NRF 2017: 3 Lessons Learned from Machine Learning’s Coming Out Party



** Guest post by Brad Thomas, Product Manager, Retail Business Unit**

Each year, one or two buzzwords come to the forefront at the annual NRF Big Show.  You know, the “IT” words that pepper every session description.  The words that seemingly appear on the marquis in every vendor’s booth.

At this year’s NRF, that buzzword was “Machine Learning.” 

In her article in the January 15 Stores Convention Daily, Susan Reda predicts that machine learning will be a top retailer focus for 2017.  For Reda, the value of machine learning is bringing “companies closer to the ideal of getting the right information to the right people at the right time.”  

I like to describe Machine Learning as throwing a massive amount of data into a washing machine and seeing what patterns come out.  This requires choosing a KPI or other variable you want to predict and then letting the machine determine which pieces of data are vital. 

Supporting Real-Time Decisions

For PTC’s Retail customers, a common Machine Learning application is predicting the likelihood that a purchase order they send to a supplier will make it to their DC on time.  In this case, the data they throw into the machine includes product attributes (e.g., type of material, MSRP), information about the supplier (e.g., plant capacity, country of origin), and information about the purchase order (e.g., quantity, lead time).

These data models are applied in real-time to open purchase orders.  Merchandizers are then instantly alerted when a change in situation puts an open purchase order in danger of missing its deadline.  They can then use the underlying information to make an informed decision on how to proceed—such as sending part of the order to another supplier or paying for expedited shipping.

Three Lessons Learned

Getting started with machine learning can seem daunting to many retailers and brand manufacturers.  Here are three lessons learned about machine learning from NRF 2017 and PTC’s experience with retail customers:

  • Have a Purpose.  What complex business outcomes are you trying to understand and what immediate actions can you take with that information?  In the example above, merchandizers are looking to identify potentially-late purchase orders in time for them to take specific actions to prevent them from becoming late. 
  • Keep it Simple and Accessible. What good is an alert if it doesn’t get to the person who can act on it in a format that he or she can clearly understand?  As Seth Hughes, senior manager, asset protection solutions at Walgreens stressed during a panel discussion on machine learning, information needs to be in “plain language…and on the floor, not in offices or BI departments.”
  • Keep it Running.  The beauty of machine learning is that models continuously adapt and improve as more observations roll in.  This means that you can quickly pick up on and react to new patterns.  For example, noticing that purchase orders are tending to run 2-3 days late from vendors who all rely on a specific raw material supplier.

Click here to learn how PTC helps retailers implement successful machine learning initiatives