Whenever I go to some social interaction (back when I could go to a social interaction), people I know or just met inevitably ask what I do and what AnswerOn does. Depending on my mood I either give them the standard “We use advanced analytics to predict consumer and employee behavior and deploy a solution” or to squelch further conversation, I might say “We use models trained with a quasi-Newton algorithm using the cross-entropy figure of merit. Model accuracy is estimated by cross-validation and we use this to suggest end point solutions.”
Sometimes I get people asking me “So, you mean you do artificial intelligence?” and then referring to the various types of modeling machines like: Neural Nets, Bayes Nets, Support Vector Machines, Genetic algorithms, Siamese Networks, etc. Depending on the audience, I will often just say “Yes.”
This dialogue hints at a deeper disconnect and overwhelming truth that gets ignored when people consider “artificial intelligence,” as well as what makes one company more accurate at valuable predictions than another. SPOILER ALERT: it is not the type of model you use that makes the biggest difference.
Rather, what sets companies like AnswerOn apart (somewhat pedantically) is our experience in solving the problems we are contracted to solve, aka domain experience in different problem sets like attrition. With more than 20 years’ experience, chances are we have already developed a data infrastructure we can use with any model.
How does experience shape an engagement?
Think data. You need to have the data necessary to make the predictions. How you present that data to the model is a process called “transformation” and it is one of the key parts to the entire process.
We have experience in many different verticals and a library of thousands of transformations we can use. If we have a prepaid telecommunications client, we know that the interplay between the type of prepaid data and voice usage is a highly predictive “transformed variable.” If we are helping lower attrition or increase performance in a call center, we pay particular attention to data elements like hours worked, patterns of behavior for work from home and brick and mortar employees, and schedule deviations. We then transform the data we receive to help us answer the question we are trying to solve.
Experience helps us to know what data to request and what transformations will be needed to expect high results from whichever model type we ultimately run.
Finally, we need to be able to tell our customers what the results mean and how they can be used to fight the problem. Prediction is only a small part of the overall solution. A customer must walk away with a solution of value, which comes only with experience on solving each problem. Otherwise, they have spent money for a black box which tells them nothing and does not impact their bottom line.
