Calling a company’s customer service department can be a frustrating experience. You might wait on hold for 20 minutes, speak with a representative (or two, or three) not necessarily equipped to help you and then ultimately hang up, perhaps with your problem resolved or perhaps having canceled your service with the company in a fit of rage. If only there was a way to make it a more rewarding experience for everyone involved — the customer as well as the company — and perhaps even save everyone some time and money. That might be coming, and big data might be leading the charge.
Deepak Advani, VP of predictive analytics at IBM (s ibm), explained to me recently how such a vision might play out. Basically, it’s a matter of putting analytics tools to work for customer service reps without them ever having to learn a thing about data analysis. In such a scenario, customer service isn’t left trying to deal with a potentially upset customer while simultaneously trying to find a resolution to his problem because the analytics system has done much of the legwork already.
Knowing you’ve done, predicting what you’ll do
Once a call is initiated, the rep will have a screen full of charts and dashboards showing a customer’s history, preferences and propensities, based on what the system has been able to determine from previous calls and general activity on his account. Depending on the reason for the call, the system might offer a variety of possible offers specifically tailored to that customer. A customer likely to cancel service, or one likely to upgrade to an even more-expensive plan, will be offered whatever it is the system thinks will make them happy.
But predetermining propensities and offers is only the starting point; Advani says real-time analyis is a very real possibility, too. Whereas the information in front of a customer service rep going into a call would be based on historical data that has undergone some predictive analysis, reps could actually type what a customer is saying, which would let the big data system perform sentiment analysis in near real time and adjust accordingly. Such a system could be self-learning, too, Advani explained, learning as it goes what keywords are correlated with what actions and constantly rescoring customers’ propensity models.
Better analytics are good for business
Advani thinks wireless providers might be among the first to embrace such a data-driven customer service model because they finally realized that keeping their own customers might be more profitable than focusing poaching customers from rivals. Additionally, he said, having accurate models of likely customer behavior will let providers even more accurately target and time advertising and special offers.
And before anyone gets any bright ideas about trying to get determine what keywords will trigger beneficial results, Advani also said that the great thing about predictive analytics is that things change so fast. Keywords will likely always be changing as the system learns what’s what, he told me, so that strategy might not work even if someone actually knew how the algorithm looked just a few days ago. It might be a bit more useful to occassionally call a service provider and threaten to quit just to keep the system honest, but I have to think a track record of never having actually quit, and of actually always moving to bigger and better plans, might outweigh a few empty threats.
However, while a call center example is beneficial for explaining how such a system might operate, it’s far from the limits of what’s possible. Advani says IBM is already working on generating propensity scores in about 10 milliseconds to enable real-time experiences, and he said a next step is to start connecting time-and-space data with propensity models. Beyond determining what people might do, Advani said, the ability to tie predictive analytics to streaming data could have far-reaching effects, including helping energy providers analyze sensor data to more optimally balance load across their networks and energy types or maybe refining IBM’s existing work around predicting criminal activity.
Image courtesy of Flickr user pasukaru76.