An IKEA warehouse with shelving numbered with large signs and a person driving a forklift.

Using artificial intelligence for smarter demand forecasting

IKEA has developed an advanced tool that can significantly improve the accuracy of its demand forecasting. The tool uses artificial intelligence, and existing and new data to offer highly accurate forecast insights. The tool is currently rolled out in Norway. We talked to Peter Grimvall to find out more about the new tool and how it helps IKEA plan its demand and supply better.

Peter is almost like a mathematics teacher. When presented with a big problem, he breaks it into small pieces, finds solutions for each bit, puts them together and voila, there is an answer. So, on a late Friday afternoon, when Peter was asked about why accurate forecasting matters to IKEA and how the new tool – Demand Sensing – will help it improve, he breaks the big technical maze into small bite-sized pieces.

“Demand planning, which is a forecast on how much we are going to sell in the future, is a significant number for us”, says Peter.

Think of it like this – demand forecasting means estimating all the products that more than 450 IKEA stores and ecommerce across 54 markets will need at various times of a year. The fact is these numbers could be a couple of billions of products. And, if there is inaccurate forecasting, it could mean an inadequate number of products in IKEA stores, leading to longer waiting periods for customers or an overstock.

An isle inside an IKEA warehouse with long rows of shelving and a forklift.


Portrait of a man with a white shirt and a beige sweater.


Peter Grimvall, Supply Chain Development Area Manager.

“Without a proper and accurate forecast, we can’t understand the demand… what needs to be supplied, to who, and when. If it’s too much, it increases our costs and hence prices to our customers, or too little, which means we won’t be able to provide the right offering to our customers. This can have a big impact on our business and the way we serve our customers”, says Peter.

Demand Sensing – the new tool

The forecasting system that IKEA has used so far would make predictions based on statistical sales. These would include, for example, sales and demand patterns from last years.

On the other hand, Demand Sensing can make use of up to 200 data sources for each product to calculate forecasts and predict the future demand more smartly and effectively. The tool can use several influencing factors, such as shopping preferences during festivals, the influence of seasonal changes on purchase patterns, and weather forecast, among others.

“The Demand Sensing tool is a unique way of applying artificial intelligence for supply chain planning. This will be the largest deployment of AI in IKEA’s supply chain”, says Peter.

An IKEA warehouse with shelving numbered with large signs, plus an orange forklift.

It can even understand the increase in-store visits during a specific period of a month, such as when people get their salaries, and buying patterns during festive periods and holidays.

“In the past, we’ve had around 92 per cent of the forecast being accepted, and 8 per cent was corrected. And now, with the Demand Sensing tool, we are at a level of close to 98 per cent accepted forecast, with only 2 per cent corrected,” says Peter.

A smarter, omnichannel tool

With the existing forecasting tool, demand predictions would start at a global level and then be broken down to regional, country and store levels.

Demand Sensing, however, builds up knowledge from a local perspective, with the local customer at the centre of its forecast. Once the local store forecast is made, it can then go up to market, country, region, and global levels.

Pallets with large packages being transported along two long conveyor belts.

Let’s understand this with an example. In the IKEA store in Furuset (Oslo, Norway), let’s say there is a trend that one product is selling faster than expected. With the new tool, the forecast for that product will go up in Furuset, Oslo, but not in the rest of Norway. With the system that IKEA is currently using, the forecast would be raised for the whole country.

What this means is that Demand Sensing works smartly and across online and offline channels by capturing demand. So, if the sales increase in the online channel, the forecast will capture it right away.

“The new tool can forecast demand from a day-to-day basis to up to 4 months. With us trying out new store formats and new selling capabilities, and fast-changing customer behaviours, the tool is agile enough to capture the unique needs”, says Peter.

Smart forecasting – lower cost

An accurate forecast means that IKEA stores will get relevant articles at the appropriate time. This can ensure better availability of products for customers both within the stores and online, helping the brand delight its customers at all touchpoints.

The reduced need for manual overrides and fewer errors will also help to save money in the company’s supply chain and optimize its logistics better.

Pallets with packages in a large warehouse; some stand in yellow boxes painted on the floor.

What are the benefits of accurate forecasting?

“What is the cost for poor planning? When we stop selling an article, we must lower the price to sell it out. This is a significant expense today. From the business point of view, if we can improve the accuracy, we will have less inventory of things we don’t need. This saves logistics costs that can be passed on to the customer. We will have fewer complaints from our suppliers that they have bought too much raw material for a specific product going out of the assortment.

From a carbon dioxide emission perspective, the more efficient we can be in our supply chain, the more positive impact we will see. Because the last thing that we need is to ship products into places where customers may not need them.”

What’s next for IKEA in terms of AI and innovation for supply chain management?

“We can do much more in terms of different prediction models so that we can forecast various supply chain events – how long should the lead time be, how much a new product will sell for example. Hence, predicting planning parameters with the use of AI will be one big area.

The other area is related to supply chain planning, which has two parts – planning and problem-solving. We always have some problems to solve, but we are not capturing the solutions today. So, what we’re trying to do right now is to develop a corporate memory that can register the problems we faced and how we solved them. With this corporate memory, we will bring in automation. We call it as digitalized decision making. So, that’s a big area that we are exploring right now.”