A Machine Learning Concept “clean out” the Customer Challenge

Hands up if you have tried to configure a supercar, giving all the options you could think of.

04 nov 2015

Hands up if you have tried to configure a supercar, giving all the options you could think of and then having tragically pressed the “Cancel” button when placing the order…

 

We did it almost all.

 

 

As almost everyone, at Elfo we have created a configurator for products sold by our customers.

 

The Challenge, however, is focusing at a higher level. The real question (very “strong” concept, especially if it applies to IT) is: How to create an “Smart” product configurator?

 

The answer

 

Waiting for technological development to have a meek droid for our purposes, we had to entrust a branch of artificial intelligence called “machine learning”.

 

In short, machine learning aim is the study and build of algorithms that can learn from data and make predictions on them. These algorithms create a model starting from sample inputs (training set) in order to make predictions or data-driven decisions, instead of following a rigid and classical programmed logic.

 

Twenty-four hours and … they fly!

 

Due to the limited amount of time available, we had to shrink the implementation field: we supposed to have emails from customers who wish to have a bid for a specific product and additional options. We wanted to use machine learning to re-engage this bunch of text into useful information, well defined, precise, that could be changed into a bid easy to calculate.

 

We did not consider the part of texts “digitization” (there are already OCR software and text extract tools for that purpose) and the creation of a bid from data “records” (several examples already available in our company). The real challenge was “train” the software to recognize customer requirements, while being dispersed in a welter of insignificant text.

 

To do this we used Azure ML, the Microsoft cloud solution designed to facilitate the creation of tools for data mining and machine learning.

 

We split us in two groups: one was making experiments with data and machine learning algorithms while the other was creating a web solution, just to prepare a short demo.

 

And the evening came. Nothing is working

 

And the morning came . “Hey, got it!..”

 

And the winner is ….

 

Late in the morning, it’s time for the last tunings, just to create a “WOW-effect” (graphics and fantastic sound effects).

 

Despite of this, we bid on ourselves, we presented well our work (we almost forgot the sound effects, sigh…) and we won!

 

We were aware of the fact that we have only scratched a surface: the machine learning world really great huge but it is also full of potential things to discover, still unknown to us.

 

Team Fervari riscuote il premio del customer Challenge (ovvero la cena); da sx: Aldi Zeneli, Mirko Perini, Vittorio Sozzi, Marco Sebastiani, Mirko Bersani, Massimo Rocca, Paolo Tacchinardi

 

Fervari team receives the prize of Customer Challenge (the dinner!); from left to right: Aldi Zeneli, Mirko Perini, Vittorio Sozzi, Marco Sebastiani, Mirko Bersani, Massimo Rocca, Paolo Tacchinardi