Royal FloraHolland

A Global Trade Platform for Floriculture, Driven by Smart Data Applications

Customer challenge: Transform the traditional flower and plants auction into a digital auction, Floriday.

Provided solution: Image recognition algorithm to check auction goods photo quality and recommend improvements. Recommender systems to connect growers and buyers.

Outcome: Created several successful AI applications that were adopted and used by the business. Improved capabilities of internal data science team and advanced data science practice.

For over one hundred years, Royal FloraHolland, its growers, and customers have been making the world a healthier and more beautiful place with flowers and plants. Every day, streamlined logistics and smart digital services ensure that 400,000 species of flowers and plants arrive at their worldwide destinations.

In 2017, Royal FloraHolland launched the Digital Greenhouse, a global trade platform for floriculture. In addition to digitizing their existing services, it also worked with GoDataDriven to develop new and innovative solutions to help growers and buyers find each other more easily. These solutions include deep learning applications, up-to-date reporting, predictions and smart recommenders.

Remco Wilting, head of data and data science at Royal FloraHolland, explained, “GoDataDriven helped us move from analysis and models to data products worthy of production. Now we can really derive value from our data initiatives.”

The Digital Greenhouse

Royal FloraHolland offers a complete range of services to ensure that the floriculture trade runs efficiently, including a world-famous auction. They recently launched the Digital Greenhouse program to facilitate future growers and stimulate global floricultural trade further. Within this program, they established a data and data science department with an agile team.

“The data science team develops smart models and data-driven solutions. They digitize and improve existing processes and also develop new, innovative products,” explained Wilting.

“Royal FloraHolland has an impressive history,” said Wilting. “We optimized the process from grower to buyer on all fronts. Now it’s time to translate this unique expertise into digital applications. Since data applications play such an important role, we decided to collaborate with GoDataDriven, the best specialists in the business,” he explained.

Global Trade Platform

Floriday is one example of a digital application. It’s a trade platform that brings together global supply and demand for floriculture. It is an open platform that allows Royal FloraHolland, as well as third parties, to add services and channels. “Royal FloraHolland’s priority is the growers,” said Wilting. “The Floriday platform’s biggest strength is that it provides buyers of floricultural products access to the global market.”

“Discussions with colleagues, growers, and buyers often lead to new projects. We learn that we can improve existing processes by making them more accurate or less time-consuming,” he explained. “These discussions often lead to new applications as well.”

Another source of projects comes from the data science team since they have an alternative perspective on the business. Image recognition for product photos is one example. Working with GoDataDriven, they apply their extensive experience in developing applications for other organizations to come up with new ideas for Royal FloraHolland.

Developing Data-Driven Applications

The Digital Greenhouse has a number of core projects in development. In addition to a big data platform, these include deep learning applications to improve photo quality; up-to-date reporting to provide insight into optimal and stable pricing; and recommenders on the Floriday platform.

Photo Quality Check

Royal FloraHolland uses product photos in its Floriday catalog, and increasingly for its floricultural auction as well. For this reason, high-quality product photos have become increasingly important in the marketplace. To automatically determine a photo’s quality and recommend improvements to growers, GoDataDriven has developed a deep learning application that uses different characteristics to assess photos in real time.

The characteristics that make a photo high-quality differ per product category. Sometimes, a white background is essential. In other cases, there has to be a size ruler in the shot. The image recognition software identifies these different elements and recommends potential improvements to the grower.

The application ensures a smooth supply of high-quality photos that can be used immediately for both the auction and online product catalog.

“Our goal is to make these types of applications from the Floriday eco-system available as APIs in the future. Third parties can then develop their own applications using the technology. It’s like a virtual photo booth for creating high-quality images of floricultural products,” explained Wilting.

GoDataDriven used TensorFlow, Google’s deep-learning algorithm, to develop the application, and the marketplace’s photo library to provide the model’s remaining “training.”

Optimal and Stable Pricing

It is important for Royal FloraHolland to come up with fair and stable prices for floricultural products. Fair for the growers and customers, and stable to ensure a realistic price expectation. The optimal and stable price reporting is a great example of a data project that is aimed at improving a previously existing process.

“After meeting internally with different stakeholders, it turned out that there were some potential improvements for reporting. The data science team drastically increased the number of data points and redeveloped the statistical algorithm,” said Wilting. “The new application now automatically gives insight into the latest price level per day and the developments across all product groups.”

There is detailed information for each product group and all products that fall under them. In addition to seasonal influences and the optimal price level, the application also provides insight into price stability. The application looks at the variance over the past twenty days as well as the trend. The optimal price is determined using the deviation from the trend line based on a benchmark.

The Commerce Department uses the insights to improve its services. For instance, they inform the market about price developments or identify changing market circumstances at an earlier stage so that specialists can find out the cause. In the case of price instability, the commerce department tries to determine the cause using their market expertise. For instance, it could be that a supply line is temporarily unavailable or that a product’s quality is inconsistent.

Smart Recommendations

Floriday uses modern technology to help buyers and growers find each other as easily as possible. For example, the platform offers a smart search based on Elastic. Royal FloraHolland is currently also developing an algorithm on a Sparkcluster in AWS that comes up with recommendations for other interesting products, based on the user’s search behavior.

Cloud

The right technology is the foundation for quick results.

“We use existing functionality and building blocks whenever possible,” said Wilting. “It would be a waste of time and effort to reinvent the wheel. “

Royal FloraHolland makes extensive use of cloud services. In addition to Google’s TensorFlow and the Spark cluster on AWS already mentioned, the self-service BI platform and the big data platform also run in the cloud – on Azure and AWS respectively.

“We expect that in the future, all IT within the organization will be in the cloud,” added Wilting. “Not just the models we take into production, but legacy systems as well.”

Amazon Customer Story

Amazon Web Services (AWS) recorded a customer story with Royal FloraHolland and GoDataDriven: Use Case Xebia Uses Machine Learning on AWS to Evolve Royal FloraHolland’s Practices

Industry

Data and AI Strategy

Project type

Custom Predictive Modelling
Setting-Up Cloud Infrastructure

Technologies used

Python
Spark
Elastic
Amazon Web Services

“GoDataDriven helped us move from analysis and models to data products worthy of production. Now we can really derive value from our machine learning initiatives”

Remco Wilting Head of Data and Science
Explore more cases