The Measurement Minute

The Measurement Minute by Gary Angel

 

Sometimes bad metrics just fool us. But often, the people using them have entirely cynical reasons for glomming onto the really stupid ones. Credit Yale and other users of the Adversity Score for being perfectly aware of  how awful the metric they’re using is.

 

The Measurement Minute by Gary Angel

 

The College Board, purveyors of the SAT, recently added a proud new edition to their reporting: the Adversity Score. Reading about how it works put me in mind of some of the great measurement blunders of modern history.

 

The Measurement Minute with Gary Angel

 

Shopper journey measurement is like most analytics. When you start to answer basic questions with the data – deeper ones immediately pop-up. And with store measurement, those deeper questions revolve around display interaction. Today, I explain how the Optical AI revolution covered in the last Minute is critical to driving store measurement deeper.

 

 

The Measurement Minute with Gary Angel

 

Store Measurement with camera has been challenged by the cost and the capability to follow shoppers from camera to camera. Edge-based AI coupled with inexpensive monocular cameras may be changing that.

 

The Measurement Minute by Gary Angel

 

When you use human annotation or data factory techniques to create ML training data, there’s always a risk that your training data won’t match the real-world data. When that happens, you’re ML models can look great but perform terribly. Model coherence is one strategy we used to identify and combat that problem.

 

 

The Measurement Minute with Gary Angel

 

I recently did a series of tweets (then collected them in a blog) highlighting the steps in the enterprise analytics journey from initial attempts at reporting to challenges with culture and organization. Surprisingly, some people read it as pessimistic. Not at all.

 

 

The Measurement Minute with Gary Angel

 

If building models is a snap (as per the last Measurement Minute), operationalizing them definitely isn’t. In our work, moving models for ML Studio on Azure into production turned out to be pretty challenging – even for a group of programmers.

 

The Measurement Minute with Gary Angel

 

Machine Learning is at the stage just before widespread adoption when lots of potential users are worried about how hard it is. It isn’t easy (not much is), but lots of people are unnecessarily intimidated by the idea of building ML models. That’s just wrong.

 

 

The Measurement Minute with Gary Angel

 

Analytics – and especially machine learning – require lots of variation. Often enough that’s not a problem. But particularly when it comes to creative, it can be hard and expensive to drive the variation you need.

 

 

 

The Measurement Minute with Gary Angel

 

Everyone (including me) likes to tout analytics ROI and actionable analytics. But the truth is that direct, actionable results from analytics are less common than most of us would like to believe or at least claim…the Measurement Minute considers the parallels to a liberal arts education.