Our experts explore the impact of blind faith in processes without any real understanding of how the business operates and consider why –despite their apparent sophistication –Machine Learning and Artificial Intelligence are not the solution to every problem on their own. Listen as they explore how to think about questioning those building the models to make it easy to both sense check how they work, and to make sure they are built at the right level for what’s needed.
Retail has seen a lot of growth over the past five years, with retailers looking to use Machine Learning and Artificial Intelligence to improve performance, simplify the customer journey, and increase innovation. But should we always trust the algorithms and are they even required? Is AI really the next level of retail optimization, or simply another solution still searching for the problem it’s meant to fix? What is the difference between explainable machine learning over artificial intelligence within retail?
Our experts explore the impact of blind faith in processes without any real understanding of how the business operates and consider why –despite their apparent sophistication –Machine Learning and Artificial Intelligence are not the solution to every problem on their own. Listen as they explore how to think about questioning those building the models to make it easy to both sense check how they work, and to make sure they are built at the right level for what’s needed.
Hear from our experts Robert Bates, Head of Decision Sciences, Currys; Rachel Goldberg, Senior Group Strategy Manager, Currys.
Nel Akoth 00:03
Hello, everyone. Welcome to this episode of the Progress Report. Today we'll spend time on a topic I think of as a hot topic. That is Data and Decision Science. You may ask yourself with so much decision around data, "Why did we pick this topic for today?" Well, we know that while everyone is talking about data, companies struggle to use it effectively. IDC estimates that only 32% of data available to enterprises is actually put to use. That means around 68% of that data is left on the table. Just think about it. 68% left out. Data can be particularly valuable for industries that benefit from a high level of personalization, like retail. I am very fortunate to be joined today by two experts from British retail company, Currys, who will help us explore this topic. Robert Bates is head of Decision Sciences. In his role, he develops and supports the implementation of multi-channel pricing strategy, and provides high quality numerical customer insights to marketing and commercial teams on Currys. And Rachel Goldberg is senior group Strategy Manager at Currys working with Robert on the business side of data driven decisioning. Rachel has over 15 years of experience working within strategy and transformation with some of the UK's top retailers both as a consultant and within industry. Thank you both for joining me today. And welcome. Now I want to begin by talking about the sheer volume of data in the world today. It is truly overwhelming. So let me just touch on a couple of points. More than 3.7 billion humans use the internet. And that's a growth rate of about 7.5% of what it was back in 2016. Now, according to an article from Forbes, every minute of the day, and just think about it from every minute of the day, we send 16 million text messages. Every minute of the day, 156 million emails are sent. We have users every minute watch 4 million YouTube videos. And again, every minute about 456,000 tweets are sent on Twitter. Instagram users, again in the same timeframe, post 46,740 photos in a minute. Robert, it's clear the volume of data and how it's used has increased exponentially, especially in retail. In your view, what's been the key impact of that change.
Robert Bates 03:07
I think it's not just about the volume of data, which has changed over the last five years. I think it's also the ability to which we've got to transform that data to interrogate it. I mean, my career in retail has been pretty long, I think I've been in the industry coming from a consulting or practitioner point for about 20 years. And I kind of think back to when I started with the toolkits which were available, that they're fraction of the power of what they are today. I remember the first project I worked on working with MS Access, a two gig limit, Excel at 65,000 rows. You had to be really tight on what you did. And if I look at the toolkits which we've got today, they're unrecognizable. It's not just the change in terms of the languages of all the autoML, which is coming on the base, or the increased usage of Python above all of the different software technologies we've got. It's the fact that with Cloud Storage and Cloud Processing, things which were unimaginable, even 10 years ago in terms of the access of data, is now available at your fingertips. And that creates a whole host of different challenges, because it's not just about having more data. It's about the ease at which you start to unlock it. And really one expectation that brings in the business as well. Retail has always been about big data. We've always been about having the individual transactional data, which sits linking those customer records over time. But now with the team I've got, we've got the ability to just have that spun out quickly in the cloud. How we use that has been completely transformed.
Rachel Goldberg 04:39
I think to interrupt, having been in retail for a while, I think the one thing that has changed massively, and there has also kind of been perhaps a driving force for this kind of overwhelming growth in data that we're seeing in retail, is a shift in behavior. And that, you know, it wasn't long ago. I mean, at previous retailers where I've worked, you know, the bulk of retailing in some areas is still bricks and mortar kind of and in stores. You know, in those environments, data is really is hard to come by sometimes. Obviously, you've got the sales and you've got the positive systems, but there's actually this growth of online retailing that has actually kind of opened the doors to this wealth of data. And I think it's impacted businesses in various different ways. Firstly, I think you get the differences in businesses, some who perhaps were born before the ".com" boom, and quite frankly, just they don't have the systems or the architecture, to kind of almost ingest all that data. Often it's siloed, it's in different places. Whereas I think perhaps there's some businesses who were born at the time of kind of purely online pure plays. And they're ability to deal with that data, it somehow kind of sometimes can be more advanced. But obviously, they have more of it. And I think that it's both exciting in terms of, obviously the insights and what you can gain from that data, and definitely I think you've seen, as that data has become more available and there's more of it, kind of a lean towards more, I say, a kind of "heart and guts" retailing and commercial trading instincts to more data driven reliance and data driven decisions. I think there's been and I've definitely witnessed a real kind of shift in that needle and therefore realisation of the power that data can give. The impact and power it can have on making better decisions.
Robert Bates 06:29
I think that's where the technological changes have really transformed it. So if you go back 10 to 15 years ago, as a retailer, your core focus was essentially, "I'm gonna bring in products from a supplier, I'm going to curate the range, and I'll sell those to my store channels". Which was primarily about getting the fulfillment right. Getting the operations right. Whereas companies like Amazon, who were primarily a tech company first and happened to use tech advantage for selling products, were very much built from the heart of being "data, data, data" in terms of that driving factor. But what we've seen is the change in the technologies coming in. Those cloud platforms and the ease at which you can ingest data, has meant that what would have been unthinkable for any retailer: building your own massive single customer view, a single database, and what will be a very long term project, you can do a lot quicker. And then the ability to unleash that insight across the business has really sped up.
Nel Akoth 07:25
That's absolutely great. And I liked the way you just honed in on the point about the ease of unlocking the data today. And Rachel, also, as you were sharing from the business side, I want to go back to you a little bit. In terms of maybe getting a little bit more specific on what the expectations are on what data can give you in terms of just both insight and impact? Can you share a couple of examples just so that we can really hone in on how much value and impact you can get out of it?
Rachel Goldberg 07:52
It's now an integral decision making, you know, all the things we're talking about, you know. How do we better understand our customers? Who are our most valuable customers? How do they shop? What are their behaviors? And these are kind of core fundamental questions that help us as a business better curate offers and propositions and make sure that we meet customer's needs. What are the propositions that are working for us or not working for us? If we try something, for example, do you have proof of concept? Is it working? Is it not working? What KPIs should we be tracking? Is the difference that we're seeing just noise or real difference? I think we could most probably go on forever to the list of questions. And it is these questions when you talked about kind of the power of data. I mean, ultimately, I think this kind of insight and the ability to use data in this way, in particular to understand your customers, but also kind of to understand your own business as well, because that's at the heart of this.
Nel Akoth 08:48
Absolutely. Thank you so much. Robert, I want to go back as you were talking about the technology area. You talked about, again, I'm going to really hone in on unlocking the data. You mentioned that toolkit. What is the role of a toolkit? Can you just elaborate a little bit on that?
Robert Bates 09:04
If you start off going back years and years to the old days of the BI team being the ones who'd control a lot of the data going to the business, we always talked about wanting to have one version of the truth. Now, it's something similar I believe within the data science world of how we want to have consistent ways of recognizing customers, consistent ways of recognizing the different product attributes, and the different drivers which create change in the business. So the toolkit which we're looking to build out within Currys all starts from taking all of those different data sources and starting to think about how we want to conform those so that everybody in the business ultimately sees the same thing. So things like conversion number footfall and average retail value sales. There's a whole host of metrics used in retail. And the important thing is all of them are calculated consistently. So everybody knows that when they're looking at them, that they know what it means. From a data toolkit perspective, that means having the right data lake, so making sure all those data sources are accessible. But building on top of that, a set of confirmed analytical records, which you can use to solve the majority of the business problems. This has a couple of different benefits. The first one, if you've got a conformed view of the transaction, you can actually tell the story of what's happening within that record very easily. So within Currys, we sell products and services, and we want the customers to basically be able to enjoy amazing technology. So the cool things in a transaction for me is, "What's the main item in that?" Because that starts to tell me what was that customer was wanting to buy. I could then look at, "What are the other things I'd expect the person to be buying?" So I can very quickly say, "Well, here's a set of data columns, which tell me about the things I would expect to find in that basket, and has that customer taken delivery paid for services paid on credit?" So I've got a single record where I've taken 5-10 lines of data and I've translated them into a story which I can take around the business, you want to make sure that you're using the right measurement system, or the right way of identifying things, so it's consistent across and you don't end up having to find some sort of weird way of wrangling the data to get it conforming across.
Nel Akoth 11:07
Would it be fair to say that that's how you've been really navigating through that personalization? Because I think there's some personalization limiting retail?
Robert Bates 11:15
I think that there's a lot of personalization, which you want to do in retail. But I think the challenge which we face is, in my own personal view, is a lot of the time when we talk about personalization, we think about a one-to-one communication. We think about going down to an audience of one. It's exactly right for that customer. That's great if you know everything about the customer. But in reality, a lot of the times you don't. So you may only have partial information. And the trick for me is how we start to build out those Bayesian models so that we start to fill in more and more information about the customer the deeper we go. If we think about our online customers, if someone's coming to the website for the first time and I've never seen that before, there's a limited amount of information which I've got about them. In that case, I actually know the IP address. So I've got an idea that geographic location. So for me, it's about taking that same, but what do I know about that geographic area? I actually know what store they'd be likely to go to. So I can start to overlay, within our stores, where we think about the population and where we curate the range for that store, I could take the output of that as he applied online. So I could start to use my store-based ranges to curate online if I know where that customer is going. If I can understand the certain demographics that have a high propensity for credit in that area, then I would increase my credit message potentially if I know they're more premium and share more premium products. All of that is a form of personalization. But I'm doing it to an unknown customer. What I'm identifying is the location that tells me enough about to start on that journey. And then as the customer goes through that journey, you can start to overwrite that by reacting to the behaviors which they're exhibiting.
Nel Akoth 12:56
It's always interesting to hear how, you know, you start to build insights and you start to build patterns, as you say, then the behavior that you study of particular customer, and how all that brings a lot of light in the retail industry. Really interesting. Thank you so much, Robert. Rachel, what is the human element of a data translate? To just change slightly here. What role does a data translator play in skills as well as knowledge sharing?
Rachel Goldberg 13:24
If I think about how Rob and I interact, I think one thing we've discussed is there's almost a stream of skill sets that you need, right? From data extraction, manipulation, analyzing, interpreting, visualization, and telling the story. I think where Rob and I support each other is where, you know, for example, helping build those frameworks where we talk about, "How do we establish a framework for understanding our customer? How do we build? What should our most important KPIs or strategic KPIs be? What's the definition of those?" So we kind of work together, side by side and that way. But I think we kind of work as a funnel both ways as well. So first of all, we will take the requests from the business and we have to channel that. And then secondly, I think the other key important element that we bring, and I think we work in tandem with Rob with this is, is that, you know, obviously with the data manipulation and when we're analyzing it, it's very technical. Most of the people in the business, you know, are not mathematicians or experts in this area. And you know, we almost need a way to make that what is extremely complex very simple, very digestible, and bring that story out. Not only just in terms of, perhaps, the piece of data or the information you're bringing to light, but also in the context of the wider business. So why is this important? What elements of the insight that you've seen is more important than others? Why? What does it mean? Therefore, what should you do? What's the decision? What are the options are recommendations? And so I think that's kind of generally how we interacted. And we often kind of knock stuff around together. So I think it's something that, you know, definitely kind of the two worlds can't work in isolation. Absolutely, you kind of have to work together. I would say that's kind of the human element in terms of the translation and definitely in the way we work.
Robert Bates 15:24
It's something which I've seen become more and more needed over the last five years or so. We've got a lot more data. We've got a lot more access to data. But what that also means is, it's easier to kind of make simple mistakes or just not really focus on what the question is and instantly go to a solution. We have so many people coming to throw around the business going, "Well, I want something. I want this." All too often it's solution focused. It's, "I want a model for this, or I want some numbers and it's going to solve everything and make the problem go away." The reality is, it's never that simple. All you've always got to do in the data world is look at what the pathway is to production. How are you going to implement the solution which you're building? And this is where that translator comes in of asking the right questions. And if you'd asked me when I started my career. did I think I'd be using a load of management science and psychology in my day to day job? I'd probably don't know. But the reality is, that's what you've got to do. There's a lot about applying that systems thinking. Breaking down the problem into the different parts of the journey facing the customer, or the different parts of the operating system which you're looking at. And there's a lot of things you end up find that you never thought you'd discover or be looking at. But then it allows you to tell the story and get you on the psychology of why someone is asking that request. What's really driving them? And how do you then actually tidy them and persuade them that you're right?
Rachel Goldberg 16:50
I think what I definitely find most valuable working with Decision Sciences in teams like Rob and his team, is the mutual challenge that you get. So I think that's where you don't generally get the best value. In an imperfect world, the business should understand the problem that it's trying to face and should have done that pre-work and thinking of trying to break it down. But sometimes you do need to bounce these ideas around. Sometimes to get to the nub of an issue, or even just the nub of the question that you want to solve, it's not easy. Especially in retail. Definitely from some of the problems that we've been looking at, they're very complex. And sometimes you don't even know where to start. But I think first is that it's also a business issue to be able to hone those questions. But I also think it's hugely valuable to be able to challenge each other. So for example, when we're working together, I will ask as much of the questions to Rob as he does to me. You know, well, especially when you're embarking on new territory, you know, I think there's a role in the business as well to say, "Well, does that data, and what it's telling me, pass the sniff test? Does it make sense?" And to be able to interrogate the data in that way through not only, "Have I have I kind of calculated it correctly? Is it doing arithmetically what it should be doing? But also, why is it telling me what it's telling me?" And, you know, sometimes it could be, there was actually something surprising that you didn't think that you'd find? Sometimes it might be an error. Sometimes it might have shined a light on some kind of data issues, but in the back end that you haven't seen. And so this kind of mutual challenge and kicking things backwards and forwards between both business and decision science, I think is hugely valuable.
Nel Akoth 18:39
That's really helpful. And in fact, as both you and Robert were explaining it, the thing that kept ringing in my mind is, "How would they know what's the right question?" But I think you've really laid it out well, in terms of just, you know, you test it, and as you go along the way, my read here, and you guys can correct me if I'm wrong, it sounds like as you build the models, they do tend to evolve because as you learn more then you adjust it to make sure that what it's telling you is really representative of the solution, or the story that you want to tell.
Rachel Goldberg 19:07
The most important thing up front is, I think, what we'd like to be able to frame the issue. So you know, the business needs to be able to clearly articulate, "What is the question you're trying to solve?" And then you can kind of agree the methodology of how you might go about it now, but what gets interesting is then what it starts to show you. Because then it can show you all different kinds of things. And that's when you kind of peel back the onion and you can go deeper and deeper. But for me, I think one of the most important things when embarking on any program, project, or data insight, is, 'What is the problem or opportunity statement you're trying to solve? What are you solving for?" If you can put that in one sentence, maybe have two or three sub sentences underneath, then I think that can actually help crystallize things.
Nel Akoth 19:57
Absolutely. So, I want to get your thoughts a little bit on, you know, how do you see technology helping data scientists? And exactly to a point of where does it help, and where does it hinder?
Robert Bates 20:10
To be honest, the most simple ask to how does it help data scientists is, I think, without the technological advantages and advances, there wouldn't be anywhere near as many of us working in as many different areas as there are now otherwise. But as I said, the ability to spin up different kind of compute instances on clouds is far more accessible than it ever was. But with all this power, there are other challenges which you've got to use. The technology makes it a lot easier to build the models, it makes it a lot easier to find those API's and pathways into production. Where I think it starts to hinder, there's a lot of technologies out there with AutoML, where you can basically load up a lot of data and it will create hundreds of new features, and it will then over the course of the next hour 1000s of iterations to tell you which model is the best for the business. That's great. It's a lot more power, it's a lot more speed. But the downside is some of the checks and balances you need to put on place to see how physically realizable those models are may be missing.
Nel Akoth 21:11
Thank you, Robert. Now, Rachel, you know, with all the discussions we've had with data, there's a lot of hype around AI in retail. What do you think that is? And are there benefits and risks associated with using these models?
Rachel Goldberg 21:26
I think AI can, has, and will continue to bring a lot of value into retail. I think that it is a hugely innovative area. And you see it across all areas of retail, whether it's using AI to improve customer journeys, whether it's in supply chain, I think I'm seeing robots in stores now scanning shelves and other retailers. So I think there's AI that can be kind of really value adding, and then there's a fine line between what's value adding and what can be I think, kind of gimmicky. But I think you can use, you know, AI and actually it kind of looks great and it's kind of flashy and it's all nice, and you've really spent a lot of time and money making something. But actually what benefit is it really giving the customer? What benefit is it really therefore giving their business as well?
Nel Akoth 22:18
Yeah, it is very exciting. And you're right, you just can't tell how it's going to sway. But I think with every advancement, it brings in another chance and a different light that as businesses and as industries we all benefit from. Now, you know, Robert and Rachel, we've talked about data, we've talked about AI, and we've talked about automation. How does all this come together to create a personalized experience in retailing? Robert, what's your perspective?
Robert Bates 22:47
For me, it's about early on, whenever we've been developing these models. So thinking what are the real levers which we've got to play with? All of the modeling, the AML and the AI, when it comes together to give the best customer experience. It's about making things easier. It's about reducing friction. So when I go onto the website, I've got a slightly more curated range, something which based either on my past purchases, based on my demographics, and my brand preference. Ideally, then as I browse the website, you start to pick up on the different signals which aren't giving. How close to the purchase am I? What stage of the journey? Am I interested in credit and services? Can you even start to identify, if say I'm in the market for a washing machine, based on the customer's immediate needs and based on their lifetime value, what office should I start to be serving to the customer to get the point next best action for them? So when it works well, I think it can all come together, and it can all come together tremendously.
Rachel Goldberg 23:45
What sometimes can hinder, or make progress in this area challenging, is that almost you need to progress at the same rate across multiple capabilities. So for example, it's you know, you can have loads of insights, loads of data from assistants about your customers, and you can see what kind of potential where that could add value. But if you don't have your front end systems, and if you don't have the communication pathways or that capability to use that data to interact and kind of change behaviors, then it's almost like, you know, you can't do anything with it. So I think for me, it's about being able to evolve, almost kind of a united front, both in terms of your capability to extract and draw insights from your data, to understand kind of how your customer's behaving, and kind of target, especially retail potentially and where it isn't bricks and mortar, those kind of more macro areas. And actually, you almost need to kind of develop that transformational plan together. So you're always not running too far in front in any one area without having that capability and the others as well. Bring all of that together. I think that's where you start unlocking the value
Nel Akoth 24:59
Very spot on. And you know, with the proliferation of data, it is definitely overwhelming. But as we think about how to extract real value from the data and how to create that more personalized experience for customers, that is truly where the differentiation lies. So, Robert and Rachel, I want to thank you so much for being here today to discuss this exciting topic that truly personally impacts all of us and all of our listeners. Thank you so much for spending time with us. And until next time, this is Nel Akoth on the Progress Report.