Self-learning supply chains? The store of the future? A conversation with an expert about what comes next

Self-learning supply chains? The store of the future? A conversation with an expert about what comes next

Suresh Acharya heads JDA Labs, the 50 person research group at JDA Software Group, the supply chain and retail software giant

JDA is a global supply chain and retail planning software company with 4,500 employees and almost a billion dollars in revenue.  Suresh Acharya heads JDA Labs, the company’s 50 person research group exploring the science, emerging technologies and user experiences that are critical to this complex field.  Network World Editor in Chief John Dix recently caught up to Acharya to talk about everything from customer segmentation to self-learning supply chains and the store of the future. 

Suresh Acharya, Head of JDA Labs, Research and Development

Suresh Acharya, Head of JDA Labs, Research and Development

I imagine a lot of what you folks are working on revolves around big data. 

A few years ago the term big data was quite hot.  But a comment someone made stays with me: Big data has arrived but big insights have not.  Therein lays the challenge.  I think that’s why data science and machine learning is now so important.

I’m a scientist by training, and one of the key things we were taught in school is correlation does not imply causation.  Just because this and that happened, that doesn’t mean because of A, B happened.  Of course the biggest challenge is to distill the correlations that are just incidental from causation.  That’s where a lot of the work is happening.  That’s the value.

There’s a lot of noise out there about trying to figure out what the signals are and this is where industry and academic research is truly going hand in hand.  I feel like academic research still has some more to go before we can truly start to throw everything at a machine and have it figure everything out. 

I was on a call with a customer last week and it was “la, la, la, machine learning, la, la, la, machine learning.” And I said, “Let me pause a bit.  Learning can be good and bad.  The reason a kindergartner throws around profanity in school is that’s what the kid learned at home.  That was learning, but that was bad learning.  Machines will learn as well.  But they don’t know what is good and bad. Sometimes they will learn bad things as well.”

So, when it comes to machine learning, we’ll need people to guide it, at least initially.

What are some of the key problems retailers are facing in the supply chain?  What’s top of mind?

One thing we see is they don’t know who their customers are.  I like to tell this story. I grew up in Katmandu.  My mom would send me to a store to buy something every day, but sometimes the store was out of what she wanted.  But the store knew my family well enough to say, “Take this, I’m sure your mom will be okay with it.”  The corner store person knew their customers.

With the advent of big box retailing that has been lost.  It’s all about product.  What can I sell?  A lot of retailers are trying to get some of that customer knowledge back by using science to study loyalty card data and other customer identifiable information.  “Help me understand, not that you’re John or you’re Karen, but rather, that you are part of a certain customer segment.  Do you buy something when it comes to the market right away, or are you waiting for the clearance to happen?”  There’s a huge focus on understanding the customer segments and trying to figure out what would appeal to customers in those segments.

We’re doing the science to understand what the customer segments are, and then trying to score a certain product for a certain segment.  Here is a pair of shoes.  It may score high to an innovator, someone who buys it right away, but that might only make up 5% of your customers in that particular locale.  It scored a 90 for that particular segment; it scored a 60 for another one. 

The heart of the retail problem is, “What should I offer my customers?” so you need to have that kind of information. 

What’s interesting is, this is where the Amazons are going to be ahead of Walmart because they know who you are, whereas a lot of purchases in Walmart are still done in cash.  The shifting dynamics around what pure-play e-tailers are able to do and why it’s so scary to the big box retailers is this:  They know you so they can make you a customized offer. 

What other kinds of things go into customer segmentation?

You recently had an article about Melia Hotels and its digital transformation effort, and what’s interesting to me is the importance of social sentiment. It’s known that people look at ratings before they look at prices, because we’re okay spending the $15 extra a night if the ratings are good.  Hoteliers cannot ignore TripAdvisor or other kinds of ratings.  It’s very important for them. 

What we learned is ratings you get from Expedia don’t always correlate with the ones you get from TripAdvisor, don’t always correlate to the ones you get from your website because they’re different customer segments.  What one segment might give as a five rating could well be what someone else gives as a three rating.

This is what I would call structured data analysis, because a TripAdvisor rating is a numerical value.  What we’re in the process of doing is trying to understand the unstructured data.  When someone writes a comment saying the lobby was terrible but I really liked the view from the room, what kind of value are you going to put on that unstructured information?

What has been found, though, is there’s a strong correlation between someone giving you a good rating and what they write.  Rarely is it the case that someone writes something nice and then gives you a low rating. The structured part of it does, in fact, form a good proxy to the unstructured.  That’s not always the case in all industries, but at least in this one it’s a good starting point. 

Ok, let’s turn to supply chain.  What’s new there?

One of the things we’re working on is what I call the self-learning supply chain.  Let me digress for a bit to give you an example and then come back to this. 

In some North American cities, certain kinds of crime rates have actually gone down.  That doesn’t mean the overall crime rate has gone down, but certain types of crime rates have gone down.  What has happened is public safety has taken huge amounts of crime related data and applied big data to it to triangulate what causes a crime to happen. 

Rather than responding to a 911 once a crime happens at 3:00 a.m., they have a certain probability associated with a crime potentially happening and deploy cops ahead of time.  To proactively do something is the space of predictive analytics.

In the self-learning supply chain, what we want to know is when something is out of stock, what caused it to be out of stock?  Was it just a surge in demand, or did the inbound order not make it in on time and, if so, why?  Was it a cross-border order that was coming from Canada?  Were there weather conditions?  Were there traffic conditions?  Were there labor issues?  What lead to this out of stock event? 

If you can understand the correlations, then you can see a perfect storm brewing.  Here’s another order coming from Canada.  Should I proactively order replenishment through another distribution center instead? 

So the key is being able to predict an event will happen and then prescribe what ought to be done proactively.  That’s the self-learning supply chain where you learn over time what has happened and then you make proactive recommendations. 

Is this part of the fabled fully automated supply chain?

I tell folks the end goal is to find a cure for cancer, but along the way if you found something that will prolong life by 10 years, that’s not a bad thing.  That’s the kind of thinking we have.

Are we there yet?  No.  Do I think we will get there anytime soon?  I think a lot of the backend automation has happened and will continue to happen.  I would be wary of thinking that peoples’ roles will go away.  People will just find other things to do.

But I was talking to a customer of ours in the third party logistics industry, and they were saying it’s been hard to hire millennials for blue collar jobs. Working in a warehouse isn’t what they want to do.  But they still have roles to fill.  People are retiring.  So some of the things they’re doing to appeal to millennials is gamification.  In this one warehouse, they created groups of associates and they have a leader board to show who is ahead. I don’t necessarily think gamification is only a millennial thing, frankly.  I think we as human beings like to compete.

So back to your question about the fully automated supply chain … that is the evolution, but people will continue to be a key part.  They have to be.  Even with driverless trucks there is still a people role to be played. 

Speaking of supply chains, Amazon is using its own trucks to deliver goods in my neighborhood, which surprises me a bit.  Why would they want to get into that business?

What you’ve raised is a fascinating phenomenon that’s happening and I’ll tell you what that is.  We used to box organizations into categories -- you’re a manufacturer, you’re a retailer, you’re a distributor. But think about Procter Gamble.  You can actually order things from Procter Gamble now.  They have also become a retailer.  They don’t have a Procter Gamble store, but you can certainly buy from them online.  PG never knew how to sell to their end customer and now they have to deal with that.

And now let’s look at retailers.  A lot of them are manufacturing their own generic brands so they have to get a little bit into manufacturing.  And with 3D printing, distribution centers can start doing manufacturing.  If I need a doorknob, rather than having that shipped by a manufacturer all the way through the supply chain, I can print a doorknob for you at the logistics center across the street from where you are. 

As a software company it’s strange for us to have retailers asking for products that we’ve traditionally sold to manufacturers. We would never sell a manufacturing planning solution to a retailer before, but now there is some demand. With a lot of manufacturers becoming retailers, now they need to understand their end customers.  Up until recently they only dealt with the retailer, so they were truly in the B2B space.  Now they’re also in B2C.

Another fascinating thing we have been working on with e-tailers is return forecasting. For certain pure-play e-tailers, 60% of what they sell comes back.  That’s astounding.  Fifteen years ago when it was primarily brick and mortar, 7%-ish of what was sold, depending on the category, would come back, and people would sweep that under the rug.  But you know that if someone ordered three pairs of shoes -- size 10, 10-1/2 and 11 -- two of those are coming back.  Now what do you do? 

Understanding what gets returned has become so important, especially with all the folks offering a free return policy, free shipping.  Understanding that is critical.  For one customer we found that 5% of their customers accounted for 40% of their returns.  They didn’t know it until we brought predictive analytics into it.  What they do with that information is up to them.  That’s not JDA’s play. 

What role is new tech playing?  Say, the Internet of Things?

IoT is certainly the hot thing.  Since JDA is not a manufacturer of these devices, we’re partnering with companies to add the smarts behind the devices.  In a warehouse, for example, being able to use Google Glass or a Microsoft HoloLens to scan items hands free would make workers more efficient.  But we’re not really there yet.  For one, some of the wearables can burn your temple if you wear them for too long. And some don’t support what’s called a multi-scan.  You can’t look, look, look and look.  You have to look, process, look, process, etc., whereas the handhelds can already do that.  I’m sure  this will improve  but  we’re not there yet. 

But at some point some kind of wearable will let people scan information and send it back to a central repository which will collect all sorts of other information and then figure out what the picker in the warehouse ought to do next.  This is multiple machines talking to multiple machines, and JDA’s play is taking that information and saying, “Okay, Bob has to go do this in aisle number three next.”

Right now people in warehouses typically have a list of things to do, but sometimes there are problems like forklift congestion and what was initially perceived as the right sequence is not right anymore.  IoT will make it possible to add real time decision-making smarts. IoT has a huge play in the warehouse.

An aside here, whatever happened with RFID? 

If you had asked me this question a couple of years ago, I would have said RFID was likely done.  It didn’t live up to the hype.  But we’ve seen RFID making a comeback in the store, especially RFID chips in apparel.  It’s still cost-prohibitive to do it in a grocery store.  You can’t put an RFID tag on a gallon of milk.  But in apparel, having an RFID tag that helps track where the inventory is in the store at any given time, we’re seeing a lot of applicability there.  I was at the National Retail Federation Conference that they have every year in New York and they had a demo where jeans were RFID-tagged and they were able to track, not just where they were now, but the history of the jeans too.

You told me one thing the Labs works on is the Store of the Future.  What kind of technology is wrapped up in that?

One technology we think will drive the store of the future is the Magic Mirror.  You stand in front of the mirror wearing something you just tried on and say, “How would it look in red?” And the mirror will change the color of what you have on.  Or the magic mirror allows the user to say, “Put this on for me” to see how it will look without you needing to actually find the item in the store and try it on.   It might also tell you, “Sorry, we don’t have blue, but this is what you would look like in teal.”  It’s just fascinating how these technologies are emerging. 

And we’re working with a robot company where the robot is a store assistant.  Why this is attractive to us is that it has cognitive capabilities.  We think about robots as things that automate things.  Here’s a routine thing a machine can do.  But now robots are getting smarter, and this one robot company has a cognitive robot called Pepper.

You’re probably wondering, what’s the play here for JDA? We play in the expanded supply chain space so we want the robot to meet the supply chain.  When you go to the store and say, “Mr.Robot, I was looking for a green shirt; I only see blue ones.”  It can synthesize the information and say, “You’re right, we don’t have any more green shirts in the store, but the store five miles down the road still has three if you’d like to go there.   Alternatively, my smarts tells me there’s a shipment of green shirts coming to this store next week if you’d like to come back, or I can take an order for a green shirt and have it shipped to your house.”

Cognitive store assistant robots are already deployed in stores in Japan and other Asian countries.    It will no doubt make its way into North America fairly soon.

Follow Us

Join the newsletter!


Sign up to gain exclusive access to email subscriptions, event invitations, competitions, giveaways, and much more.

Membership is free, and your security and privacy remain protected. View our privacy policy before signing up.

Error: Please check your email address.
Show Comments