The Measurement Minute

The Measurement Minute with Gary Angel

 

Combining loss prevention and shopper analytics sure looks attractive when you consider the cost and friction in duplicating your in-store camera investment. But there are some real challenges to killing these particular two birds with one stone.

 

 

The Measurement Minute with Gary Angel

 

If you consider the factors that control what makes a good ML problem and apply them to major sports leagues, it’s pretty clear what role stands out as the most likely to be replaced by machine learning.

 

 

The Measurement Minute with Gary Angel

 

Thinking about when to use machine learning? Defined outcomes and clock-time efficiency are both important. But so too is the scale of the problem you’re tackling – not so much in quantity of data but in the number of variables and control levers to consider.

 

 

 

The Measurement Minute with Gary Angel

 

Acquisitions are a dime (or $300 million) a dozen these days. But McDonald’s acquisition of Dynamic Yield is definitely an out-of-the-box strategic exercise – which is why I think it’s pretty darn interesting.

 

 

The Measurement Minute with Gary Angel

 

Creating training data for ML can be really hard. That’s why you should consider using the data you have to generate more. Check out the full blog I wrote on data factories. But it’s also worth thinking about human annotation.

 

 

The Measurement Minute with Gary Angel

 

The use-cases we here at Digital Mortar have pushed on with ML all involve business problems where we don’t have defined outcome in our data. In this Minute, I look at the problems of accurate geo-location and Associate identification and explain the basic techniques we used to generate training data.

 

 

 

The Measurement Minute by Gary Angel

 

There’s a lot of hype about ML – some of it deserved. But when you look at it closely, almost all of the techniques that really matter in ML fall into the supervised learning category. That’s fine if the data from your problem space contains the defined outcomes necessary for training supervised learning algorithms. If not, then you have some work ahead of you.

 

 

The Measurement Minute w. Gary Angel

 

The evolution of big tech didn’t go the way most of us expected. And the creation of giant monopolists (or monopsonists) creates real issues for public policy. But for omni-channel strategists and analytics folks, understanding why things have gone they way they have turns out to be pretty important for thinking about customer strategy.

 

 

The Measurement Minute with Gary Angel

 

When you see ML programs beating the world’s best chess or Go players, it’s easy to assume that machines are faster learners than people. They’re really not. At least when it comes to learning per experience. But some kinds of problems can be experienced much faster than normal clock time – making them ideal for machine learning.

 

 

 

 

The Measurement Minute with Gary Angel : When Machine Learning Works

 

Machine Learning (ML) has some signal successes. But maybe not in your enterprise. That could be because you suck at it. But it’s also possible that your problems aren’t ideal for ML tools. The Measurement Minute takes a look at this second possibility.