“When I first heard about AI, I couldn’t have cared less. No-one will visit IKEA more just because you have AI. But then we used AI and machine learning to develop a personalised, individually relevant search experience, which delivered an extra €700 million per year in revenue. OK, now I’m paying attention.”
Jonas Hessler, former Global Web and E-commerce Digital Manager, IKEA
An inside story of IKEA’s online transformation from catalogue selling to one of the world’s most successful
When I joined IKEA 15 years ago, it already had an internet presence – albeit a bit basic, online version of existing catalogue-based home shopping. Launched at the end of the 90s, it extended to 13 markets, so it’s fair to say it was quite early in having a significant, revenue generating internet presence.
At the time, I was working in a variety of global business development roles, including the pressure of running one of the biggest stores in Germany – with lots of products and lots of visitors. By the time I took over as Global Web and E-commerce Digital Manager, IKEA was well established online and expanded into more and more markets.
The complexities of scale
But delivering an interactive e-commerce experience had proven a real challenge – there had been many attempts and many failures, of course with some success too.
The issue was scale. You can’t really appreciate the sheer scale of IKEA until you are part of it. To put it in context, I remember an early planning exercise at a time when IKEA was using lots of birch. We realised very quickly that, if our online growth plans were realised, it would be impossible to fulfil the orders – we would have needed every birch tree on the planet.
Today, we welcome 2.3 billion visitors every year, and revenues have exceeded €23 billion – that’s just from online.
But getting to that stage was a journey, and every journey needs a destination.
Convenient and unique
For us, all the research, all the feedback, told us our customers were struggling to find what they wanted online – they came with buying intent, but were frustrated by the user experience.
That gave us a simple, overarching goal to guide all the innovations to come. Quite simply, we needed to create a simple, unique digital meeting with our customers. You can break that down into two parts – first, a simple meeting place in our terms means convenience for the customer.
The second, uniqueness, is more nuanced and begs the question ‘what is our uniqueness in the eyes of the customer?’. Let’s take payments as an example. Do we want to provide a unique payment solution for the customer?
No, absolutely not. There is no benefit in that. People like payment to simply address the convenience part.
But then we have other areas. For instance, how to convey our home furnishing expertise related to life at home. This is clearly an area where we want to differentiate and we said very clearly: “We want to be the leader. We don’t want something that is on a par with the rest, we want to be the best.”
That was the starting point and the focus for our online channel development – and that too falls into two camps. Convenience is about navigation, search and day to day merchandising – it has to be personal, relevant and easy. But across 10,000 products in, what became 30 markets, it is a massive task.
The uniqueness is much more about insight – from product recommendations at the basic level to campaign content and guidance at the higher level. Again, relevance here is crucial.
Manual, complex and inefficient
That might sound simple, but it was anything but. We faced a number of big challenges.
For one, merchandising at that scale, day to day is time consuming, resource hungry and boring. It’s poring over data and tweaking rules day after day. I had, and still have, a big team of merchandisers, but they were spending 85% of their time working on our product listing pages.
That’s horribly inefficient, but also starved us of the resources we needed to accelerate innovation.
Then on the technology side, we had all the platforms you can imagine. Five different front ends to maintain in parallel. To be honest, it was a nightmare. To do that, and keep up the development pace was extremely difficult. The development and release cycle was a massive area to address.
An intelligent solution: Starting with search
Then we came across AI and everything changed. Not at first – in truth, when I first heard about AI for e-commerce I couldn’t have cared less: “No-one will visit IKEA more just because we have AI.”
But then we started to look at how it could address some of our challenges – and that is what is important about AI. Not AI for its own sake, but what you can do with it – on the face of things, it could eliminate huge amounts of manual effort, and do the day to day merchandising quicker and better than a human – after all, computers are designed to deal with lots of data.
In short, it promised to free people up to focus on strategic, high value tasks, thereby accelerating our release cycle – and boost conversion by offering a more relevant, convenient experience.
So, we set out to try it in certain areas where we wanted to improve, with the starting point being “what is the value of every visit we have?”. Then we used AI and machine learning to develop a personalised, individually relevant search experience.
Almost immediately, we saw the value of each visit increase by 10%. Then bear in mind we are talking about 2.3 billion visits per year, so 10% equates to big numbers – it’s about €0.3 more per visit, which is almost €700 million per year.
OK. Now I’m paying attention.
Now I’m very keen to talk more about AI and machine learning because I can translate it into business benefits, and it is something that helps our customers to find what they are looking for.
Streamlined and efficient: Automated merchandising
The next stage of our journey was to look at using AI to automate our day to day merchandising, because that was really an area where we spent more and more time to keep up with the pace. Not surprisingly, the people working in merchandising the site fed back that doing it manually was not much fun, it was a lot of administrative routines.
So, we tested automated merchandising. That means using more or less the browsing history and AI to do the product listings.
In this case, the feedback from colleagues was curious. The immediate feedback we got was “It doesn’t work”. In reality the fear wasn’t that it wouldn’t work, it was fear of losing their jobs. So, my priority was to convey a very clear message: “We are not looking at cost cutting and getting rid of people. We want to automate the manual tasks to enable you to focus on more value adding tasks.”
Once the message got through, the results were again beyond our expectations. Conversion went up, while we removed 70% of our merchandisers’ manual workload.
Those merchandisers are still there today, but they work on much more value adding activities. The majority of online merchandising is automated and the merchandisers focus on the most important products and product families. They have time to apply their knowledge of life at home and their knowledge of products, to complement the suggestions delivered through automation - to take it from good to great.
It requires some additional effort to make that happen. But it is the automation of the majority of the pages and listings that enables them to do that.
Insight driven recommendations
Naturally, we were again very happy with the results, so we embarked on an AI recommendations project. We already had recommendations on our site but they were done manually and based on our insight, our perspective – not data.
So, we took the same approach; to have personalised, relevant product recommendations – essentially analysing browsing histories and using that to suggest products. Actually, as it turned out, that wasn’t just interesting for our customers, it threw up some interesting questions for our range designers too. Customer behaviour questions like “Why on earth do customers make a connection between this product and this product?”, could now influence product design.
From a commercial perspective, the effect of AI in recommendations was a 7% sales uplift for new recommendations panels, and an uplift of around 4% for panels that were previously manually delivered.
The big numbers
Make no mistake, these are tangible examples of how we can use AI to replace manual work, to enable humans to focus on value adding activities - to offer a much better customer experience and, of course, grow revenue.
If we look at what we achieved, there can be no doubt that AI was transformative for IKEA, and for my team. We managed to remove 70% of the manual workload while delivering a convenient, relevant, attractive customer experience – at one stage, we were growing traffic by 50% per year.
But, in the end, I am a retailer – I am after sales. AI merchandising delivered there too. By the end of our journey together, before I moved on, we had sites in 30 markets and were growing overall revenue by 50% per year – reaching €26 billion from online in 2018.
AI was either directly responsible – driving consistent 2-4% sales increases – or indirectly, by freeing people up to do higher value work and accelerate our wider innovation efforts.
And it was quite a journey.