Xebia Helps Royal FloraHolland Use AI for Smarter Predictions

Improving Operations and Transforming the Customer Experience

A Century-old Company Blooms into the Digital Space

Embraced around the world for their ability to speak volumes without saying a word, flowers can be found for sale nearly everywhere. The flower industry is expansive, and with its 145,000 transactions conducted per day and 400,000 varieties of flowers and plants available, one company stands apart in its legacy, volume, and sheer size: Netherlands-based Royal FloraHolland, a cooperative company where growers come together to sell their products in the same place.

Royal FloraHolland connects buyers and sellers around the world through its online trading platform Floriday and its world-famous live floral auctions. The enormous Aalsmeer auction facility, which sprawls over 128 acres near Amsterdam, serves as tourist destination where visitors can immerse themselves in the international flower trade.

“We’re in the business of bringing together as much supply and demand as we can at one time,” says Remco Wilting, head of data and data science at Royal FloraHolland. “We’re a large company and over 100 years old, so we’re very traditional in some respects, but we’ve recognized the importance of going digital to provide our growers and buyers with more opportunities.” As Royal FloraHolland recognized that growers wanted to use other trading methodologies outside of the physical auction house to sell flowers, the company began exploring how it could improve current processes and provide growers and sellers with new opportunities to reach buyers.

“The world is getting smaller. Digitization is necessary if we want to stay competitive and differentiated in our industry,” says Wilting. “Because if we don’t do it, somebody else will.”

The Digital Greenhouse: Building Data-driven Applications on AWS

The company recognized its need to reorganize in order to go digital and become more data-driven. “After the decision to go down a more digital-focused path, the IT department evolved into the business technology organization, or BTO, which is responsible for both IT and digitalization of the business,” says Wilting. “During that evolution, we started what we call our digital greenhouse, where we’re building new applications for our growers and our customers.”

Another step in Royal FloraHolland’s reorganization process was identifying partners to take care of IT needs and infrastructure. “Because of its global reach, pace of technological innovation, and flexibility, we chose to migrate to Amazon Web Services and use it to deploy all future applications built in our digital greenhouse,” says Wilting.

Wilting built a team to support the company’s digital transformation. “As we began to build Floriday, we digitized our global trading platform for buyers and sellers and realized just how big of a role data was going to play in the future of Royal FloraHolland,” says Wilting. He brought in Xebia, a partner with data science and AWS expertise, to help him educate his growing data science team and identify use cases in which well-trained deep learning models could drive better business decisions.

“If we can build and train models that help us drive better predictions, then we can quickly translate the improvements that follow into money saved."

Remco Wilting
Head of Data and Data Science
Royal FloraHolland

“If we can build and train models that help us drive better predictions, then we can quickly translate the improvements that follow into money saved."

Remco Wilting
Head of Data and Data Science
Royal FloraHolland

Saving Costs, Driving Efficiencies by Improving Trolley Predictions

Thousands of trolleys move swiftly through the Aalsmeer facility each day in a choreographed process that brings flowers in from trucks, sorts them, and then delivers them to outgoing delivery vehicles. Each trolly is an intricate and vital part of the Royal FloraHolland supply chain.

“Accurate trolley prediction is a very old problem for us,” says Wilting. “The number of trolleys that come in determine how many people are needed to handle all of the flowers and plants, so the workload of our logistics team is very much dependent on how many trolleys come in each day. It’s important information for them to be able to plan in both the short- and long-term, and it’s not a prediction we’ve been able to improve substantially in the past.”

Royal FloraHolland and Xebia saw in the trolley prediction problem a classic use case for data science to address. “Regardless of if you have too many or too few people walking around because of inaccurate trolley predictions, neither is good for our business. Either we spend too much money on labor or we have too much of a workload for the people there, meaning we won’t meet the deadlines of delivering goods to customers,” says Wilting. “If we can build and train models that help us drive better predictions, then we can quickly translate the improvements that follow into money saved.”

Showing Buyers a Complete—and Accurate—Picture

Flower buyers want to know what they’re buying but often see only a picture before purchase. “When a picture isn’t representative of the product, the buyer is very unhappy. The pictures are traditionally inconsistent in quality. They’re taken by the individual growers and can vary greatly from grower to grower,” says Wilting. “We're developing a deep learning model to check image quality and provide feedback to help growers improve the quality of the images that are presented at the auction.”

The business value is clear: The better an image, the more likely a grower can sell flowers at a higher price. “The value is related to the overall revenue going through our platform,” says Wilting. The company continues to train its models and gather feedback from internal teams to drive further improvements to the application.

Showing Current Supply and Recommending Alternatives in Real Time

Providing buyers with attractive alternatives when their first choices are not available can drive additional business revenue and improve customer satisfaction. Using deep learning on AWS, Royal FloraHolland is currently exploring how it can automatically present buyers with stock availability and alternative options tailored to their preferences.

“We want to make it possible for buyers who aren’t close to our physical locations to see the current supply and what they can buy now,” says Guijt. “We believe we can use this data to build and train models for a recommendation engine that’ll assist them in finding what they need and providing alternative options to them. For instance, if they always buy from a particular supplier a flower that’s not available anymore, we want to help them intelligently identify alternatives they may be interested in.”

Improving Operational Efficiency

Becoming data-driven affects the entire company. Royal FloraHolland’s data science team learned to involve internal stakeholders early on in projects.

“We’ve had a lot of eagerness to help out and provide feedback to us as we build these applications,” says Guijt. “Through our demonstrations and explanations of the business goals driving these technological changes, we find that internal stakeholders begin to see how much easier their lives can become. They see the benefit and they want more. And by providing us with feedback as we develop and train the models, they hold a stake in what we’re doing help shape the models.”

For Royal FloraHolland, the goal of using data to improve business outcomes and keep the company highly competitive in the flower industry is being realized through each project it undertakes on AWS with the help of Xebia.

“It’s empowering to know that we can transform the business through data-driven initiatives,” says Wilting. “And we’re only just beginning. There will be much more to come.”

Crayon helps TINE revolutionize the Norwegian Dairy Industry using ML

Learn more »

SFL Scientific helps LinkSquares use ML to eliminate manual contract reviews

Learn more »

Anodot helps Lyft identify business risks with ML

Learn more »

Convoy uses ML to revolutionize trucking

Learn more »