Our new email routing assigns almost 100% of incoming emails to the correct processing point in practical use. How did this happen, and what hurdles did it face during development? Noman Khairul provides answers.
What was the motivation behind the development of this routing?
The impetus came from a customer who was struggling with numerous emails with no subject, just attachments. These attachments are mostly photos and scans of documents. Previously, routing was done automatically based only on the subject and text of the email. Without manual review, all remaining messages could not be routed to the correct contacts for processing. Response times and SLAs could never be guaranteed this way.
Why go the extra mile to examine attachments as well?
All messages with attachments used to have to be opened, the attachments looked at and the email then assessed as a whole and resent. This took a lot of time, which is now completely saved. The email is now only opened from the right place. The attachment and photo analysis saves time-consuming steps. And of course we have developed it in such a way that it can also be used in other projects. The concept will help many customers and partners.
What works today that wasn’t possible before in email routing?
Without artificial intelligence, which can also detect important information in files and photos, such routing would be hit-or-miss for every email. That is, routing would not be able to determine the correct destination for a good portion of emails because the relevant info would be overlooked. Only with intensive training and AI that classifies files correctly will complex attachments be understood. The computing power and strategies for this have been established in recent years.
How does it get to 100% success rate?
Anyone who deals with AI knows that 100% is actually not possible, or even points to a mistake. The very high success rate is of course related to the fact that our client receives very similar emails from its customers. And very many of them. With an energy provider, there are of course always concerns that are similar to a concern that is already known. That’s why we can judge the content of the emails so well and route just about every email correctly.
What were obstacles during development?
As always with AI projects, we first needed enough sample data. We used our internal labeling team to classify this data correctly for the first time. The high quality of the labeling process then also led to the high success rate of close to 100%.
What future potential does the solution hold?
Our solution is system-independent and can grow with the customer. If the customer decides to switch to another platform in the future, such as a ticket system or similar, it is still possible to use our routing engine in the background. This is because, as mentioned, the engine works not only with emails, but generally with documents from the attachment. This also works via interfaces with well-known ticket systems such as JIRA or link|that Myrmex.