Shahin Shahkarami leads a data science and ML engineering team at IKEA, driving transformative changes for one billion customers across the globe. His focus is on recommendation systems, ML search, and visual intelligence to personalise the omnichannel shopping experience. Prior to IKEA, Shahin delivered analytics solutions in the telecommunications industry, enhancing customer satisfaction at scale.
In this interview, Shahin gives us the low-down on IKEA’s data strategy and latest AI innovations. From a cutting-edge personalised recommendation engine to AI-assisted interior design, Ai has been crucial for IKEA’s enduring success as an omni-channel retailer, transforming the homes of customers from all walks of life.
Shahin explains the long-term vision for data and AI at IKEA, and gives us his predictions for the ways AI could transform the future of e-commerce and retail:
I work for IKEA Retail (Ingka Group), IKEA’s largest franchise. My work covers approximately 31 countries where IKEA operates throughout Europe, North America, parts of Asia, and South America. The Data & Analytics Team oversees AI within the company’s product teams, and deals with data foundation, governance, management, and insights for IKEA Retail, as a whole. We work alongside the product teams, or sometimes we work independently to build data platforms and other tech that the different product teams can enjoy and reuse. Recently, we opened an AI lab internally, which focuses on GenAI and applied AI functionality. Our goal is to have an R&D function across IKEA as we explore how new technologies can give new experiences to our customers.
Within my team, we have several specialist teams working on different aspects of innovation. One team is dedicated to what we call ‘retail data machinery,’ focusing on personalised recommendations, personalisation engines, and search optimisation. Another team specialises in insights, analysing the broader customer journey – both online and across our omnichannel platforms – looking at how people transition between stores, apps, and the web. We’re also investing heavily in building generative AI capabilities that will shape IKEA for years to come.
We focus on growth, or as we call it, ‘growth and range’, which includes how we present IKEA products to our customers online. That’s an essential part of IKEA – our products and how we personalise them. We’re here to make sure every customer finds the most relevant products.
Another key focus is ensuring IKEA remains a truly unique omnichannel retailer, where the IKEA experience feels seamless and consistent, whether you’re looking at our website, our app or you’re in an IKEA store – the goal is that these channels work cohesively, while staying true to the IKEA brand.
I also work on home imagination capabilities, which is about helping online customers imagine and better design their homes. We want customers to feel inspired and translate their home furnishing needs online in an interactive and playful way.
Our vision in IKEA is to develop high-quality data and AI products that enable accurate actionable insight and on the other hand infuses automate intelligence to enhance efficiency and performance. Specifically within AI, we are actively investing to build capabilities to maintain IKEA’s leadership within the home furnishing space.
It’s always about having a clear strategy of what we want to achieve. Where do we want to be in a few years’ time? Are there areas that we need to address as an online retailer, and also as a leader of life at home within the furnishing industry? The use cases and projects typically stem from two different angles. First, we consider our product with an overall view together with UX and engineering, to then understanding the product’s impact on the business, and the business goals we want to achieve. Then we consider how AI or data can help us attain that goal. That’s how we choose which projects to take on. And then there is an additional angle, which is to consider new capabilities. For example, with the GenAI situation, which opened up a lot of doors for us. We need to understand the tech and assess if it can solve our customer problems more efficiently or create new experiences for us. In those cases we start by experiementing the tech without having a use case in mind to begin with, then we see a realm of possibilities from there.
We do some opportunity sizing from there, and then see if we can embark on these projects moving forward.
The secret to good collaboration is ensuring we all share one goal. And within that common business goal, we have different specialists who work on achieving it.
We have our data and analytic coworkers embedded in different product teams. Sometimes we have our specialists fully embedded in certain product teams, because we know that will ensure the goal alignment and also the delivery of a certain data analytics, inside the product, or AI product. And sometimes we take more of a hybrid approach: having a central team that supports the larger product teams.
There’s value to having a standalone function in the organisation; in being able to connect the dots between different product teams that might be operating in silos, due to the company size. We can take a more holistic view, and then collaborate between different teams, acting as the glue – and the voice of reason –for the organisation.
Each product team consists of four parts, as we call them: PDEX (which is product), engineering, data, and UX. These four work together from the start until the end of the project.
Sometimes there are different flavours in the organisation. For example in PDEX there are products that don’t need a data person because you’re building a front-end tooling, but some products are more AI-heavy. And in this instance we’re involved from the ideation to execution.
That’s one of the favourite questions of any C-suite or board member: how do we actually measure the impact of data and AI? Because it can all become a bit abstract. Over the last few years, we’ve built a great experimentation and measurement culture within IKEA, meaning that everything we have for our customers, always gets A/B tests and different types of causal effect testing.
This has helped us to understand the value of the work of AI and our data scientists as well, because when you make a feature, you can always visualise the impact in that moment. But you can also do it holistically, at a larger level.
We also evaluate opportunity sizing – essentially, what the potential impact would be if we scaled a solution across all our global operations or removed certain organisational barriers. From there, we assess the return on investment (ROI). But it’s all about finding the right balance. For example, even if the ROI on a specific product’s search function is low, it’s still essential to have that feature on the website.
But for our large investments, we always ask: does it reduce costs by improving productivity or efficiency? Does it generate revenue, or enhance customer satisfaction? The initiative needs to deliver on at least one of these in the medium or long term. From there, we work backwards to identify specific use cases.
A great example of something that’s hard to quantify, is the fact that many customers who visit IKEA don’t use a shopping list, but they’re there for inspiration. They’re not just looking for a list of items to buy, they are seeking an experience. Because of this, we aim to have all these tools available to help customers get inspired, and then measure the impact of that inspiration – for example, if the customer is coming back or if they’re adding things directly. In these instances the ROI isn’t as tangible as you want it to be, but sometimes you need to address it in order to be an industry leader in this area. You want customers to come to you for inspiration as well.
I think one of the most successful examples has been our journey in recommendation and personalisation. Our product range is at the heart of IKEA: the stores, the website, everything revolves around it. Our vision to help every customer access the most relevant range, the most relevant content, the most relevant services – this has been really important. So, we built an in-house recommendation engine.
This engine addresses a really complex set of recommendation problems, because we’ve got so many different journeys, different channels, different experiences, and you want to build something that’s effective and reusable, and also cost-conscious and scalable at the same time.
Essentially, our recommendation system consists of more than 15 AI models that work together to power different recommendation panels that we have on the website, and in other locations you might not associate with recommendation, like a chat or a 3D setting. The engine can offer the right product for a customer in all these contexts. And today approximately 40% of online visitors to IKEA are using it and it’s generating both direct and indirect revenue across the website.
In the past, we used to buy these recommendations from an external company. It’s been a strategic direction for us to build them all in-house and in a way that IKEA can generate the right content and products for different types of customers.
The way I see it is this: if the capability we’re developing requires complete control and a clear understanding of everything that goes in and out, then it’s something we need to build in-house. For example, IKEA is different from other retailers because it sells its own product online in many countries (compared to a retailer like Amazon, which has an open- ended number of products). So, the complexity and challenge is thereby different, and we need to build these things internally to offer the best service to our customers.
The main challenges were twofold. First, there were technical challenges: we wanted to build personalisation for our customers, but did we have real-time data for our customers to personalise things in real-time settings? Could we serve these models in a scalable but cost-conscious way? And from an algorithmic perspective, how should we design this system to address all these journeys and channels in a way that’s reusable and functional?
We solved these technical challenges by having really smart people that work endlessly to solve these problems. We focussed on our data foundation, for example, by ensuring we understand our customers and customer data, setting up goods and the ops setup structures. And from an algorithmic perspective we gained a lot of knowledge, whilst solving these problems in an innovative way.
The second challenge was to overcome the organisational issues, in particular the change management side of things. Moving technology is easy because it’s logical: you move from point A to B. But organisationally, it’s always challenging when you’re coming up with new ideas that are changing parts of the company, fundamentally.
But thankfully, we’ve achieved a high level of collaboration within the organisation. We have a clear vision of where we want to go and we can help guide everyone towards that.
I have a deep technical background in the space, and I’ve been really interested in AI since my academic experience. But when it comes to these complex systems, I always look at how they can be of practical benefit. And that’s where maybe my product owner side comes into play: really looking from the customer perspective. Does this really improve customer satisfaction? Does it really help our customer?
I sometimes feel the role I have is more of a translator: taking some really complex technical concept that might involve, say, the cost function of neural networks, and translating this for a senior leader or different department within the company towards business priorities. And then conversely, explaining some business idea which might sound really abstract to someone who works in data science because there are so many ways of doing it, and translating it back into technical language.
A next potential case is how we can help our customers better imagine their homes. Because the essential problem we’re trying to solve is this: if people buy product X or product Y, does it really fit in their room? Does it fit in with their style? Does it fit within their budget?
That translation is really difficult to achieve, because people’s homes are complex. People are complex as well. And that’s where we’re utilising AI to help customers imagine how the products will look, and how they will match their personal style.
There are different types of capabilities such as scaling designs, which give customers these tools to help them place products in their rooms, or to re-design the room, and to give them different options so they can compare styles. Our ultimate aim is to simplify the process of designing a room. Interior design might be easy for a group of our customers, but not for most people. On the other hand, people do have a feeling for what they want and what they like, based on their needs and desires.
So in summary, that’s the use case: interior design, toolings, and 3D settings that are very, very interesting. And that’s what the team is focusing on at the moment. We’re still in the early stages of investing in and understanding these capabilities. We’re committed to doing this responsibly, ensuring that everything we create prioritises the safety and well-being of both our customers and our business.
As a GenAI use case the project has huge potential. Why should you buy this pillow that’s white and square versus the other pillow that is also white and square? There’s this whole angle of better storytelling and answering customers’ questions, helping them when they’re deciding and choosing products.
And for certain types of products, you need to help people visualise the end design. Let’s say if you’re buying a kitchen and you’re spending a significant amount of money. Or if you just bought a house and it’s completely empty and you don’t know where to start. It’s all about removing the fear of design for the customer, by helping them imagine what the room could look like with these different styles and sets of products.
So, with these capabilities and technologies we’re building, we are using GenAI to scale that design. But on the other hand, we are also doing a lot of engineering and tech work to create these toolings as well.
This might sound cliché, but IKEA’s mission of creating a better life for many, is honestly something really iconic. And I think AI can help with that. AI has huge potential for helping every single individual from different budgets, from different backgrounds, different stages of life. You could be a student, you could be a young professional, you could be retired and use these capabilities. That’s really exciting!
And more specifically within my own domain, what really excites me is that there’s a lot of movement in spatial computing, spatial understanding, of the interior space. GenAI capabilities are getting a lot better at that. When exploring that area of spatial computing for IKEA, there are a lot of the models that have become cheaper and better. There’s the option to offer hyper personalisation at scale and give new experiences to our customers. And storytelling our range to our customers is another topic that really excites me.
There are three main trends. One is, of course, AI acts. the EU AI Act, for example. How that impacts an organisation. IKEA has an immense focus on that; we’ve been investing in responsible AI, and we’ve got responsible AI teams organised within the company to become a leader in this space.
The second is the future trends of productivity at work. I think all these productivity tools can really enhance the day-to-day work of a lot of our coworkers in large organisations. At IKEA, for example, even if you move productivity by a few percentages, it will have a massive impact. If your coworkers are more productive in answering customer calls, in answering customer questions in the stores, or maybe even if software engineers are more efficient in the way that they write code – that will drive meaningful transformation.
And the third trend is GenAI. It now has better multimodal capabilities that are cheaper, safer and more secure. It can be used on your device and not in the central clouds. These models are evolving into having better reasoning, and getting better at solving complex tasks. There are a lot of new applications that will come with GenAI. I believe we’re on the verge of a revolution in e-commerce and retail. In five years, we may be embracing experiences we can’t even imagine today – experiences that will soon feel as natural as the ones we rely on now.
