - but natural language or sentiment analysis are more likely. That being said, the two major approaches are what Melissa already mentioned - change the input method to help structure it up front, or apply a smart/dedicated system (machine learning/sentiment analysis/natural language processing) to try and structure it once it's submitted unstructured.

Consider this - what structure do you need to get at least partway through this? A form is excellent for categorization, and if you push it further you can give options for either tying the requests directly back to your support documentation/FAQs or triggering end processes (change of address/refund/replacement/etc) directly without routing through a support person.

Thanks!

------------------------------
Robin Toll
Senior Product Consultant
Blue Prism
------------------------------