Meat and Bones: Machine learning and DI working together for evidence-based management

by · March 14, 2015

I had a great call with the CEO of a possible partner company for Quantellia this week, where I found myself saying that Decision Intelligence is the “bones” to the “meat” of machine learning.   The image above shows what I mean.  Each star shows an influence link, and each on of these is a possible place where the results of machine learning—whether it’s a decision tree, neural network, simple linear regression, or even a deep learner—contributes to the decision model. As you can see: lots of stars = lots of ways that machine learning can help a decision model to be as powerful as it can be.

I built this particular model last November, on the day that I was scheduled to give a talk at the COMET Executive Summit, put on by Pipeline Publications, in San Diego.  In the morning, there was a great presentation by a telecom vendor, describing how they thought about building systems that helped telecoms to provide better customer experience.   So we had some fun.  I built a decision model in powerpoint (as shown above), then Mark built the model in World Modeler over lunch, then I presented it as part of my talk in the afternoon.

If you’re familiar with the basic ideas of decision modeling, you’ll see the usual levers (yellow), outcomes (green), externals (red), and intermediates (blue) (If you’re not, then a great way to learn the basics is by subscribing this blog (see the Sign Up Now! link in the upper right of the first page), and you’ll get a free copy of Decision Intelligence: A Primer). 

So much of the time, when we build a machine learner, we just do these one link at a time.  So I know a lot of telecoms that build churn analytics using machine learning to determine which customers are likely to leave the company, based on a number of factors (like their monthly bill, features owned, geography, and-for the most enlightened ones-if their social network is churning).  But a churn analytic is just one link in the chain.  Ultimately, we need to make a decision like an investment in improvements to the call center, or an investment in a new model of smart phone, to reduce churn.  And churn isn’t the end of the story either.  If a reduction in churn doesn’t lead to an improvement in net revenues (e.g. if the cost of the churn reduction exceeds its benefit), then we won’t get very far.

So you can see that this single churn analytic is only a piece of a much wider decision that involves many links of influence, each of which can be informed by machine learning.  For instance:

  • How does investment in improving the self-service available via smartphones influence the number of call center users?
  • How does call center volume impact net promoter score?
  • How does an investment in making customer data more consistent improve the consistency of customer’s experience through multiple channels?
  • …and much more

And as we use a model like this in practice, we gather additional data over time so that each link can represent our best understanding of how the influence propagates. We can use adaptive learning techniques, therefore, to continuously improve our ability to make decisions like these over time.

Looked at it this way, we can supercharge machine learning by making the results of our analysis part of larger models, which help to provide an evidence-based decision management structure throughout an organization.

Learn more about Decision Intelligence in Customer Experience Management:

Tags:

You may also like...