Faux pas at the party
Imagine that your name is Derek and you are attending a party and run into Joanne, a friend from the past whom you are hoping to reconnect with. Remembering her name was the first step in the right direction. However, as the conversation progresses, Derek starts to force the advancement. A waiter with a tray passes within reach, and Derek remembers from last time they spoke that she drinks white wine. Without asking, he picks up a glass, along with a red wine for himself. Derek hands her the glass and hopes that she's impressed with his nonchalance.
"Don't you love this music?" Derek asks. He's sure she does. She's the type.
"No," says Joanne.
"It's so refreshing," Derek says.
"I hate Ke$ha," says Joanne.
"Great to dance to as well," says Derek. "If you haven't, you should really buy her first album."
Joanne eyes start to wander to others around the room. This isn't working. Derek remembers she was very excited abour her job last time they met.
"How's the job?" he asks, restablishing eye contact to maintain Joanne's attention.
"I got fired," she said.
She looks at him as if they've never spoken before. In fact, Joanne smiles at the man standing slightly to Derek's left. "Hi John," Joanne says, handing the glass of white back to Derek so that she can take the red wine John's offering her. Maybe she didn't like white after all.
In today's business environment, this happens far too often – businesses assume they know their customers without the proper information and data to back up their assumptions, and are all too often left alone at the party.
The art of conversation
Wouldn't this conversation have gone a bit better if Derek could have taken Joanne's answers into account ? If his end of the conversation had been contextual, not based on stale data about his last encounter with her, would Joanne not have abandoned him?
Some companies believe real-time decisions are only nice-to-have. These companies may script conversations in advance. If so, all their conversations are like the one above. It may occasionally work, but only when they talk to customers that don't tell them anything new. How likely is that with customers continuously adding to high velocity big data? Every interaction in every channel, every tweet and post, every change in physical location can all become relevant to the current conversation.
How can Derek assume nothing happened in Joanne's life since the last time he met her that may have influenced her tastes or needs? Can he really afford to largely ignore the information Joanne gives him while he's talking to her, most likely about the things that are most front of mind?
The ability to execute predictive models in real-time instead of relying on pre-calculated scores adds critical relevance. No matter how accurate the models are that predict a customer's behavior, they're at risk of being out of date if they are applied to stale data. And, as Derek is noticing, it's often the most up-to-date information that's the most important to a customer, like the reason he or she contacts the company in the first place. Think about calling your provider's contact center because you experience a sudden increase in dropped calls. Or worse, from the provider's perspective: not calling them but checking out the Ts & Cs on their website to see if you have grounds to cancel your subscription without a penalty. What if a predictive churn or retention model could be applied the very moment the dropped calls coincide with researching the smallprint of the contract? In that case the provider could send the (likely) disgruntled customer a proactive retention offer before the (probable) termination call.
A lot happens within a customer's journey within a week, a day, an hour, a minute, or even seconds (think posts, clicks, calls, payments, locations, etc.). It doesn't take NSA-scale data collection for new facts to support more up-to-date predictions and thus better, more relevant decisions. This requires Complex Event Processing to act on real-time events and big data to store it all for ongoing analysis and pattern matching. An increasing number of companies routinely execute hundreds or even thousands of predictive models in support of a single decision as to what the next best action should be. Not everything needs to be real-time, not every customer attribute changes all the time. But even one attribute is enough if it happens to be an input of one of those predictive models.
Coping with velocity and volume
A company that wants to thrive in an increasingly real-time world needs to collect data even if it comes in at extremely high velocity, in many different formats, and in large volumes. This is not your typical data warehousing challenge where ETL (Extract, Transform, Load) tools could take their sweet time collecting data from operational sources and storing it centrally for some leisurely, offline analysis. That may have been the primary data challenge of the eighties and nineties, but today's buzzwords are not denormalization and star schema, but Map-Reduce, Hadoop, and Cassandra. But it all starts with intent. The primary driver for data warehousing was internal, a need for business intelligence, to understand which processes work and which do not work and make changes to the way the business does business. Today, we're adding an external driver – the need to understand, facilitate, and anticipate the customer journey. To do that, we need to ingest more real-time data than ever, and use it to make relevant decisions and take real-time actions. If not, you risk being left alone at the party, just like Derek.