Innovative techniques under the loop

At 4C, we don’t want to fall behind, we want to offer you magic! Well, statistics and algorithms aren’t really magic, but neither is Disneyland. The Learn team investigated 5 innovative techniques in 2015 and want to share their opinion about the most interesting 3 with you.

Machine learning: closer than you think

One of the first successful cases of machine learning was when YouTube was able to recognize cat videos via this technique. Very entertaining, surely, but not really business-ready.

Machine learning sounds as though it’s only for the Googles amongst us. Not for the ‘little people’. Surely we as little people are not confronted with it in our everyday lives… Or are we? For the people who are already envisioning the movie “A.I.’ or ‘I, robot’ to come to life, I can assure you we won’t be taking it that far. But just to show how close to you machine learning already is, a few examples:

• When Facebook draws a square around your friend’s face and suggests which friend it is? Machine learning.
• How they’ve come up with a brilliant show like House of Cards? Machine learning.
• How I’m able to control my GPS system by writing letters instead of clicking on letters? Machine learning.

Facebook was able to let machine learning loose on all the pictures that were found on their site. By examining each pixel and scrutinizing every shadow on your already tagged photos, they are able to find you in untagged pictures. Implementing this feature meant a more user-friendly site where their customers needed to do less effort. Machine learning = Business-ready.

Netflix searched for patterns amongst every series and its ratings. The results? Kevin Spacey turned out to be an actor that everybody appreciates, director David Fincher appeared to produce only successes and the topic of politics seemed to interest the majority of the people. What do you do with this? Create a show based on those patterns and boom! House of Cards is a success. Machine learning = Business-ready.

When stepping into my new car, I started typing in my home address into my GPS. I was astonished to find that I did not have to click anymore; I could just write the letters on the knob! Machine learning is able to distinguish different letters from all kinds of handwriting. Audi stepped up its game and provided a more efficient way for its customers to use the GPS. Machine learning = Business-ready.

Although Facebook, Netflix and Audi are amongst the biggest international players in the world, it still means that the first steps towards integrating machine learning into the everyday lives have been made. Will it wipe away a good ol’ logistic regression? Of course not. Both serve a different purpose and have its time and place. And when that time and place comes, be certain that 4CC is ready for both!

Social analytics, worthwhile?

By the end of 2015, Belgium had forbidden Facebook to track people without a profile and what they did outside of the social network. When they eventually found a loophole and closed down the site for non-Facebookers (obligating you to make a profile to see everything), Belgium sued again and even wanted to make it European this time.

Not only personal network sites are becoming more aware of their user’s privacy, also the professional sites like LinkedIn are realizing that what a user does online is of great value. With the Winter release of 2015, Salesforce got disconnected to LinkedIn and had to stop using its information to enrich its SF profiles.

Online privacy settings are becoming more and more sensitive and legalized. Getting and using online data and behaviour is big business. Accumulating a database of online social richness might leave you poor – if it’s not financially, then probably legally! People are slowly starting to realize they’re leaving a goldmine of information on the internet. They’re getting touchy about the use of what (they think) is personal. And let’s be honest, buying ‘personal’ customer data that wasn’t trusted to you in the first place doesn’t exactly scream customer company.

So if you can’t buy the data to enrich your database, should you just pretend you can’t hear the Twitter bird tweeting at all? No!

Being a customer company means you should be accessible any place, anytime. If a customer chooses to address you directly on social media, please engage in this interaction! Show him you care, show him your quirky side, show him he can trust you. Engaging with your customers will make him feel appreciated and have a stronger connection with your company. Perhaps even trusting you with additional data…?

Monitor online behaviour that is geared directly towards you, but don’t go behind you customers’ back by analysing data they did not trust you with. It’s too expensive, too risky and the added value will not weigh up against the added sorrow.

Text mining, a call to action?

One of our team members, Stijn, recently followed a course with expert Jan Wijffels about text mining. As an example, they got to perform text mining on online movie reviews and predict their IMDB scores. First of all, let’s all apologize to Justin Bieber for receiving such cruel comments on his movie (better stick to what you know, kid). Secondly, although this is a very proper, academic example, just think of the possibilities! Text mining gives you the opportunity to take unstructured data and actually do something with it.

Seeing the rich data coming from something like Salesforce Service Cloud, we could grab this efficiency and take it another step forward. Say for example your contact center gets 1.000 mails per day. You could just treat them with a First in, First out-principle risking that high priority mails get treated too late and you lose that customer. If you received a mail that said “I’m switching to company ABC!” or “I’ve had enough!”, would you wait to answer it? Would you let that customer sit in anger and let him leave? Or would you rather have a system that lets you know the importance of your mails? Because that’s what text mining can do!

Based on what is written in the mail, we could set up a model that identifies which words or sentences increase priority (increase churn). That way, you can act in the most appropriate and efficient way. A mail could automatically be evaluated and escalated in case of a high churning risk. Even small, obvious things could be automated and dispatched accordingly. Think of language recognition or categorization of a complaint. Does this deal with department X, product Y or topic Z? Off it goes to the person with the right knowledge to answer! A customer company responds to his customers appropriately and at the right time. And text mining can help you do this.

We’re convinced. Are you? Let us know!