Why Doesn't Product Search Work?

Introduction

Mitra: Welcome to Syntra and today's episode, which is all about product search. If you haven't listened before: this is the podcast for people driving their function forward who want to understand how the relationship between systems and users shapes the way we work. We share insights on how new technological possibilities are transforming the way we work and interact with technology. Today I'm joined by Glenn Svanberg. Welcome back.

Glenn: Thanks so much. Super exciting to be here.

Mitra: And a really fascinating topic we're diving into today. We're also joined by Martin Skarin. Warm welcome.

Martin: Thanks, thanks. Today's rookie, maybe. Exciting stuff.

Mitra: Ha! So who are you, Martin? What do you do at Fiwe?

Martin: At Fiwe I work a lot around e-commerce. I've been here for 20 years, and before that I was also involved in e-commerce from the late 90s.

Mitra: That's fascinating. You've had a chance to see a lot over those 20 years. Search is clearly something close to your heart, so it felt like a great opportunity to dedicate one of our episodes to product search specifically. Let's kick off with the first question: what do we actually mean by search? What do you say, Martin?

What is product search?

Martin: For me it's pretty obvious – though maybe I'm a bit too close to it. But in short: the little search bar that exists on every site, where you type one or a few words and expect to get back a list of products. Simple as that.

Mitra: That simple.

Glenn: Yes. It's actually a pretty clear definition. You have some kind of text field where you can type whatever you want and expect a list of products in return.

Mitra: I agree. But it's also worth being clear about what it isn't.

Glenn: I notice a lot of people confuse it with navigation, or with some kind of chat-based solution for finding products. But that's not what I mean by product search.

Mitra: Exactly.

Glenn: Product search, as I mean it today, is a search box and you get back a list of products.

Mitra: Right. I'm curious though – I'm one of those people who tends to search through a chat. Why isn't that product search?

Martin: I think for me it comes down to the fact that the search box is so ingrained in how we work, and a chat is really a dialogue. Product search gives you a very quick, direct answer – it doesn't take as much time. It can be really efficient. But it can also put me in a bad mood as a user – if I feel like the results aren't aligned with what I'm looking for, it can make me do something else entirely and give up. The dialogue is what makes the difference.

Mitra: Right.

What makes search good or bad?

Mitra: What makes a product search good or bad?

Martin: A good product search, I'd say, is one that becomes your natural way in to finding products. It's where you choose to go – you'd rather search than browse. You'd rather go there than to a chatbot or anything else. It becomes the natural entry point and you trust the results. You get fast, accurate answers to what you're looking for. That's a really good product search.

Mitra: I think that's exactly right – it should be faster than browsing your way there. But for me, having worked a lot with data, I think you have a good product search when you as an organization are proud of it. That's a good product search.

Glenn: And a bad one?

Martin: It's a search you don't fully trust. And that makes it completely unusable. When you constantly get vague results and realize the right products aren't surfacing. Picture coming to an e-commerce site and seeing a campaign with products on the front page. Wow, that looks interesting. So you search for one of those products from the campaign. You search, and it doesn't come up as a result. You'll never use that search bar again.

Glenn: You know the product exists, but the search isn't consistent. You'll just find another way in to get to the right product.

Martin: And I think that rings true for myself and for users generally: if you don't get the right answer, if you feel like you got a result but it's not the right one – that's the worst thing I can imagine. If you can't find what you know exists, you go somewhere else. There's a feeling of "this isn't what I'm looking for." There are so many ways to have bad search.

Glenn: So many ways to get search wrong. Zero results – that's clear, you get nothing. That's a poor product experience, but maybe not as bad as bad search, because I react quickly and understand I didn't get a hit. I can work around that. But if I get 60 results that aren't what I want – that's really bad.

Mitra: So it's actually worse to miss the mark when you do get results?

Glenn: Yes. Better to miss completely than to serve up the wrong answer – it's far more misleading.

Mitra: Why?

Martin: Because it leads me to think I'm in the wrong place, with the wrong supplier, and I give up. In the worst case, I've got a list of things to buy, but because I can't find one specific thing, I take my entire order somewhere else – even though it might be a supplier I visit regularly. Sometimes I'm in a hurry and I need that specific item today, and that's what makes search so important in that moment.

Mitra: That's really interesting. How often do you both encounter bad product search? Is it common?

Martin: I don't think I've come across a search I'd call genuinely good yet.

Glenn: Google is genuinely good. You can always find a search that could be better – there are nuances. I catch myself searching for something and not quite getting an exact match, just something roughly right. And then I end up going into a product, following the breadcrumbs, and navigating through categories instead. Even in well-known tools I sometimes take shortcuts because I'm not used to them. You've essentially trained yourself to work around bad search.

Martin: You don't get the results you're actually after. But you can use search to get to a category where you can then find the right things. Somewhat like that.

Mitra: And for those of you not in the room with us: the looks we're giving each other say a lot – there's a strong shared sense that it's pretty common for search to just not quite work.

Why is it so hard to improve search?

Mitra: Which makes me wonder: why is it so hard to improve search?

Glenn: It's a problem that comes from so many different places. Search is the ultimate expression of your product data. If you don't have complete data, you have no fair shot. There's no way to build good search – you need structured, well-filled data to have a solid search corpus to start from.

Martin: Exactly.

Glenn: And most companies don't have that.

Martin: No, and I find that it's a kind of pipeline – the foundation of the data often comes from the manufacturer. But that's not where we meet the end consumer. Data moves closer and closer along the chain, more and more products, closer and closer to the consumer. By the end, both the language and the volume and the content of the data are incredibly fragmented. And nobody worries about search, because it only shows up as a problem at the very last stage.

Glenn: And the problem lands in a part of the organization that doesn't have the capability to fix it, so people end up pointing fingers at each other. "Other parts of the company should have sorted this out."

Martin: That's how it is – it spreads across many organizations, between those responsible for the product data, those who populate the data, and the suppliers who write things in different ways. And so on.

Who owns search in the organization?

Mitra: If we think about what this looks like in practice inside companies: who, or which function, ends up responsible for search? Whose job is it? We can talk about how it is versus how we think it should be – because those are often very different things.

Glenn: Usually it lands with whoever owns the system people search within – meaning e-commerce or some kind of content platform, but most often e-commerce. They have some tools, but far too limited tools to build good search. You can tweak things a little, I'd say.

Martin: I'd agree that it usually ends up there, but it really should sit closer to the product information world – because search is ultimately just an expression of product information. It's not about how you visualize things; search is an expression of your information, and it belongs closer to the product information problem than the e-commerce problem. But at the same time, those who maintain product information might not be able to solve it on their own – it might fall to procurement to push suppliers for the right information, or to an industry organization. So there's really nowhere that brings everyone together to address a problem that cuts across all these departments.

Mitra: It sounds like it's not black and white which function should own it – more like a combination and interplay between procurement, people working with product data, and those responsible for the visual side of e-commerce. So it sounds like you're advocating for a new kind of cross-functional setup here.

Glenn: You need to understand what the actual problem with search is. Many systems sell themselves as including great search, and there are plenty of search engines out there. If you don't work on this organizationally in the right way, and instead just hope the IT solution will fix everything, you'll keep running into the same problems – and it's almost always about product data.

Language and search

Mitra: Where does language come into this? We use different words for the same things. How do you bridge the gap between how customers talk, how suppliers talk, and how we talk internally?

Glenn: I imagine that must cause a whole lot of problems – speaking different languages but trying to find the same thing.

Martin: That's really where you get into the fundamental mechanics of how traditional search engines work. They're essentially just text-matching machines. They don't look at the meaning of words at all – just at which characters appear next to each other. If it's the same sequence of characters, it's a match. An unusual sequence makes it more likely to be what you're after. But that's not actually how we search.

Glenn: I have so many examples of this. I think a lot of people assume professional buyers use words that match the supplier's terminology. That's just not true – people tend to use much simpler, everyday language when it comes to products. That said, when you're dealing with professional users, there are certain niche terms that matter. Here's a small example: "folding ruler" is a word most people know. My dad might call it a "foldable". I'd probably say the same. But the technically correct term might be "measuring rod". Typically you'd just add that as a synonym in your search engine and move on. But the problem is there are so many words like this. And if you're a professional, you know you don't just want any folding ruler – you want a contact gauge. From one supplier it's called a "contact gauge", but the next supplier calls it a "contact gauge rod". You might not realize those are the same thing. A search engine might try to split the compound word, but that leads to false matches. You need something more to understand what the product actually is. And if you specifically need a contact gauge – the kind with the scale on one side so you can lay it flat against a surface without it tilting – you need the search to surface that. One supplier uses "measuring rod" in their search and understands what a contact gauge is, connecting "contact gauge rod" to the right product. But you get no results if you search for "contact gauge", because that term simply isn't in the search engine. And it can't go the other way either – someone who doesn't know the technical term might not know what a contact gauge is, even if that's exactly what they need once they see it.

Martin: Don't all these problems come down to the fact that traditional search engines are a bit dumb? They only look at the text itself, not the meaning.

Vector search and AI

Glenn: What if you bring AI into the mix – some intelligence – and introduce something like vector search, where you can measure how similar words are in meaning? "Contact gauge" sits very close to "measuring rod" and "folding ruler" if they're in the same vector space.

Martin: Yes.

Glenn: They're related. And that's what I should be getting results for.

Martin: I think you also need to understand it in context – some of these words might not be something AI with vectorization alone can handle. Looking at words through a vector search engine might give better results, but you can't just match words against what's in your product data. You need to account for what people actually search for. There can be many words and phrases that match nothing in your catalog, but that matter a lot when enriching your products.

Mitra: I'm curious – vector search, tell me more. What actually is vector search?

Glenn: Let's go there. Think of a little universe – a box with a universe full of stars. Each star has a position, a coordinate in that star system. What vector search lets you do is give every word a position in that kind of universe, and then check which two words sit close to each other. You can do that mathematically – calculate the distance between two words – without looking at how the letters are arranged. By looking at meaning instead.

Mitra: Do you have an example of where vector search has been useful?

Glenn: Yes, we just had one. It's a classic example of where you wouldn't need to manually create a synonym saying these two words mean the same thing – because they already appear together in the same documents. When you train an AI to become a vector search engine, you train it on a large number of documents and let it learn which words appear in the same texts and how closely they relate.

Mitra: Let me extend the example around vector search and different ways of communicating: why is it that search within a company's own system can be so much worse than Google? There's no dialogue there either, but it just works. Why do I always find what I'm looking for there?

Glenn: You can throw in several words and Google pieces it together and gives you a result.

Martin: Google does exactly this – they have AI intelligence built into their search queries and genuinely understand the intent behind your question. But they have different muscles than most e-commerce companies. For one, they process billions of searches every day, so they can build an excellent feedback loop around what you actually click on. If you don't click on anything, that was a bad result. If you click the top link, that was a good result. So they can fine-tune at a scale we simply can't match. Some search engines do have that kind of feedback loop, but I don't think it's enough on its own. Another thing Google has is that product suppliers buy keywords. When someone buys a keyword and links it to their products, Google reads that as a signal – this is probably a good word for what people actually call this product. That partly explains why you find things on Google even when you search using a word that doesn't match the product's official name – Google has learned the connection.

How do you start improving search?

Mitra: But how sustainable is that really? I'm thinking about all the companies listening who are wondering how to improve their search. Where do they start?

Glenn: I think you can go a long way with traditional search engines and the solid work that's been done with them. You can't just throw them out and say you're switching to modern AI vector search – that's no silver bullet. Look at screws, for example: a 4.5mm screw sits very close to a 4.6mm screw in the same vector space. But I might actually care about exactly which screw I get. You need traditional search for that. Vector search is great for the fuzzy, semantic side of things, but for numbers you still need traditional search. I believe in a hybrid approach – marrying the two together.

Martin: I agree. I think you need to look at what people are actually searching for on your site and examine what products come up. That's difficult when responsibility is spread across many people, but I believe in using AI to process that data and surface the search terms that actually match. Though that requires a meaningful volume of data and some careful tuning – what works for one assortment might not work for another. It's often the case that you optimize one part of your catalog and end up undermining a long tail of other products. It's tricky. Somewhat dependent on what you want to focus on and where the biggest return is.

Where is search heading?

Mitra: Fair enough. But Martin – with your remarkable 20 years in the industry, having seen so much change. Where do you think search is heading?

Martin: I think we're at an inflection point. Looking back 30 years: it started with database search, which around 2005 evolved into a search engine with pre-indexed terms – fantastic progress. Then we learned to work with it and realized we needed to layer in sales data and behavioral data. That's where we are today. And now we're quite taken with vectorization. But it's something you have to learn: you can't vectorize numbers the same way as words, and it can be language-dependent – handling Finnish differently from Swedish. So it can be demanding. I think the pace of development over the past year points to the next step being about pre-processing product databases in a way that puts search at the center. We're going to stop manually maintaining categories and search terms – because no one can keep up with that long term. You have a product catalog with a lifecycle of maybe 3, 4, 5, 6, 7 years. Maintaining search terms manually throughout that is an enormous job. I think that's work that's going away. It can largely be replaced by letting AI pre-process the data and introducing vectorization as part of the search mix – specifically to bridge the gap between how customers talk about a product and how suppliers want to describe it. It's not the same language. There's a mismatch, and we need to build tools to handle it. I think that's where we're heading.

Glenn: I think a lot of people who've worked deeply on search know what they should be doing, but don't have the energy because it involves too much manual work. Now is the time to step back and reflect. You know what you could improve, and you can let AI handle the parts it does better than people can manage at a reasonable pace.

Martin: Whatever way we solve it and whatever technology we get help from – let search bring it all together. I as a user, with my unique experience and behavior, have a need and search in a specific way. That's going to look different for someone else. It comes down to exactly which word I choose.

Glenn: Exactly. I want the technology to learn from my previous behavior patterns. It could start with... Mitra, you've been buying quite a lot of cream and oat milk and a mix of flour – maybe you should try making this recipe, based on your behavior. That idea of a system learning what I've done and eventually suggesting something I hadn't thought of – it exists a little today, in quite limited form. And it can get a bit odd sometimes.

Martin: It can get a bit odd.

Glenn: What we talked about earlier – being able to deliver fast, relevant answers – sets demands. You need the data ready to support that. In some cases you might have a narrow assortment, a type of product with limited repeat purchasing. That's possible, but it can still feel strange. We've seen it from time to time. But for other types of assortments, it's more viable. You compete based on what's most efficient – sometimes a reorder page showing your 40 most common products, your favorite item, what you buy every week, works better. It depends on the context.

Martin: It's about knowing your users and understanding what type of user you are – that's central. And that's one of the things Google does extraordinarily well. They've invested enormous resources in getting users to stay logged in. Then they know who they are. They give away as many free services as they can because having a Google account is so valuable – so you're logged in, they know who you are, and they can use that to improve your search.

Why should companies invest in better search?

Mitra: That makes me think. Let's close with one final question – or actually two, I realize. First: do you think search is an important question?

Glenn: Yes.

Martin: Yes.

Glenn: Critically important.

Martin: It's a real moment of opportunity.

Glenn: I can't wait for better search.

Mitra: And then my follow-up: why do companies need to think about fixing their search and having a good one?

Glenn: I think it creates internal conflict within the organization. Because when nobody's happy with it, people start pointing fingers. It's your fault – you, the category manager, should fix it. It's an e-commerce thing. But it's the web system that's broken. And suddenly you've got a blame game, which leads to the people who are actually supposed to be selling things not wanting to use the digital channel – they don't recommend it because they feel it's not good enough.

Martin: We're in 2026 now. Digital channels are actually...

Glenn: Right!

Martin: But for them not to be a primary channel feels completely absurd to me. And having a system that understands you and actually delivers the information you have – so customers can find your products and buy them – should be a no-brainer. It's 2026. We should have systems that understand us.

Glenn: I agree. You need to understand how to work with these tools. We've been trying for the past 23 years, and it's led to silos without really getting it right. Maybe things will finally take off now.

Martin: There are so many things competing for attention – sustainability regulations and everything else.

Glenn: Yes, a lot is competing. The question is how you know when it's time to actually do something about your search.

Martin: Users are lazy by nature. We have extremely little patience. I think you should take every opportunity you can to make it easier for users to consume information and ultimately buy products. The tools are there now. There are completely new capabilities. It's a paradigm shift – systems that can understand users in a fundamentally different way. If there's ever a time to start, it's now, when we have new possibilities.

Glenn: I also think about the many companies that have efficiency as a core strategic priority – leveraging automation and everything it can do. In that context, it feels like a real missed opportunity not to focus on good search. There's so much potential being left on the table – search is a direct path to efficiency. It's an extension of the sales channel, as Martin touched on, and you also gain by learning and understanding what the market is actually looking for – so you can develop the products and services that meet that need.

Martin: But I think it's difficult, because as we saw earlier, it usually lands with whoever owns the system people search in as the primary responsible party. That might not be where you should attack the problem. It might be someone else – whoever owns the product system, or whoever talks to suppliers. But their plate is also full. They want to get data in, work with large volumes of products, get to market quickly – and search sits very low on their list. Why would they prioritize finding better help with search terms when it's not high up for them? I think it needs to be escalated to the central level of the organization. Because when you go down a level, you end up trying to fix problems that don't actually move the needle on search.

Glenn: AI is already in every leadership team, high on the agenda, and everyone is pointing to it: we need to work with AI. And the way they do it is by introducing a chatbot that everyone now has to sit and chat with. That does nothing to improve the experience on the site. Chatbots are incredibly slow – they take seconds to respond and spit out long walls of text. That's not the answer people need.

Martin: I think we just need to find the right pressure point. If we're going to work with AI, let's do it for real. Can we bring AI into search itself? Can we use AI to prepare the search corpus so we have a solid search engine? Can we introduce vector search so the system actually understands? That's how you work with AI for real. Google has been an AI company since day one – that's what they were from the start. But expressed through search, not expressed through a chat, which has become the visible symbol of AI.

Glenn: Exactly.

Wrap-up

Mitra: And on that note, Glenn, you're leaving us with a big thought. I don't know why, but I keep thinking about a tree – the root system and the leaves. That we keep treating the leaves instead of going down to the roots and addressing things there. But we're out of time for today. We could have kept going with this for so much longer. If you're listening and have thoughts or feedback on more angles related to search – or any other topic – please do reach out. A huge thank you to Martin and Glenn for being with me today.

Glenn: Thank you so much.

Martin: Thank you so much.

Mitra: That was great. Goodbye.

You've been listening to Syntra, a podcast from Fiwe.