Improving Contact Center productivity with Machine Learning

by Andrés González

Last April we had the opportunity to explain how we use Machine Learning to improve productivity in the Ricoh Contact Center.

Ricoh is a well-known Japanese printer and copier manufacturer. Among other services, they provide an after-sale service to solve printer issues in their pay-per-use agreement.

The process we have optimized, first in the Spanish branch and second in the United Kingdom one, is related with how the incidents are dispatched to an on-site technician or they are solved remotely by an agent:

Improving Contact Center productivity with Machine Learning

The business goal is clear: they wanted to reduce costs by reducing the number of visits to the clients. An on-site travel is 10 times more expensive than a remote assistance.

The ML module (dispatching bot, we call it) is a piece of software that automatically sends the incident to the on-site team or to the remote repair team based on the printer characteristics (model, black and white, color, scanner…) and also – and most important – the incident description. This description is a free text where the client describes the diagnosis of the incident, like “There is a paper jam and cannot print” or “There are green vertical lines in every copy”.

The challenge here is to understand human written texts and, based on that, decide to send an on-site team or solve the incident remotely. We use two machine learning techniques: unsupervised and supervised learning. We use Topic Modeling (unsupervised) to enrich the supervised model with language data from the text. The second algorithm uses that information, among other, to make the final decision. In the next video you can see the details.


We would like to thank BigML for the opportunity they gave us to show our solution in the 2nd Edition of Seville Machine Learning School (MLSEV).