Sjölin welcomes us to Boozt’s Copenhagen office, called Innovation Lab, where some of the e-tailer’s 210 developers work with app development and new innovations. Now the CTO of Boozt Fulfilment & Logistics, he’s been with Boozt for almost 14 years and has experienced days when the company has been close to bankruptcy and days, especially during the pandemic and its rise for the e-com sector, when the growth seemed unstoppable.
— It’s been quite a ride, he summarises. We try to apply all our learnings now, being a multi-billion Swedish company. We have 60 nationalities and all their different backgrounds creating an atmosphere where we learn things from the people we bring in. We come from a startup environment and we still like to think as a startup company.
In his role, Sjölin is responsible for the coming payment gateway, called Booztpay, and the Fulfilment Centre in Ängelholm, Sweden (pictured above). It houses 1,200 robots, 1,2 million bins, and the world’s biggest Autostore solution, and the capacity is set to further increase with the just-announced 44 million Euro investment.
— During the pandemic, all the KPIs and everything was screaming ’explosive growth’ in everything we did, so we planned a massive expansion. Now, when we’re settled in, our indications are 5 to 15% growth which the expansion in Ängelholm is set to fuel. From a logistics point of view, our main target is to support whatever the business wants to do, including having enough space for items, but our unique selling point is to send out goods as fast as possible to consumers — in 1 to 2 days. Even when we’re growing, we also need to be fast, which is very, very difficult to do on the scale that we’re doing, when you have 11 million pieces in the warehouse.
— It’s easy to misjudge the speed of things. When you walk around there (in Ängelholm, Ed’s note) every day, you only see the problems and the things that do not work. But when you bring people from different companies there and guide them, they’re mind-blown about the Autostore technology, the automation, and how everything is flowing. Autostore is a Norwegian solution for compact storage. It’s a matrix of plastic bins standing on top of each other, like in a 3D compartment, so it’s very dense in storage, which we like. Then, you automate it, so the people in the warehouse don’t need to go and fetch items. The goods will come to them instead and they’ll pick it out in a bin and put it in a bag, put it on a belt, and it goes out through the warehouse. The fastest order we’ve had, I think, was 63 seconds from being placed on the checkout to be ready to go out through the warehouse doors. But that’s also the fastest, so our developers like to remind us that everything else has been slower…
Will further automation in Ängelholm replace employers?
— We’ve had automation since 2017, and we’ve never reduced the number of people because of it. However, we might not have hired as many people as we otherwise would have. Our operators are instead developing themselves into new things, such as joining our service and maintenance team and being educated to maintain the robots. We do different career paths for them — it’s super fun to see people gathering new knowledge, and also helping us to be more efficient in the systems. We have 400 employees in the warehouse. If we can get 400 people to think about the problems we have there, it’s amazing.
Sjölin recently went on a US tour visiting four different warehouses; Lululemon, Crocs, and Levi’s, plus Dillard’s — a brand store, similar to Boozt.
— It’s a fascinating country. They do everything big — but it’s also nice to know that we did it bigger in automation — and it’s great to compare KPIs to see how you benchmark against the rest of the world. Since we’re not competing in all areas, we can help each other and exchange information around logistics and tech.
An exception to the lack of cooperation in the creative sectors, not least in fashion.
— Yes, and we gain so much knowledge when we find people who are transparent, open, and don’t have any prestige. It’s an honest discussion about what is working and what is not. People are not stealing straight off — it’s very difficult to do a copy-paste there — and you get golden nuggets that you can use.
What were your takeaways?
— They still use a lot of manual labour. A workforce in Scandinavia is more expensive, in general, than in the US, especially in the Midwest. So, anytime you do automation, you try to balance the return on investment. What’s the cost that we’re replacing versus what we’re investing? Sometimes you do manual labour because it’s more cost-efficient. It is tricky to get automation to work super well — humans are far superior in some cases, like return handling.
Boozt aims to be a pioneer within so-called LogTech, short for ’logistics technology’.
— It’s about the traditional logistic operations and how you work with them, and apply modern software development. The automation hardware won’t work unless you have very good software as well, so we try to facilitate the bridge between those two. If you look at big software development companies like Google and Facebook, they’re never down. You’ve never seen Google down — it just works. We try to do the same thing and not be down in the warehouse. Since everything is web-based, you just plug in your computer and go to work. We do things fast. We like fast. We do around 100 deployments per day into our live systems.
100 per day? What could that be?
— Anything. Change the colour of a button, move some text, or change the whole routing system. You don’t take down the site to do maintenance and put it back up again but incremental small changes all the time. Facebook or Amazon do them as well. A lot of them come from the operators — they go around in the warehouse and if they get an idea, we will implement it and try it out together.
What’s the key tech in LogTech? AI and robotics?
— A great example is when our software is enhancing an existing hardware solution like we do with Autostore. Through our tech team, we aim to create a new benchmark in the industry. It’s a collaboration. AI is hyped now, so everyone in the tech world is asked about it and how to use it. We’ve used AI for the last 10 years. For us, it’s there, and it’s a tool to use. Now we have ChatGPT and generative AI, which has changed the game for a lot of things, especially in content generation. We have a team in Aarhus, Denmark — including my colleague who is an astrophysicist working with AI — who are doing a lot of machine learning in AI.
— AI becomes super powerful with the amount of data that we have, which is very hard to make sense of sometimes, so you can go and look into certain patterns. With all this data, how can we apply machine learning to it to give us good insights? How can we change our routing algorithm? How can we change how we do different distribution cut-off times? That helps. It’s a tool like everything else.
Is AI the most useful in data management for you?
— It is. However, I’m old school. No one is talking about databases now, because everyone has a database, but for me, that’s a better tool than AI. You need to store stuff somewhere efficiently and then you have AI to crunch it but the technologies need to work together. For a developer, we always say that we like generalists — people who are curious about a broad area — rather than specialists.
Can you also use generative AI in LogTech?
— It’s very intertwined. We have data that we gather in the warehouse for the website. We take images in our photo studio and based on those, we generate product descriptions from AI so that the generative AI can help us do product descriptions. It’s all connected and, as I said before, logistics supports the business. If we can use AI to help enhance the content for the rest of the webshop, it’s a tremendous gain in value. Or to identify colours from an image, so that you get the colour selections correctly on the website. If you go in and you only want to see blue, you can get that from the image instead of having people manually sitting and setting colours.
What are the next steps here?
— The next steps are interesting. We all know ChatGPT but when you have an even more thinking AI, we take the next steps. Now, you need to prompt it in a perfect way to get some kind of results, but AI will take further steps. We’ve talked a lot with Google and they have specialised AI that we can train on our own data, which all the companies are already doing. There’s a danger with ChatGPT too, because when you start asking it, you’re also giving away data. If we’re asking, ’How should Boozt forecast the data?’, then someone else can ask, ’What is the Boozt forecast?’.
— We had an AI expert talking at our internal conference and he talked a lot about how Europe is much better at protecting data rather than, for instance, China where it’s the Wild West, or in the US, where everyone is just in competition. EU is a good pot for building good, safe data and AI.
The development might become slower, but that can be a good thing.
— Yes. Sometimes it’s going very fast, and sometimes it needs to also go slow — and it needs to be good for the humans as well.
What other projects do you work with?
— One with distribution chains. We are responsible for our warehouse, but we also have all the distributors that we’re not really controlling. But for the customers, when they buy from Boozt and the package is delayed, it’s still our fault. So, how can we collaborate around the data here? Can we do it differently? Can we identify bottlenecks in other distribution chains?
— We have three different Autostore cubes — one with 530,000, 450,000, and 250,000 — so a total of 1.2 million bins. This creates a problem when items are physically in different cubes. We are currently working with hardware and software to create a smarter solution. We will combine these cubes through conveyors and automatically move items between them so no humans are involved. Especially when we are approaching our ’Super Bowl of ecommerce’ — Black Friday — this will be an important help for us in order to keep our delivery promise.
An important part of LogTech, Sjölin shares, is to simulate certain developments when working with digital twins.
— You make a digital copy of your physical warehouse, type in all your data in it, and can then simulate, for instance, the peaks. You can see how the warehouse works depending on how you move different items around, and if you’re more or less efficient. We’ve talked to a lot of small companies that have different ideas around digital twins and we are definitely looking into it. We are moving towards building simulations of different areas. The concept of digital twins is really appealing where you can try out different things in a digital environment without affecting production. We have a lot of ideas we want to try out fast and safely, and digital twins fit perfectly into that strategy.
Yes, because you mentioned that you make 100 changes a day in the warehouse, which you can then also implement into your digital twin.
— Yes, then you can do that (in the digital twin) first and you can see the impact before you start moving things around. We don’t want to move items. We count touches and want to have as few as possible on each item. We have one project called The right item at the right place. When you place an item, you should place it in the right place immediately and not move it around five times. That’s also something you want to do with a digital twin. If we place the item in this configuration, how will that impact the outbound capacity that we have?
That also seems like a place where AI can help.
— Definitely. You take all that data, you build it up, and you start to manipulate it at a very fast speed and see how it turns out.
Fascinating.
— It is fascinating. In the warehouse, we also look at pattern recognition. You can almost take an image of the warehouse and see all the items and where they’re put. Then you do image recognition and based on a certain pattern, you get a certain output. When you have 100 million different patterns, you see what kind of output they have. Then you can start figuring out which would be the best pattern.
One of your main challenges is sustainability. Can LogTech be used to increase it?
— Yes. If we can meet different pickup times, we can have more complete trucks, so we don’t send air. If we reduce the number of robot movements — which would be good for the business as well — we won’t use as much electricity as we would have. By handling both shipping and returns in one location close to the customer, we minimise environmental impact and ensure fast and efficient deliveries, all in a single shipment.
— We have a very strong team at our fulfilment centre that constantly asks ’why,’ and as a result, uptime has greatly increased. Technology and robots minimise the risk of incorrect orders, leading to fewer returns. Fewer mistakes mean fewer returns.
What else do you forecast in the future of LogTech?
— When you talk to different logistic companies, without being too controversial, not that many are great at software development. There are a lot of benefits to absorbing all the great developers and engineers out there that can make a tremendous difference in your company. We do all of our systems ourselves. We say that we have the best systems in the world because we’re fairly sure that no one else can run them. They’re tailor-made for us, says Sjölin. He continues:
— It’s not a recommendation for small companies because it’s difficult to build everything yourself when you’re small. But when you grow to a certain scale, you need to be in control of your businesses and add AI to it where it makes sense for your own business and not just because a supplier introduces something.
— As with all great technology, it is a collaborative process and effort. The more transparent you are about your problems and solutions, the higher you can aim. From a LogTech point of view, the best tip I have is to start working on your own solutions, not replacing all the systems that you have. Start with the small things. Take data gathering, for instance. Are you in control of your data or do you have all that data in a third-party system somewhere where you don’t control it? Even if you think you control it, you don’t. If you gather your own data warehouse, it’s also easier to change systems and get knowledge and insights. It’s also easier to start working on your own systems. Control is good — and to get great generalist engineers is very, very good!
