Wednesday, December 19, 2018

helping machine learning researchers understand the discipline of marketing

When I mention marketing to people in the machine learning community (and even scientists more broadly), some individuals become immediately disinterested or even express a degree of antipathy.  To me, this indicates they really have no understanding of the discipline.  The common perception is that marketing has to do with advertising and persuading people to buy things they don't want.

In reality, advertising and persuasion are only a part of marketing and the bit about getting people to buy things they don't want or need is bad marketing strategy.  Marketing is really more about identify what value you as a company can bring to a market and how best to profit from that value; one can see this by looking at the 3 C's and 4 P's framework that is foundational to the discipline:
  • Customer: What are the customer preferences?  What do they need?  What can they afford?
  • Competition: How are the customer needs already being met (or not) by others?
  • Company: How can we leverage our expertise and assets to give us a strategic advantage?
  • Product: What are we selling?  How does it meet the customer needs?
  • Promotion: How will customers find out about our product?  How can we help them understand that our product meets their needs?
  • Price: How much will we charge?  How will this impact the number of sales?
  • Place: How do we distribute our product?  Will our intended customers be able to access it?

The research side of marketing is inherently interdisciplinary, drawing from economics, psychology, statistics, and (more recently) computer science.  Drawing on such a wide pool of expertise makes sense—the problems marketing folks are tackling are big and messy with real-world impact.  Here are just a few examples of exciting areas for machine learning research in marketing:
  • Recommendation systems: an established application area in machine learning; looking at them through the marketing lens reveals new challenges and opportunities.
  • Impact of policies and information disclosure: when new policies (e.g., healthcare, nutrition, sustainability) are implemented, how does that impact the relevant, often very heterogeneous, markets?
  • Measurement models: how can we build interpretable machine learning models that can be used to identify some theoretical construct in marketing problems (e.g., customer sentiment or preference)?
  • Predicting demand: prediction problems fit neatly into the machine learning world; demand prediction is relevant for many markets, such as electricity, transportation (e.g., bike sharing, rental cars, flights), and selling physical objects (e.g., food, electronics).
I'll end this promotion of the marketing discipline with one thought.  Some very interesting marketing problems would benefit from expertise in machine learning.  Combine that with the need for more of the machine learning research community to ground the motivation for their work in real-world problems and there's a recipe for mutual benefit.

No comments:

Post a Comment