By Dr. Kannan Govindarajan, Co-founder and Chief Product Officer
We were just at Sales 2.0 in San Francisco a few weeks ago, and one of the undeniable themes of the conference was the role of predictive analytics to help sales. Many predictive analytics vendors claimed that their methods gather 1000s of signals from inside and outside the enterprise to be able to predict whether a lead is a good lead or not.
They pick up signals from social media, the Internet, and other data sources, to be able to generate a score to decide which lead will be a marketing qualified lead. Others made the case for scoring accounts using a variety of external signals making the case that the buying propensity of enterprise customers can be characterized using such signals.
So, the natural questions to ask are: Are more signals better? Do they lead to significantly better predictions? How useful are external signals? Most importantly, do more signals help sales people make better decisions?
Our experience with customers has been very different. We have used external signals at some of our customers too, and yes, it did add to the predictive power of our solution. However, most of the predictive power comes from signals in data that are within the enterprise. In most of the situations the number of signals that are meaningful contributors to predicting the outcome is at most in the 10s, never in the 100s, let alone the 1000s that many of these vendors appear to be using. Also, for predictions to be useful, they have to be understandable and actionable.
If you provide sales people with a prediction saying an account is better or a lead is better without a reason, they are less likely to follow the prediction and instead go with their gut, especially if their gut tells them something different. If you use 1000s of signals to create predictions, you will not be able to simply explain why you got the prediction you got. What is the value of the prediction if the sales people do not use it?
Another reason why more signals don't lead to better decisions is best captured in the following Financial Times article Big data: Are we making a big mistake?. The key takeaway from the article that is applicable here is that using 1000s of variables for predictions relies on using "found data" and that can lead to significant prediction errors as with Google's flu predictions. In addition, the more serious problem with using 1000s of signals is that the predictive models thus created are by definition "theory free", meaning, there aren't tight hypotheses being tested.
In essence, having 1000s signals makes it impossible to ascribe causality of outcome to the variables. All we can do is identify correlation which is not the same as causation. And this has a huge impact on usability of the predictions because organizations cannot use the predictions to get better because there is no causal link between the signals and the outcome.
In contrast, when we at DxContinuum added external signals for our customers, the variables we chose were very specific and meant to test hypotheses such as:
- Does the fiscal year of the customer have an influence on when they make a purchasing decision?
- Does the recent financial performance have an influence on whether and when they purchase
- Does recent organizational change have an impact on purchasing decisions?
If you were to ask sales people, they are likely to answer yes to all these questions, and the data does indicate they are correct, but the fact is that even without such external data, it is possible to get predictions that are 85% accurate early in the quarter.
In summary, when it comes to the number of signals used for predictive solutions for sales and marketing, using the right signals that are predictive of outcome is much more important than using 1000s of signals. Clearly, these signals can come from within the enterprise or outside, but what is important is that you are able to explain why you use the signals you use and are able to explain each prediction in terms of the signals. Check out www.dxcontinuum.com
Less is More!