Putting "Person" into "Personalization"With the ability to do such fantastic customer analysis, there is a risk that we will lean too heavily on statistical models. I'm deeply skeptical of modeling because all models are built on implicit assumptions, and we usually aren't detached enough to question our own assumptions. Modeling is invaluable for finding correlations, highlighting unusual behavior and testing hypotheses. But the need for creativity and intuition only increases as we deal with bigger volumes of data. Predictive analytics can highlight areas to focus, but it can't be relied on to drive a company's decisions. In the end, it comes back to the fact that customers are people. They don't respond to stimuli consistently. They don't always act logically or reliably. They don't even know all the reasons they like or dislike things. Sometimes we have to use thin qualitative data without waiting for statistical significance.
OverconfidenceWe also run the risk of becoming overconfident in our recommendations. There is a great temptation to believe that with so much data we have foolproof theories. Web analytics data is inherently fuzzy, and big data doesn't remove that fuzziness. In some cases, it amplifies the uncertainty. Big data is not the Rosetta Stone to customer behavior. We may never be completely certain about things like how individual customers use multiple devices or how many customers use a single account. Web tracking itself still has gaps that are inconsistent and unreliable. The danger in being overconfident is that we will be unwilling to question our own conclusions. We may interpret new data from the perspective of theories we've already built. If we get our theories wrong, we run the risk of turning off customers much faster than when we inadvertently created dissonance.
Highways to NowhereA related risk to overconfidence is feeling like we know what our customers need better than they do. What complicates the matter is that sometimes we do. However, in creating a strategy for interacting with customers, we should leave room for them to deviate from our expectations for them. Customers should be able to interact with a company like a highway: there should be a multitude of relevant destinations and a clear path to reach each one. We fail if our customers feel like the only way to get what they want is to get out of their car and abandon the highway we built.
The goal of big data is to understand our customers as people and to treat them with that dignity. Advances in how we can analyze data make this more possible than ever.
This is part 5 of a five-part series on big data and web analytics.