Table of contents
About the talk
The DATAcated Conference is a free, virtual ‘data party’ hosted by Kate Strachnyi. This is the third DATAcated Conference – it has an industry focus and covers financial services, healthcare, energy, retail, sports, and food & beverage.
Tom Davenport is a world-renowned thought leader and author, is the President’s Distinguished Professor of Information Technology and Management at Babson College, a Fellow of the MIT Center for Digital Business, and an independent senior advisor to Deloitte Analytics. An author and co-author of 20 books and more than 200 articles, he helps organizations to transform their management practices in digital business domains such as artificial intelligence, analytics, information and knowledge management, process management, and enterprise systems.View the profile
All right. We are going to go ahead and check things off with our very first Speaker or taking things off for retail here with c one and only Tom Davenport, such an honor to have him with us here today. He is the president's distinguished professor of Information Technology and management of Babson College, is a visiting professor at Oxford University School of Business, a fellow at the MIT initiative, on the digital economy and Senior adviser to the voice analytics. And a I practice his most recent book is the AI Advantage from MIT press, and he's written twenty books in or three
hundred articles for Harvard Business Review, MIT, Sloan management review, and many other Publications. He has been named one of the world's top 25 Consultants by Consulting magazine. What is 100? Most influential people in it, buy ziff-davis magazine and one of the most of one of the world's top 50 business school. Professors by Fortune Magazine. Amazing. Go ahead and bring your virtual seats here. Welcome to the dedicated conference. Really really, really an honor to have you here today. I just I'm looking
at at the comments. Everyone's ready to get this started. Randy, because here's the one and only time that important. Yes. Awesome. We're going to I'm going to hop off the stage here for you, Tom, and I'm going to let you that you take it away for 10 minutes. Yep, I shared your screen. If it's up there to talk about retail for a little while. I thought, well, I could do a kind of a bland overview of everything that's happening or I could jump into a specific sort of used case and I thought
the latter might be a little bit more fun for my 15 minutes data. Cated Fame. So this is based on some work that I did a few years ago. Next best offers or sometimes people call next best actions. And while it's oriented to retail, it could actually be used in a wide variety of other Industries. I know I've done some work with Morgan Stanley, which has a next best action system for investing ideas and their wealth management business. So it has a pre Broad range of applications. I just thought it might be useful as it in the way illustrates how much data and how complex, it
can be to create one of these applications have been going to work very well. So next best offers or targeted offers or as I said action for customers based on what they bought in the past other attributes that, you know, about them. And then good offers are also based on the purchasing context and knowing the answer abuse of the products or services that you're making the offer of. And of course, this could be products or services or information, or you can offer relationships and a way to sort of people,
you may know, suggestions on LinkedIn and Facebook and so on are a form of offer. They can be delivered over multiple alternative channels and there are some retailers not very many. I think Nordstrom was the most prominent one who believe in delivering offers over through a human salesperson. I think they call that clienteling but that's increasingly rare or decreasing like comment and typically we determine the offer through technology most commonly now, machine learning, some companies still use business rules about there. May also be a human filter again, if it's our kind of a
high-value item and you can afford for human the site, doesn't really make sense. So, this is my time for the next best offer framework and it should have shows how complex it can be and kind of suggests that it's always on going. Your kind of always refining your offers. And so companies that have done this for a while that, for example, CVS, extracare program also known as the world's longest register tapes programme say, you know, you're just constantly thinking about what we can do to refine our offers and learn from what people have responded to in the in the past. So I'll go
through each of these very quickly just to give you a sense of sorta what's involved. If you are. I think they want to do a good job at the gym to think about it and strategically, you have to kind of aligned the offers that you plan to make with what your customers. Wants what? You know, maybe difference, executive product managers. And so on. I'm having mine for you to kind of build the business. And depending on what kind of Technology you're using. You may create a hypothesis to Define your
variables and customer attributes to drive offers. Or if you're using some form of Automated machine learning. You just may say, okay. We got a bunch of a variables are features on our customer and just figure out which ones really dry and effective prediction of what offer they will respond to. So, ideally, we know a lot about our customers, their demographics, or maybe there's something about their psychographics. They're certainly their purchase history, and patterns, is that the single
most important bit of information for a machine-learning based loyalty program and that is typically obtained through a loyalty program, but even retailers who don't have loyalty programs so that they can get if, you know, 50 to 75% of customer identities through credit cards, so you could associate a purchase history with with an individual customer that way to it would be good to know what they done recently. Maybe what they looked at online, their
Channel preferences for how you deliver the offer and any previous expressions. Pinterest. So you can sort of see, just from customer data alone that this can get fairly extensive and and complex. Also, I think one of the things that many companies don't recognize is it important to sort of know the attributes of your product that you planned offer and there may be some Financial considerations calculating offer profitability. Maybe you want to collaborate with
some other providers, there been a number of banks that did offers. And you know, what are you going to do? What you going to offer us a bank of Sino slightly, reduced interest level on a mortgage for something like that. So they typically do deals on other programs, bankamerideals was an example of that. Company card, lytx. Specializes in offering you restaurant discounts and services and product discount. Then the attributes of each offer and the products associated with her to quite critical a particularly. If you're not using
machine learning where you can sort of look through a lot of different teachers and find out what might drive a purchase. But Zappos, for example, is quite good at offers and they concluded that in order to do it. Well, they had to classify the attributes of their products to say, okay, if you bought, I don't know polka-dot shoes in the past. Maybe you by polkadot shoes in the future. If we get a new pair of polka-dot shoes and so in order to make that work, we have to classify
our products as to whether they are polka dot or not. Maybe now you could do that, but he's learning. But at the time they were using, I think two different departments did product. Attribute classification. It's not widely known, but potato in clothing manufacturers. Don't really give you much in the way of of product attributes. You might get a color, but that's typically about it and what products or services are associated with with previous purchases. Obviously, knowing that someone spot polka dot in the past would make it likely you'd want to offer polka
dot in the, in the future. If I want to know something about the purchase contact, so is the customer near you. In a we've been talking about this for a while. You can do it. If you have an app and the customer gives you permission and they happen to have their smartphone with them, but it hasn't really worked out all that. Well, yet, I wouldn't say online behavior is easier to monitor social network Behavior. You can start to monitor. This is, I think I have a version of collaborative. Filtering
filtering. If your friends are buying something you're likely to, to buy it to Then comes I guess what we call the core of the offer. Next best offer process. If we're analytically oriented, which I suspect most of you are and that might be a machine learning model to predict offer response. It could be rule-based. Some of the early church were done with rules. But obviously kind of rules run out of steam after about a hundred different rules. It could be, you know,
the customers has expressed a pattern preference like polkadot polkadot offer. That's a very simple kind of rule that you could offer. But I think machine learning is both more precise. If you have the data to train the models and more likely to be able to, you know, considered a variety of a features are variables. Have that model you got to and typically integrated with some transactional system, a CRM system or an eCommerce system. Something that is actually going to maybe a marketing automation
system that's going to make the offer and you have to time and sequence offers when their customers can respond and would want it. So I remember one talking about this to Bank of America and they said we were trying next, best offers mortgage propositions and they put them on the little slips that you get. If you ask for your balance, be printed out in an ATM machine and they said they thought what a brilliant idea. This was and then they found out that it extended the wait time if there was a line by 30 seconds or so what
people read this this offer and they didn't tend to respond very much either. So there's I think a very important feedback and adaptation process as well. Obviously machine learning models, don't last forever. And so you have to monitor them for drift. Now, of course, you can do a lot of that with ML Ops programs, if you are so equipped. And I think, you know, if you're going to place a lot of Reliance on machine learning models in your business, and they're kind of
tea assets than I think. In my lost makes a lot of sense, as a way to maintain the value of those assets. You design, new offers. If necessary based on the customer response, obviously, new products and services, keep coming up and organizations that you may want to offer. You want to basically treat each offer as a test or at least as some sort of learning opportunity and May know the rules of thumb, even of offers May evolve over time talking to CVS about their Extra Care offers. They had
originally believed that people would respond to cross-category offers, you know, if you bought a toothpaste than you might be interested in mouthwash, but it turned out that people generally only responded to offers of the same things that they'd already bought. I'm so they change that General strategy or rule of thumb to emphasize that more. So that's my slides. I just wanted to give you a little nugget of something that can be done with analytics. And now a, I
maybe you'll find it. Sobering that there, that many steps but like anything with data and analytics. If you want to get a lot of value, you have to give it a lot of fun. Commit to it over time and constantly basically revise whatever you're doing as the as the world changes. So with that, I will stop, I don't know, Kate. I think I have 2 minutes. Maybe if anybody has any questions or comments. Absolutely times. And yes, we have many, many questions and comments will take at least one or two.
I've given the time right? But thank you so much for the presentation, really insightful. I'm glad that you actually chose a specific Topic in retail versus, like you mentioned trying to cover the whole thing in 10 minutes. And I think he really got some really good Insight. Here will take a question here from Gordon, and he's asking what was the most surprising not intuitive next best action that you discovered or Witness. Well, I've always paid the most interesting thing in talking, a lot with Morgan Stanley over the years about
their next best action for investing offers. You know, I always thought that the most important aspect of it was the actual, you know, the quality of the recommendation that the client would think I was this is something that I really ought to do and that, you know, it's based on machine learning. So it must be a very intelligent offer. But according to Jeff McMillan, who's the head of analytics for the wealth management business. It's really more. The kind of client engagement that the process creates that, you
know, you know, your advisor is looking out for you and that they're paying attention to the markets on your behalf. And, you know, they can send out these offers and about with about 45 seconds of work, at used to take Several hours to figure out an investment idea for a client. So it makes the client feel like the financial advisor is much more engaged. And apparently, that's what really drives on business for Morgan. Stanley, testing products, hinder, future ability to see you patterns. I'm at, you know, machine learning is based on the past
and it's often you get enough. You might say, you get in a sort of analytical statistical, right? In a way, unless you are willing. Every once in a while, depending on how fast your business changes that the frequency Woodberry, but you might want to start of open. Thank God that we want to collect more data that we want to totally retrain our models and start from scratch. So yeah, I think that is a good point. You can, I'm get into a rut with these offers. Absolutely. It sounds like that is all the time. Do we have today? But I really, really appreciate you
taking the time really great presentation. You have a lot of comments and questions still sorta flowing in here. That if you had the time at some point, I, if you wanted to jump in on LinkedIn and answer some of those via messages, please feel free to do so, and I especially appreciate you taking the time because it is your wife's birthday today. So, happy birth. Your wife going to say thank you again for doing this.
Buy this talk
Interested in topic “Business Services”?
You might be interested in videos from this event
Buy this video
Our other topics
With ConferenceCast.tv, you get access to our library of the world's best conference talks.