WM Talks - Product configurator with AI - the future of e-commerce sales

We invite you to listen to another episode of our WM Talks podcast, where we discuss technology and business topics related to the IT industry.

Szymon: Hi, welcome to another episode of WebMakers Talks. I'm Szymon Kubala, and today in the studio with me is Damian Maślanka. It's great to welcome you back to the studio, Damian. Today we have quite an interesting topic, because we're going to talk about product configurators, but from a slightly different angle - we'll add that sexy topic that often comes up, namely artificial intelligence.

But so that we don't jump straight into something complicated, we'll start with the configurator layer itself. I'd like to ask you first of all, what are configurators in general?

Damian: Okay, product configurators are essentially an overlay on a product page that allows us to personalize a given product.

Normally, in a standard store, when we have a product, there might be a small possibility of personalization, like changing the color, but that's more of a variation. A configurator, on the other hand, allows for broader customization - we can modify the shape of a product, change its color, change its dimensions, add a particular style, add some extra element. So it's a more interactive form, where from one product we can actually create many different possibilities. And thanks to that, our customer has the ability to personalize it for themselves, so it's no longer a generic product, but you could say something strictly dedicated to them.

Szymon: Great. Tell me, what are the most common products that are configurable? Do you come across ones that people most often choose and say: "Yes, this is it". And I'd like this to be configured, to be personalized for me, to be more mine. Or are there also other types of configurators?

Damian: Okay, for example, one type is a box configurator. We can configure what kind of box we want, what size, whether it should have engraving or not, whether it should allow for printing, what shape it should have, whether it's a gift box or more of a classic box. Other examples include jewelry. You can adjust whether you want a certain necklace length or a different one, what kind of stone it should have. You could compare it to walking into a jeweler and being able to create a pendant or bracelet exactly the way you want.

Szymon: Like with some kind of assistant, as if they had a catalog in front of them?

Damian: Exactly. Configurators can also be used in the clothing industry, where you can modify the outfit you're browsing to some extent or choose appropriate accessories to match it, so that it fits you as well as possible.

Szymon: Okay, great. And something bigger, for example, I don't know, houses or pergolas or similar things. How does it look in that case?

Damian: For example, garage configurators have recently become popular, where we can actually choose what kind of garage we would like - its length, width, height, whether it should have certain types of doors, whether it should include a window or not, whether we want the roof to be relatively flat or more, let's say, conical. There are tons of possibilities here. In fact, this applies to any kind of structure. Just like you mentioned - pergolas, for example. You can also configure summer houses, basically everything can be personalized and these are already larger-scale products. Even automotive configurators - I'm sure each of us has come across them at some point. We can configure a car exactly how we want it. Choosing the model is one thing, but then what features it should have, whether it should include additional systems, what color it should be, what rims, and so on. In this type of car, there are countless possibilities and these are much bigger things than what I mentioned earlier, like jewelry or boxes.

Szymon: Sure. Recently I came across a really interesting apartment selection configurator for real estate developers, which is also a fascinating topic. We can see what the apartment will look like after the investment phase is completed. We choose a specific apartment - which floor, which side of the building, whether it gets sunlight or not. And then we also choose what should be inside, like a variation of three different interior finish packages. That's also another way to approach a configurator and quite an interesting one.

Okay, great, I'm glad we moved into the configurator topic, but we're here for a slightly different reason - to talk about something that's very current and keeps coming up all the time, which is the use of artificial intelligence in configurators. Have you already come across such implementations? If so, where and how does it actually work?

Damian: And here's a simple example - the apartment configurator you just mentioned. One aspect is having a floor plan or some kind of visualization. Today, generative artificial intelligence is already able to create much more accurate visualizations, more tailored to sales, making them more engaging. Based on such floor plans and specific parameters, it can generate very appealing apartment interiors that simply help sell. And that's just one of many aspects where artificial intelligence can be incorporated.

I would also like to point out that, for now, what's more important is actually having the configurator in your store, and artificial intelligence is more like an add-on that makes the configurator even more dynamic.

Szymon: So as always, right? Artificial intelligence is like the dot over the i and the cherry on top, right?

Damian: Exactly.

Szymon: But it's an interesting topic. I imagine something like this: we have an IKEA product catalog with product photos, and based on those products, we take a photo of our room and furnish it using artificial intelligence with those products. Is that just a vision of the future? Or is something like that already possible?

Damian: That's already possible right now. Artificial intelligence also gives configurators another advantage - when we create a configuration, there can be millions of different possibilities. The more controls and options you add, the number of combinations multiplies each time and becomes very large. AI helps here as well - although there is theoretically an unlimited set of possibilities, we can significantly narrow it down to the best ones and automatically create example sets built with the support of artificial intelligence, which are then recommended to our customers.

Szymon: So we can even replace the entire collection if we want. Great.

But one thing caught my attention - you said there are unlimited possibilities and so on, and when there are unlimited possibilities, there's also some risk, right? That the customer might start abusing it in some way. How does the market respond to that? What's happening in this area?

Damian: Yes, you definitely need to keep your finger on the pulse here, because by definition a configurator is meant to enable personalization, to allow you to configure a product. If there's too much artificial intelligence and it does too much on behalf of the user, then we actually shorten the personalization path again. So you need to find that golden mean - a balance between what the user can do and is actually looking for, and what artificial intelligence proposes and recommends. The user should always have the feeling that they are the one who ultimately decides what the final product will look like.

Szymon: Is there some kind of golden rule that would help me, as a business owner, decide where artificial intelligence should be and where it shouldn't yet? Is there any method, any golden advice? How should I even approach this?

Damian: It's generally a difficult issue and you really have to consider where artificial intelligence can shorten the path and where, despite everything, the decision should remain with the user. A simple example could be color matching. Artificial intelligence can determine that, for example, red will match blue in a given set, but not necessarily black. In that case, it can suggest a color combination so the user doesn't have to think about it. But in areas where its role is not strongly creative and doesn't significantly increase the benefit for the user, it's better to leave the decision to the user.

Szymon: So tell me, does artificial intelligence in this way limit the user's decision-making? Does it restrict it, or on the contrary - does it enrich the number of choices? Which could lead to decision paralysis, because we see that too, right? We open a wardrobe or want to choose a game on Steam, there are more and more options. So the question is whether this isn't a threat. We have an infinite number of possibilities, so in the end we don't complete the purchase because there might always be something better ahead of us.

Damian: Yes, exactly. This is where that golden mean comes in - you can overdo it, or you can approach it in a reasonable way. So of course, implement AI to support the process, but at the same time not give the feeling that "it all configured itself", because then it wouldn't be any different from having a ready-made set or product variation that the user simply selects.

Of course, you can always design it so that there is a proposal generated by artificial intelligence and the user can introduce their own individual adjustments - and that's actually a very good model. There are prepared sets, the user has the option to edit them, and they are already designed to best fit the user.

Szymon: Sure. So it's not that everything happens on its own. You often mention this golden mean. You also say that as someone interested in implementing such a configurator and adding AI to it, I have to do some work. And I'd like to slightly shift the topic here - none of this happens by itself. What kind of work does the client, the person implementing the configurator, have to do? That's one thing. And secondly, what about implementing the configurator with artificial intelligence? I assume the workload is different and that someone has to invest effort not only by commissioning the project, but also by doing some work themselves. What does it actually look like?

Damian: Here, you need to divide the whole process into two main steps. The first step - and we've also talked about this in previous podcasts - is to define what kind of configurator you want to have on your website. That's step number one. You need to mature to the decision that, okay, we have a store, we have products. Then think about which products are worth customizing and giving customers the option to personalize. And then implement a standard, basic configurator. Once you have that configurator, you can first test it in practice with users. See how they configure products. That already gives you internal insight into what should be sped up, which decisions or steps are unnecessary, cause friction for customers, or which are so obvious that the customer always selects the same option regardless of having more to choose from. And this is where there is room for artificial intelligence to automate some of these steps. It can propose intermediate steps that are already pre-filled, as well as the entire final visual layer. As we said, the number of configuration possibilities can be unlimited. It's very difficult to build models that account for every possible case. At that point, a decision may be made: okay, I want artificial intelligence that, for example, will visualize all these selections and generate the appropriate image or model that ultimately presents the final result of the configuration process.

Szymon: So in reality, using artificial intelligence is rather for those who have already validated their configurator model in some way and know that it works for them. Is that true? Or can you also start the conversation with: "I want a configurator, but I also want AI in it"? Is that also a good path?

Damian: Exactly. I'm a supporter of approaching all processes iteratively. Every iteration gives us knowledge, experience, and insight into which direction we should take next - in this case, how to further develop the configurator.

It also gives us market feedback. For example, we may see in the statistics that the configurator increased sales by several dozen percent. In our case, it actually happened. We have higher sales, we have fewer email inquiries about product configuration, and at the same time we've been collecting data in the background on how users configure products.

Now, based on that data and on hard results showing that the configurator really increased sales and profits, we can optimize it even more. We can notice that for many users the process still takes a few or a dozen extra seconds - which are crucial and affect our return on investment. Then we can decide: okay, let's implement artificial intelligence here. Of course, you can take that step immediately and launch a configurator with AI, but then you lose that intermediate step and you don't know how AI actually influenced the results. Did it really increase sales, or was it just a nice add-on? Or it may even turn out that effectiveness decreased because the user wasn't fully satisfied with the results or lacked decision-making control in some aspect and abandoned the cart.

Szymon: So we move from "I think so" to "it's necessary because the market says so". Okay. While you were talking, you also mentioned conversion, and I'd like to ask about that to wrap things up. How does implementing a configurator - and later a configurator with artificial intelligence - translate into increased sales and conversion for your clients? What does it look like in your case?

Damian: Generally speaking, implementing a configurator and enabling product personalization results - in most cases, because we can't put everything in one basket - in sales increasing by several dozen percent. Of course, it all depends on the product type, the store type, and customer loyalty. If we have a loyal customer base that buys regularly, the impact may be slightly smaller than in the case of constantly acquiring new customers who compare us with competitors. There's also another aspect that hasn't been fully discussed. One layer of artificial intelligence is the external one we've been talking about - making the form more autonomous, dynamic, auto-filled, generating visualizations automatically. But there's also a huge analytics area. Every time someone clicks through the configurator, changes something, buys or doesn't buy - that information is stored in the store's database.

And artificial intelligence can analyze all that data, compare it, look for patterns, identify so-called twin customers who followed similar decision paths and bought a product. It becomes a massive knowledge base that allows us to draw conclusions and identify patterns. Thanks to that, we can refine the product page even further, and those several dozen percent increases may turn into an even higher sales rate because we have concrete data and AI-driven analyses.

Szymon: So in a way, a sales assistant appears - not even a sales assistant, because that sounds wrong - more like an assistant to the seller, the person running the business, who often doesn't have time to validate everything. And as I understand it, we ask artificial intelligence that has collected the data, and it draws conclusions and suggests actions. Is that correct?

Damian: Mainly, it's about understanding that artificial intelligence has its strengths and weaknesses. The key is to know where its strengths lie and treat it more as a recommendation system rather than a golden oracle that says "do this and it will definitely work". We should treat it as a source of suggestions. If it detects that many users tried to configure something but didn't complete the purchase, that's already a signal that the configuration path is not optimal. Then the human factor comes back into play. As store owners, we can go through those paths ourselves, see why the sales weren't finalized. We understand our business best. And once we improve that path, we know it will start converting better and increase sales.

Szymon: And as humans, we have a broader understanding than what artificial intelligence gets from raw data alone. It reminds me of a billboard I once saw that said: Don't believe everything you see on the internet. The founder of the internet - Nicolaus Copernicus. It's similar with artificial intelligence - we shouldn't believe everything automatically. We see the flood of AI-generated content on social media, a lot of it misleading. We also see it in various chats when we validate and verify the information. It's important to remember that. But I think this really shows the potential of artificial intelligence in configurators and in businesses built around them - especially since this area is growing rapidly. I'd also like to ask about that growth. From your perspective as a software house, do you see more companies and stores moving toward configurators to simplify the sales path or change the sales channel? What does it look like for you?

Damian: In general, a certain trend is emerging in the market. It's not fully formed yet - you could say it's still in its infancy - but it's already breaking into the mainstream. And that's largely thanks to LLMs. Since ChatGPT became widely available, we've been moving from a model where information had to be searched manually toward one where AI can gather everything and present it in an accessible form. The market will continue evolving in that direction. More and more, we'll be "talking" to websites instead of manually extracting information. And naturally, configurators will undergo this kind of evolution as well. I'm deliberately not using the word revolution, because it will be a gradual process - happening in controlled conditions. After all, the internet has existed for around 30 years. People are used to certain patterns, and it's not something that will change overnight. But we will increasingly converse with websites instead of clicking through everything manually.

Implementing artificial intelligence in a configurator is essentially aligning with that trend. On the one hand, there will still be some manual interaction - some clicking will remain - but on the other hand, more and more often there will be an assistant that asks part of the questions, and we simply respond - in writing or by voice - and it preconfigures the product for us. It selects certain options, marks specific controls, and then we can finalize or modify the configuration. So this is the direction things are heading, and it won't apply only to configurators. It will also affect many other aspects of e-commerce. It will influence how we acquire knowledge from any website - there will be less searching, less manual effort, and more information served directly to us.

That's the near future - one, two, three years - and we'll see significant changes. But it all starts with step one: enabling configuration. Because lacking a configurator at all will mean falling far behind what's happening in the market. Step two is making that configuration partially automated while still allowing personalization. That hybrid approach will offer a strong competitive advantage.

Szymon: Great. It sounds a bit like something out of a Lem novel, but we can see it gradually becoming reality. We'll definitely talk more about it in future WM Talks episodes.

For now, thank you for listening, and we wish you all the best. See you in the next podcast. Bye.

Damian: Bye.

Thank you for listening to this episode. For more interesting content, visit our blog at webmakers.expert.

FAQ