Many organizations attempt to achieve “data nirvana” by having 100% complete information for any given business decision. In the customer analytics space, this is sometimes referred to as a “360 degree view of the customer.” However, we really never know everything about our customers. What we call a 360 degree view is really just the most complete view we have at any given time. All of the information we are missing must be inferred or assumed through analytics. The more complete our picture, the less we have to infer, but realistically we are usually inferring far more information than what we have in our possession.
What if we were able to actually know everything there is to know about our customers? What if we were able to understand them as fully as or better than they understand themselves? Would that lead to predictive models that were perfectly able to predict? Would it possibly eliminate the need for predictive analytics altogether?
I recently got into a very interesting discussion related to this thought experiment with my friend John McKean. John is the Executive Director of the Center for Information Based Competition and an author. He is working on a new book titled Extreme Relevancy: Volunteered Customer Intelligence. He called me to exchange some ideas on a few of his intended focal points, and one of the topics we dug into was the implications of perfect information. We pondered the question of whether many of our methods would be unnecessary in a world of perfect information. We ended up concluding that the answer is “No.”
There are several reasons for this which I will discuss briefly below. I may delve into more details on each of the reasons in future blogs.
Reason 1: We don’t know ourselves perfectly
If I were to ask you right now what your favorite movie is, your favorite food is, or your favorite vacation spot is, you would be able to tell me. Knowing that, I could market to those clearly stated preferences. However, the most powerful marketing offers are for things that you didn’t yet know you liked or needed.
Recommendation engines operate on this principle. It is a delight to learn about a different book or movie you’ve never heard of that seems to be a perfect fit for your tastes. In this way, analytics are used to help us discover new preferences. Having “perfect” information on my existing preferences doesn’t negate the need for analytics to raise my awareness of new preferences.
Reason 2: Our preferences can change in the blink of an eye
Let’s assume that you know exactly where I plan to eat dinner tonight and what I plan to order. With that perfect information, you could provide me an offer to add a specific wine or dessert to my meal. However, as I drive to the restaurant I may hit traffic and decide it isn’t worth it. While your offer was perfect when you sent it, it is no longer relevant. Let’s assume that you wait until I am in the parking lot to send the offer. What if the line is too long or I find that the restaurant is out of my favorite dish so I go elsewhere? Again, your perfect offer based on perfect information is no longer relevant.
The point is that to truly have perfect information, you would have to be keeping up with exactly what I am thinking in real time and delivering offers only at the very moment that I have committed to an activity. Not only is this impossible, but it is arguably not a very good way to do things. Once I have committed, your offer is just giving me discounts or incentives that don’t change my behavior, they only reduce profitability. The real value is when an offer drives a change in behavior for a customer.
Reason 3: There is natural variability in our behaviors
With the exception of those with Obsessive, Compulsive Disorder, most people will deviate from even their favorite habits from time to time. Diehard fans of a single type of beer will try another one because it looks interesting, is on sale, and/or they are in an adventuresome mood. I may visit my favorite restaurant most of the time, but try something new every once in a while.
Even if you have a true and complete summary of all of my past behaviors and preferences, the best you can do is determine the probability that I would make any given choice. Then, derive offers based on those probabilities. Our inherent variability adds error into the mix.
While some of these examples may seem a bit extreme, they demonstrate my belief that robust analytic processes are needed even if you have perfect information. While I focused on customer marketing examples, the same concept applies in other areas as well. Our best analytics will always fall victim to changed circumstances, altered plans, or new dynamics that we haven’t encountered before. In those cases, we have no way to predict what the response will be. Even perfect information can be made imperfect very quickly.