Predicting real-life behaviours with digital traces.

Aug 31, 2018

Virtual and concrete worlds dovetail in continuous cause-and-effect relationships.

As more and more bits of our lives migrate to digital realms, the interplays between what we do in concrete and in virtual worlds have grown tightly together.  As a result, many of our personal experiences have started to unfold through an indivisible, continuous thread of events, straying seamlessly between those two realities irrespective of where they start – or end.

The observation and analyses of online behaviours has proven very adept at predicting the real-world spill-overs by which they are followed. And vice-versa: events that initiate in concrete domains have more and more frequently correlated with their subsequent digital manifestations. With the right data and methods in place, we can aim our prediction lenses at virtually any behavioural continuum traversing those two spheres.

Take, for example, consumer purchases (perhaps the most celebrated instance of concrete-to-virtual cross-over). Nowadays, it is rare to find someone who starts the consideration phase of any meaningful shopping without an online research. Which may very well be followed by a brick-and-mortar retail visit. Which may, in turn, culminate with a final e-commerce transaction back at home. Since each step of that funnelling journey can be measured, modelled and correlated with what will happen next, everything can be mapped out and probabilistically appraised.

“While digital players have grasped behavioural predictions with all their might, CSPs have largely sat at the margins of that movement.”

Behaviour predictions can be developed on an individual or an aggregated level. Sticking to the purchase example, in some cases it is more valuable to determine what each individual will likely want to buy within a specific time frame; in others, it is more critical to establish the odds of how a pre-set segment of consumers (defined geographically, demographically or from any other combination of dimensions) will decide to spend their money.

Because large and granular enough data is an extremely scarce resource (unlike the methods to model it, which are openly shared in the public domain), the prerogative of performing broad behaviour predictions sits, currently, with only two major groups of organisations: mammoth digital players like Google, Facebook, Baidu etc., which capture a mind-blowing portion of our online activities, and the Communication Service Providers (CSPs) which make those interactions possible in the first place.  

However, while the digital players have grasped the opportunity of embracing behaviour predictions with all their might (placing them at the heart of their business models), CSPs have largely sat at the margins of that movement. They have the raw data but neither the capabilities nor the monetisation models to leverage the ‘clairvoyance’ they could get from it.

Overcoming this handicap requires from CSPs a concerted effort to generate the right data, turning it into predictive knowledge that feeds into an adequate set of business opportunities. That would involve, then:

  • Getting the right data in place: setting up network analytics solutions to capture, classify, structure and organise the data traffic flowing across their networks, developing a massive pool of labelled events ready to be digested for further behaviour modelling;
  • Getting the right analytical capabilities in place: establishing data science teams with the tools and the know-how to build actionable predictive intelligence on all the phenomena that traffic metadata can be correlated with;
  • Getting the right business opportunities in place: deploying agile experimentation frameworks to test different business applications. Broadly speaking, that involves:
    1. Boosting CSPs’ own abilities to serve, develop and defend their subscribers, through the anticipation of (and the proactive triggering of responses to) commercial, technical and customer experience events that will likely happen in the future;
    2. Diversifying CSPs’ business models by forming partnerships with third-party organisations that can benefit from telco-generated predictive insights.

Connecting all the dots is not easy: in a telco environment, behaviour prediction can prove impactful only when the perfect alignment of factors falls into place. Once they do, CSPs can experience an exciting glimpse of what it means to walk the talk of machine learning and artificial intelligence: there is no better way to harness those concepts than by defining, and delivering on, real-life cases of behavioural projection.