AI in e-commerce: product content and photos, pricing, personalization and customer service

Episode recorded in Polish in 2023 and published on 5 September 2023; the AI tools and features discussed reflect what was available at the time.

Why is everyone suddenly talking about AI?

Tomek: Hi! Welcome to another episode of Flying with Shopify. Today we're talking about the role of artificial intelligence in e-commerce. Is online retail really in for a revolution? Let's get started. Hi, Maciek.

Maciek: Hi, Tomek.

Tomek: Tell me, should we brace ourselves for a total Armageddon, a complete rebuild of business models, because AI is going to revolutionize our everyday sales processes? Or quite the opposite? Maybe we're already part of this change, maybe we already use it, just not always consciously.

Maciek: I love how you ask questions with the hint and the answer built in. It's true, as you say: AI is everywhere and we don't even notice it, just like the internet or electricity. Before we start, I think we should make a small disclaimer: we absolutely don't consider ourselves experts in artificial intelligence, and we aren't. You'll easily find those online. You'll probably have to filter out the self-proclaimed ones, and there have been a lot of them lately, but plenty of sensible, solid knowledge on the subject is already out there.

Tomek: So where does this phenomenon come from? Why is it that suddenly, wherever you look, whatever news you read, it's all about AI and this revolution? Is it about artificial intelligence in general, or maybe about ChatGPT, which has become a very tangible tool?

Maciek: It's down to growing popularity and people realizing what the GPT model can do. ChatGPT is just a tool for talking to that language model, let's put it that way. On top of that, generative AI, generative artificial intelligence, has taken off, not only for language but also for images: all those other tools like Midjourney, DALL-E or Stable Diffusion. Of course, it's worth pointing out that OpenAI's solution, GPT, isn't the only tool, the only artificial intelligence engine, so to speak. Google is developing Bard, IBM is developing Watson, and as we said, artificial intelligence is already present to some extent in most of the apps we use. Whether it's Netflix's algorithm suggesting what films you want to watch, Spotify's algorithm, our very own Shopify, the spam filter in Gmail or face recognition in Google Photos. So it's everywhere. But to answer your question directly: the interest in the topic and the buzz, the hype around it, comes from these capabilities being shown off and people getting to experience them, specifically with GPT.

Generative AI in the store admin: Shopify Sidekick

Tomek: Then, since we started with the role of artificial intelligence and its impact on e-commerce itself, let's look at e-commerce through the tools we have at our disposal that use artificial intelligence or rely on how it works. On the one hand, tools that can help us run the business, and on the other, ones the customer can see, the ones that shape their user experience or their shopping experience in general. Maybe that's how we split it.

Maciek: Let's start with the first group, the various applications we can already use, or soon will, to streamline business processes. Maybe let's start with generative AI, because I think that's probably the most interesting part: this is where the benefits will be the most spectacular, let's put it that way. We can use examples of features Shopify is developing, but not only those. Incidentally, Shopify is betting heavily on AI-based features. In its latest edition, the Summer Edition, where Shopify presents a set of new features to be released soon, many of them are about artificial intelligence. The flagship product will be Sidekick. If you're Batman, Sidekick is your Robin.

Tomek: So, a helper.

Maciek: A superhero's helper, exactly. Shopify's marketing tells this story that store owners are superheroes. And rightly so, because it's an incredibly demanding job. So what will Sidekick be able to do, or what can it already do? It's being built on OpenAI's GPT, but also on Shopify's data. It will help you run your business. You'll be able to talk to it just like you would to a person, through a chat window, and ask it for all kinds of advice and information on how to do things. It will also be able to carry out tasks in your store that have so far taken up a lot of your time. For example, you can ask it to mark down all your products, prepare a marketing campaign or set up a new discount code. Or even make design changes to the store: suggest a new color scheme, a new look, different sections. Of course, you can do all of this manually in the store, but with this tool you'll be able to write it in plain language, in English to start with, and approve only the changes you can see in front of you.

Tomek: In a word, something that speeds things up and simplifies them, and in some cases means you don't need to know the technology or be able to write a piece of code to make those changes.

Maciek: I don't know whether this particular tool will have that, but other tools, Zapier for example, already do. Zapier is a tool that lets you connect different software using so-called zaps. Until now, you clicked your way through it, and now you'll be able to write the command in plain English. For example: if a new order comes in on Shopify, add a new row to a table in Google Sheets. It can also write bits of code to be run at a specific step. You write in English what you want it to do, it writes the code for you, and it works.

Tomek: OK, sure. That really is a huge simplification.

Maciek: Yes. And it's already possible.

Product descriptions and customer replies: Shopify Magic

Tomek: Back to generative AI: you mentioned Sidekick, but I assume Shopify has more tools that help, or could help, merchants run their stores. I've heard about Shopify Magic, a tool that makes it easier to write product descriptions, helps with images and lets you manage product content a bit better.

Maciek: For now, Magic really is only for generating product descriptions, but you can also use it in the visual page editor, to write bits of copy: titles, parts of blog posts, articles and so on. To generate text on the site. Anyone who has published a new clothing collection or added new products knows how time-consuming adding new products is. One description takes 30 minutes on average. Here, you just specify what the description should include, and a moment later you approve the suggestion you get. That's another fairly significant simplification. A similar mechanism works in Shopify Email and in Shopify Inbox, the chat for talking to customers. The chat can handle the most frequently asked questions: it can suggest a reply, and again, you just approve it. Generative AI is going to free up a lot of merchants' time.

Product photos and virtual try-on

Maciek: What's interesting is that this will happen not only with text, but also with images, which you brought up. I don't think I've seen it as a native Shopify feature yet, but there are already apps trying to do it. We know Google is working on it and has already shown a demo of how the tool will look in Google Ads. With even one basic photo of your product, you'll be able to tell it what extra shots to generate.

Tomek: So it can create a whole gallery from a single product photo?

Maciek: Exactly. Take this polo shirt we had to buy in a hurry before today's recording, because you had a meeting with Mr Alzheimer this morning.

Tomek: Thanks for reminding me.

Maciek: You're welcome. You only have a photo of it on a plain background, and you can easily put it on a model or show it in a particular setting that fits your brand's visual language or will appeal to your target audience.

Tomek: That sounds like huge savings on photo shoots, which you often need and which not every store owner can afford.

Maciek: OK, let's move on, because the next thing is something Google is already testing with a few brands, including H&M, I believe. You can already search for some clothing items and see them on male and female models with different body types and skin tones, so you can picture how the clothes will look on you. I think it's only a matter of time before most online stores have a feature like this, and that's when you'll really see the savings, because we know from elsewhere that being able to see a product in a given size, on a different body type, significantly reduces returns. So I think it's a matter of time.

Tomek: So what you physically receive matches what you saw in the photo. That sounds really fascinating. I think on the one hand it'll be a huge simplification, and on the other, above all, it'll save both money and time, because as you said, preparing content for a new product usually takes at least an hour. And if we're adding 100, 200 or even 500 products at once, we're talking weeks of work on a new collection, which can definitely be cut down.

Pricing, stock and returns

Tomek: Maciek, I'd go even further. As you said, content generation is the most tangible side of artificial intelligence, the one most users experience right now. But artificial intelligence is above all about very deep data analysis: data that stores collect from user traffic and transactions. I'd like to talk about how AI can help store owners with day-to-day store management. Because what challenges does a merchant really face? First, they have to decide what price to set for a product, especially in a very competitive market where price matters a lot. Second, they need to stock their warehouse properly, because, as we know, overstocking simply means huge costs for a store. On the other hand, stockouts mean sales drop, and very often so does the store's reputation, because the products we offer and advertise turn out to be unavailable. So it's worth mentioning which tools already work in this area.

Tomek: When it comes to pricing, we already have mechanisms and apps that let us adjust prices dynamically. Prices are adjusted according to guidelines: what it cost us to source the product and what margin we're aiming for. These tools also compare that with competitors, to position us properly against other stores. They can take seasonality and upcoming sales into account, so they'll suggest what promotion to run on those products. In other words, they adjust prices very dynamically to the market situation. That's one thing. For product management, they can predict what similar periods looked like, for example last year, or what sales periods and peaks look like in a given industry. So they'll tell us when certain products should be marked down because we have too many in stock and their season is slowly coming to an end. On the other hand, they'll tell us which products we might run out of for an upcoming campaign we're preparing. These are really very helpful tools for day-to-day work and management, especially for a store with a very large number of products.

Tomek: To round off the features for store and inventory management, I'd add that there are also tools that help you shape your product range, for example by flagging which products come with a fairly high return rate. Then we can make a conscious decision not to offer those products if we're worried the returns will cost us too much. That's another very valuable feature of such a tool, because it helps you plan your range better.

Maciek: So besides potentially problematic products, we can also identify potentially problematic customers, who may be involved in practices aimed at extorting money: it ends with you refunding the customer for the purchase while they keep both the money and your product.

Tomek: Exactly.

Maciek: And that's a fairly popular sport in North America.

How do you keep a customer before they leave?

Tomek: Since you mentioned customers, that reminds me of another feature that really helps with customer retention. As in any business, acquiring a customer will always cost more than keeping one. So to keep customers in our sights, and our store in theirs, and to keep them coming back and buying, a very useful tool is one that prevents, or at least reduces, what's called churn, which, put simply, means losing customers who stop buying from us and, over time, stop being our customers.

Maciek: So we're not literally talking about cart abandonment, but…

Tomek: But about people not coming back to the store after a few months. The tool analyzes users' past behavior and the rhythm of what they did: they visited our site, engaged with our newsletters, responded to events and to the codes they received. If the tool senses that the gap is getting longer for a particular customer, it can automatically prepare an individual offer for them, an individual promo code, to encourage them to come back. That reduces customer loss.

Maciek: You could probably also use it to activate customers: to increase average order value or purchase frequency. I think this brings us neatly to using AI in marketing.

AI in marketing: Klaviyo, edrone, Shopify Audiences

Tomek: Sure, absolutely. That's a very good direction. When you look at which tools are already supported by artificial intelligence today, on the one hand you have the ones meant to help run the store, and on the other they're very much built into marketing tools. Because to work smoothly and effectively, marketing tools have to use the data collected about customers, which, as we said, means personalization and customer profiling. Marketing tools can analyze individual users' behavior based on their sessions and on how they behave in other, similar stores and industries. Based on the information collected and the customer profile, they can generate highly tailored offers, promotions, actions and marketing automations for them. I think the biggest player globally would be Klaviyo, while in the Polish market edrone is definitely doing very well here.

Maciek: I agree. Our clients actually praise both. Among tools that run on artificial intelligence and support marketing, we can also mention Shopify Audiences. It helps identify groups of customers where showing ads will pay off in terms of return on investment. Let's try saying that in plain words.

Tomek: In a word, we show ads to the people most likely to make a purchase and become our customers.

Maciek: Exactly. That's also based on data collected earlier, of course. Personalization and recommendation engines work on the same principle.

Personalization, recommendations and site search

Tomek: I think, Maciek, that brings us neatly to the second part, where we wanted to focus on personalization, on the user experience and how it can be tailored to the information and data we've collected about the user. You started with personalization.

Maciek: The idea is to show users an interface or products they're likely to be interested in. Once again, the master here is Amazon's marketplace: as long as you're logged in, it'll show you a product you're very likely to want to buy. Whether it's books similar to ones you've read, complementary products or categories you've browsed in the meantime. We have the same options in Shopify. We can use tools like Nosto or Rebuy, which are powerful personalization and recommendation engines. They work on previously collected data about customers who have bought in the past, and they're very accurate at suggesting products that ultimately raise the store's average order value.

Tomek: Two features immediately come to mind that you'll find in practically every store: cross-selling and upselling. The first is about suggesting an additional product that complements the one the user has already chosen. That definitely increases the order, but it also improves the experience, because the customer knows they can put together the whole set of products they want in one place. Upselling suggests higher-value products to raise the order value; sometimes they're products the customer simply hadn't come across. And how do these mechanisms work? Instead of manually matching specific products to the chosen one, they use historical data, the past behavior of users with a similar profile, to suggest the products most likely to satisfy the customer, so they're ready to swap their choice for them or add them to their order.

Maciek: Exactly. Site search tools use a similar mechanism, whether it's Algolia or Luigi's Box. I think there's also… Klevu?

Tomek: Klevu too.

Maciek: And it works the same way: the tool collects historical data, looks at what people similar to you typed in, searched for and bought, and personalizes the search results for you instead of showing everyone the same thing.

Customer service

Tomek: Yes, because the key to success, to your store being seen as really attractive, is that users who come to it find what they're looking for as quickly as possible. And in this case, as you said, search tools give them suggestions and simplify that path. In stores with a complex structure and lots of products across many categories, that's often quite difficult, and that's exactly what makes the difference. On the other hand, stores have to work hard on customer service itself. An online store is the equivalent of a physical store, but there's no physical contact with a salesperson. And online shoppers increasingly want an advisor when they buy: someone, or some kind of guidance, on how to choose a product and how to get the right fit. Or they have simple questions about how the store works, how it handles returns, what the status of the return they submitted is. And contrary to what you might think, it's customer service, often after the sale, that largely determines how customers rate a store. That's a challenge if you handle it manually, because the bigger the store and the more customers you have, the more of these situations and people you have to deal with. Maybe we could talk about whether there are AI-based solutions that help, or could help, with this kind of service.

Maciek: Absolutely. All modern customer service tools already have it built in. It's another big area where generative AI is used, besides generating content and images. So far, the biggest casualty of generative AI has been customer service centers. And it's heading towards a point where we can automate and speed up customer service, because a bot can learn everything a person knows in a few minutes and then serve those answers much faster, flawlessly and very quickly.

Tomek: Yes, exactly. ChatGPT is increasingly becoming very strong support in customer service, including asking and answering questions, and sometimes people increasingly confuse automated service with a company employee. The chat's learning level is already really very high, and that makes it much easier to manage communication with customers across many channels. We have to remember that customers never just email us. They choose whatever form is most convenient for them at the moment: they'll message us on chat, send a text or write an email, because that's what suits them at the time. So tools that bring all these channels together, tie them to a single customer and help you decide how best to respond to that customer are worth their weight in gold, and they decide how our store is perceived.

Conversational commerce

Tomek: Maciek, the last area I'd like to touch on in our conversation is so-called conversational commerce. Here, too, we already see fairly advanced artificial intelligence that lets us shop in a completely different way from how any of us has done it so far.

Maciek: Exactly. Through messaging apps, whether text-based, like the chat in a store, WhatsApp, SMS or social media messengers, or voice-based, through Google Assistant, Siri or Alexa. And it's true, you can book a plane ticket to your chosen destination through Alexa.

Does a store need to prepare for an AI revolution?

Tomek: OK, Maciek, I think it's time to wrap up. As we said, we're not artificial intelligence experts and we didn't try to be in this conversation. We wanted rather to touch on different areas of e-commerce and point out where artificial intelligence is starting to play a bigger and bigger role. But going back to the question we started with today: should we really be ready to completely remodel the way we run e-commerce, or is this rather part of the natural development of online retail that we see across the digital world in general?

Maciek: There's no doubt that what's happening around artificial intelligence is a revolution, comparable perhaps only to the internet going mainstream or, later, the arrival of smartphones. I think even those can't compare with the possibilities artificial intelligence opens up for us. But it's not as if this is something new. It's been happening for decades and it's everywhere, and we already use it in many tools without realizing it. I don't think that, as an e-commerce owner, you need to prepare for it in any special way. There's certainly no point working on these solutions on your own, because it'll be very expensive and time-consuming. I'd rather go for implementing ready-made solutions that handle narrow slices of functionality. And if you're on a SaaS e-commerce platform, you practically don't have to worry about it at all, because these platforms are technology companies precisely so that you, as a merchant, don't have to. You get all these new features added to your system for free, often without even knowing it. And where you do want to extend it, there's already a ready-made tool you can plug into your setup very easily.

Tomek: I think we can end today's episode on that positive note. Maciek, thank you very much for the conversation. And thank you all for listening to and watching today's episode. As always, we invite you to comment on our episodes and subscribe to our channel. Thanks a lot.

Maciek: Thanks.