If a marketer wants to sell, they have to personalize. And not only the content but also the form of communication towards the customer. It's no longer just about addressing them by name and offering similar products in e-stores. "The end customer doesn't expect a super complicated solution. He just wants a good customer experience and tailored communication. In short, for the retailer to make the purchase as easy as possible," says Jakub Šmíd, Tech lead at Blindspot Solutions, in the ADASTRA podcast.
- How long does it take to develop a prototype for personalization of communication on e-shops, and how much time is needed for its integration?
- And why is it better to start with smaller tasks and move from those to deeper analysis?
- How to set up dynamic pricing on food that is approaching its expiration date so that a retailer can increase sales on it by up to 60%, for example?
- How to optimize inventory in warehouses and sales forecasting?
Listen to the podcast (CZ)
Read the podcast as an interview
Ivana Karhanová: Personalisation is a key element of successful sales. At least, that’s what we, who are involved in marketing and sales, are getting from all sides. But, of course, most people in the industry already know that it’s not just about addressing them by name. But personalization isn’t simply offering similar products either. I’m going to talk to Jakub Šmíd, Tech Lead at Blindspot Solutions, about what personalization can mean and what data-driven e-commerce is.
Jakub Šmíd: Hi, thanks for having me.
Ivana Karhanová: What is personalization all about? Or rather, where do you think it could still be, and where do you think it’s missing?
Jakub Šmíd: That’s a good question. From my point of view,, personalization means that when I am a client of a store, I go there, so maybe I want my user experience to be tailored to what I am looking for, what I want, or how the store speaks to me. I want to be contacted when convenient for me and on the channels that I’m used to and things like that. I noticed the last example related to my favorite food ordering app, Wolt. I have roughly six restaurants that I frequently order from.
When I come to the Wolt site, I like to make the restaurants I frequently buy from easily accessible to me. Instead, when I type in the name of a dish, it throws up two hundred different restaurants that focus on Asian cuisine. The app cannot put my favorite restaurant in the results for me as a returning customer. I have to type in a detailed search each time or just scroll, which takes me three times longer than I’d like. As much as I love the app, there is room for improvement. When one reads articles about personalization, addressing by name is a good step, but it would need something more. In the case of Wolt, it would be enough to rank the searches where I shop most often, which strikes me as an almost free and very cheap move.
Ivana Karhanová: Or by where I am right now.
Jakub Šmíd: Exactly.
Ivana Karhanová: The competing app offers me restaurants all over Prague by default, where the delivery costs me an awful lot of money, even though restaurants in the area have free delivery, or I can come and get my food.
Jakub Šmíd: With Wolt, I noticed that there is also a discrepancy between the mobile app and the website. The website offers me everything in Prague, and the mobile app-only offers what’s nearby. When I’m at home, I tend to order food via my laptop, but I usually download the mobile app when I’m on the road. So, for example, if I’m on D1 in Pruhonice and coming home, I’d want food waiting for me when I get there. But the app only offers me restaurants near Pruhonice, which is not what I want. So there, I would say it’s almost the opposite that on the web I would want restaurants near Holešovice, where I live, and in the mobile app, it’s more of a preset where I want it.
Ivana Karhanová: Why do you think they don’t have that? I guess it’s not hard to sort the search results according to the parameters.
Jakub Šmíd: It’s hard to say. Maybe they have a lot of other work to do. I assume they are focusing their efforts on integrating data from all restaurants. But it seems to me that many things would be very cheap and their effect would be extraordinary. For example, if I want to order food, it takes me 20 minutes instead of three clicks, especially if I’m ordering the same thing repeatedly. So if there was a “Last time you ordered,” I would just tap, saying that’s what I want again today. Then I’d be the happiest.
Ivana Karhanová: Actually, what you’re referring to is closely related to UX and UI. It’s not exactly offering similar products anymore, but it’s working with how the user uses the app.
Jakub Šmíd: Yes, definitely. Knowing about the customer is one thing, and the other side of things is how do I use that knowledge and maybe rearrange it a little bit like UI, highlight some things, or on the other hand, choose when to contact the customer. The best examples are real examples where the impulse to contact is relatively clear. And there are a lot of those. For example, right now, I was looking for a tent, looking at reviews, and I picked one. I know the price, but they don’t have it in stock anywhere, and the manufacturer says it will be in a few months. Never mind, I’ll wait. But now, the tent is haunting me on the web. Once I typed the text into Google, it chased me everywhere. Every time it comes up, I hope they have it in stock so I can order it, but I always learn that they don’t have it in stock. So it’s a waste of time for me because I can’t order it. But as a customer, I’m sending a clear signal that I’m committed to that product. It would be enough if the retailers had some sort of spreadsheet of who has requested a particular product and when it’s in stock, so I send them an email about it. And I can still imagine if I got a nice message saying, Hello, James, we finally have that tent in stock, and as an apology for waiting so long, here’s a nice five percent discount. So I’d buy there right away because the discount is great, I’ll pay even less than I wanted to, and I was already committed to the product, and I waited, so I have nothing to think about. But I don’t think I get too many emails like that. I just get a lot of newsletters about what e-shops have new on their site because usually when a person sends an inquiry there, they’re put straight into some newsletter subscription.
Ivana Karhanová: But we’ve gone from basically personalizing products where I go back and often buy, which is like food that you can buy say every day or every other day at some reasonable rate through apps, to products that you buy once in a long time, in the case of a tent, maybe a few years. How to do personalization in this case, because here it’s no longer about UX and UI and being smart in the approach to the user in the app, but personalization should be based on something else.
Jakub Šmíd: Although UX and UI play a role there, see, for example, the lack of buttons like “Let me know when something is in stock” and, of course, the number of things I can do with the data it is different. As you rightly said, with the services I go back to, they know a lot of that data about me, and I want them to do something with that data because they know me as a customer and my whole experience. And then there are other sites where, for example, I don’t regularly buy a tent once every six months. There it’s more about knowing what kind of customers are going there, what to offer them, and how to reach them, and that’s where that data can help.
Ivana Karhanová: So far, I imagine a classic “Other customers have also bought” bar, where I see a selection of five photos of what customers have also bought.
Jakub Šmíd: Yeah, you can do better analytics to make the results relevant – what customers are buying, similar products. There can be a main page that I can set up, for example, based on what I get from Google, so at least basic categories like what age the customer is.
Ivana Karhanová: You’re still doing deeper analysis from that data and using third-party data.
Jakub Šmíd: The more data I can piece together, the more I know about the customer. Of course, it takes more time. If I’m a small e-shop, it’s good to start with some personalization, like adding a “Let me know when it’s in stock” button and writing some simple app that will occasionally go through and email the people who have contacted the e-shop, and those products are already in stock. I don’t think that takes a lot of work. Still, suppose I’m tasked with connecting a bunch of different data. In that case, it’s worth prioritizing things, like what could make the customer experience better, how to make it nice for customers to buy, so they know all the information, possibly give us contact information, that sort of thing. It doesn’t have to be a lot of work either. Doing deeper analysis is possible, but it pays to start with these, as the English would say: low hanging fruit.
Ivana Karhanová: I can think of another sector that does not fall into what we have been talking about: fashion. I do go back to my favorite e-shops. I may even go back relatively regularly, maybe a few times a year, but I buy different products every time. And just as you get chased around the web by a tent, I’m now being chased around the web by a dress I bought on Zoot, and of course, I won’t be buying the same one again. But still, how can they personalize the offer if, let’s say, I’m not going to buy a second identical black dress, or I don’t want to buy a second identical long dress. How can they do that?
Jakub Šmíd: In the past, we did it by, for example, looking at how customers go through a certain website. And we saw that when you go there with some intention like you want to buy a red dress. You click on it. You might assume that if you click on dress one and dress two, there would be some similarities between them. Some algorithms can take a bunch of different passes of what people clicked on in those sequences and somehow learn what products are similar. Then I can tailor the offer to some particular segment of customers based on how they perceive the similarity of the products.
Ivana Karhanová: And you do that similarity based on what – based on the descriptions or the metadata of the products, or based on the visuals?
Jakub Šmíd: Based on how people click there. The underlying assumption is that if I click on two products right after each other, I’ll pick up some similarities.
Ivana Karhanová: But how do you find similar products? Based on the hard descriptions that would have to be entered in by the person entering or receiving the product into the system, or would it have to be from the manufacturer, or do you generate that similarity based on the image?
Jakub Šmíd: No, nothing like that is needed. We just need to have the sequences of what people clicked on. Often it turns out that when you make a similarity based on the descriptions that people type, it doesn’t work well. For example, if I imagine myself having a box in front of me with fifty shoes, and I have an idea of what shoes I want, I’ll click on the first one. I don’t like those, but I click on others that also meet the search criteria. And then, if I put the information together like that from tens of thousands of searches and sequences of people clicking, it can somehow capture important similarities because that’s the information that users are telling me themselves.
Ivana Karhanová: So basically, you’re retrieving this from the history of all the users who are browsing the site.
Jakub Šmíd: Exactly. Then if I want to analyze it more, I’ll take people who are similar to me and what product is relevant to them. So it’s a recommendation for a specific group of users based on customer similarity.
Ivana Karhanová: You’re offering me similar products based on a huge amount of data – what people have clicked on that I have a fairly large data match with.
Jakub Šmíd: Exactly. I don’t need to explicitly say anywhere if I’m doing it based on images or some text, which of course, requires a lot of preparation to have relevant texts. On the contrary, it turns out that details like approaches where I don’t put anything in the similarity that some person explicitly encodes there, because often those labels are not very relevant to what people are searching for. An example would be the label “red shirt.”
Ivana Karhanová: With a picture.
Jakub Šmíd: Or something like that. I mean, it’s some information, but it’s just that if I were to search for red t-shirts, what could be relevant. For example, I looked for a Totoro the Wood Elf plush figure for my daughter. I just typed in stuffed animals or Totoro, and it offered many results. Some of the stuffed animals were small. Others were large. But I had an idea of what I liked, and as I clicked, the sequence encoded the most similar ones.
Ivana Karhanová: It narrowed down the selection of products you were offered.
Jakub Šmíd: Yes. Then if someone was looking for Totoro and clicked on one particular one, the algorithm knows that Kuba Šmíd, after clicking on the first one, then clicked on the second particular one. It can then put that information into the most similar products. It’s based on what Kuba and maybe a hundred thousand other users clicked on.
Ivana Karhanová: How difficult is such personalization?
Jakub Šmíd: Because we only need history, we can quickly make a prototype within days. It’s just a matter of integrating it somewhere on the web. It depends on how complicated it is, whether the client will use some of our APIs or whether we will integrate it directly with them and run it in some of their databases. It shouldn’t take more than a few weeks.
Ivana Karhanová: Personalization is just one part of data-driven e-commerce. The other one that is being addressed is sustainability. We address it in the food/beverage segment, for example, food waste. Will you also address algorithms that can tell supermarkets, for example, when they should start getting rid of that food in advance, so they don’t have to throw it away?
Jakub Šmíd: Yes, this falls under a broader topic called dynamic pricing, where we try to price products a little more optimally to increase sales and so on. In the case of sustainability, you’ll often notice in a supermarket that, for example, on something that’s going to expire, they’ll set a bigger discount, and that’s it. It works online as well. On Rohlík, they have ‘Save the Food’. It’s quite a nice concept. What can we do? When we look at the data, it often shows that supermarkets set the discount too high, which means that, for example, a product doesn’t expire until, say, a week from now, and they put a 50 percent discount on it. The next day it’s completely sold out. If they set the discount smaller, they’d make more money, and the food wouldn’t get thrown away anyway.
Ivana Karhanová: Does that mean they’re going to spread the sale over a longer time at a higher price and a higher margin?
Jakub Šmíd: Exactly. The second case is that they set the discount too small or introduce it too late, and then they throw away a lot of food. In terms of optimal pricing, I would set the discount to limit the food being thrown away and, at the same time, maximize the margin. And this strikes me as a great example. I like it because it has a nice intersection where everybody’s interests merge. There’s some interest in the company where I’m throwing away less food. On the other hand, every time I find something in the fridge that’s gone bad, I feel guilty for a few days afterward.
Ivana Karhanová: There’s refrigerator soup for that. You just throw everything that needs to be eaten into a pot and make soup.
Jakub Šmíd: I’m late for the eternal one, so I don’t want to throw it into the soup. I guess I’ll have to get better at it too. But when you see how many tons of food hypermarkets throw away, you get the feeling that something is probably wrong. Everybody’s happy when you set a discount. Less food is thrown away. Someone buys it cheaper, and the supermarket gets a bit more margin at the same time. Even the supermarket doesn’t want to throw the food away because what it doesn’t sell is a net loss. For example, with our client, we found out from the data that if we set everything up more optimally, we would increase his turnover on the food that is thrown away by 60 percent, which of course, makes a lot more on than margin especially if you sell the goods with little or no loss. That makes an awful lot in the volumes that the bigger chains sell.
Ivana Karhanová: On the other hand, the moment you introduce this, it has to start showing up in your inventory, right? Because the supermarket is ordering other things with some advance notice, and when you introduce something that works with dynamic pricing, it affects their stock levels.
Jakub Šmíd: I think that if I could connect all of that, on the other hand, so that I have pricing that will sell that food and that inventory with some relatively predictable curve, that will, in turn, increase the stability of the inventory. I’m not saying it’s completely easy.
Ivana Karhanová: And it will increase the efficiency of the supply process.
Jakub Šmíd: Yes. The first step for me as a supermarket is knowing what’s going on. It happens to me that the discount is too big or too small, I throw those things away, or on the other hand, they sell too fast. If I sell the goods too fast, it means not only that I’ve introduced the discount too big, but I might have a problem that I haven’t ordered a replacement batch yet. It’s great that I sold it quickly, and it doesn’t spoil, but suddenly another customer comes in, and everything is sold out.
Ivana Karhanová: So then he’s upset that it’s not.
Jakub Šmíd: Exactly. The more stable the sale of goods, the better it is for the people dealing with the stocking. They can tell themselves that so many potatoes are sold every week, and they can order more stock for Monday. But if it’s the case that sometimes it sells, and sometimes it gets thrown away, they don’t know whether to order it for Tuesday or Sunday.
Ivana Karhanová: Is that getting us from dynamic pricing to inventory optimization or sales forecasting?
Jakub Šmíd: Yes. They go hand in hand. I think these topics are quite similar, and if we can partially solve one, we can help the other in the opposite direction. If I have a good handle on when to buy what, that can also be some input into dynamic pricing because maybe I can keep that discount smaller if I know I’m not going to stock up for, say, two days. If, on the other hand, I know I have another few dozen kilos of stock coming in, and I need an empty shelf, then it’s again a good idea to set the discount higher. I think the example of dynamic pricing with the throwing away of food is only part of the problem. And the more logistically I can do that, the better for me. But again, if I’m starting with this, I may just pick a subsegment rather than building some big system that will try to holistically capture as much information as possible.
Ivana Karhanová: That means starting with something and gradually connecting other parts to it as I get that one solution into other processes.
Jakub Šmíd: Yes, that would be my recommendation. I think it works much better that way than finding that I’m trying to build something too complex. Often, you just find that the standard process is we do some analysis, we talk to people about how the process works, when to fire, and how to adjust pricing. To say I’m going to introduce such a discount on this product is easy in the online space. In the case of brick and mortar stores, somebody is going to have to go in and relabel some labels if that store doesn’t already have digital price tags. It’s not like I snap my fingers, and suddenly, the prices in the whole store automatically change. The analytics will determine how challenging this is, which also enters entered as a parameter. Online, I can move the price as often as I want, but only in dynamic pricing can I make that change once a day, once every two hours. But every change can cost something, and it’s good to consider that. You find out in the analysis, but sometimes you don’t find out everything. So if I have a big process that involves stacking, and pricing things over time, then someone might come in and say: That’s nice that you do the math here, but I can’t reflect that. I’d have to call some people then. And, of course, the bigger the project, the more things like that come up over time. So starting with something smaller eliminates the possibility that then some problem will show up because somebody kind of assumed that it would be easy to write this information through and publicize it to the people who then work with it. Then it can turn out to be the opposite, and it kind of hinders the whole use of time.
Ivana Karhanová: Says Jakub Šmíd, Techlead from Blindspot Solutions. Thanks for coming into the studio and sometime over to chat at the next data-driven e-commerce on Audible.
Jakub Šmíd: Thanks for having me. Have a great day.