improvement on forecast runtime
accuracy on a 52-week forecast
Action has grown in just 30 years to become Europe’s fastest growing non-food discount retailer. But with substantial growth comes a substantial inflow of data.
Each of the company’s stores across 11 countries needs to track everything, from consumption patterns to product placement to supply chain disruptions — all of which vary according to local, national, and international trends. It’s no wonder, then, that Action has recognized the importance of developing more robust data analytics operations to further their success.
We spoke with Randy de Heus (Manager of Business Insights and Data Analytics) and Dawid Kirsten (BI Consultant for Insights and Analytics) to learn more about how Action is building the right architecture and toolset to leverage the company’s data with greater efficiency. More specifically, we’ll look at how Capgemini and Dataiku worked together to help Action develop more accurate forecasting models with greater speed.
Prior to working with Dataiku, Action was relying on Excel-built algorithms for their forecasting models. The process of deploying the sales forecasting model, which is run every week and realigned once a quarter, was over-complicated as a result. “[Data scientists] even needed a couple of laptops just to run the application,” de Heus explained.
The process was also, by consequence, relatively slow. It took data experts several days to run the forecast each time, and hours more to realign it several times a year.
Kirsten and his team worked with Dataiku and Capgemini to mature a proof-of-concept, black box of code into a more easily explainable and understandable visual diagram of the data journey. That modest proof of concept became, in a short period of time, the team’s sales forecasting model.
“When we initially did it, it was based on some R code on our server.” Then, with Dataiku, “We went through the process of linking it to live data, creating pipelines to move the data to the data lake, and also transferring that large block of code into visual recipe steps where possible, or else splitting the R code into buckets to be processed separately.”
With Dataiku, which allows the team to test, modify, and redeploy their models routinely and with precision, they brought the runtime of each forecast down to between nine and 14 hours, and then further improved this to between six and nine hours — a time savings of almost 900%.
It’s no small thing for your business team to have a reliable sales forecast. As Kirsten stressed, these are not easy to generate without the right tools. “We have gone from an Excel-based guesstimate of what the forecast should be to 88% accuracy on a 52-week forecast [with Dataiku],” he said.
Though it is primarily the data scientists who make use of the Dataiku platform directly, there are many functions within the supply chain analytics team and other business units that benefit from access to the platform and its outputs, which flow directly into operational tasks.
Dataiku has allowed data scientists to build trust in the system, clearly showing line-of-business users how the forecast works thanks to the platform’s visual recipes. The ability to manipulate data with visual recipes and split and sort analytics tasks has also proved crucial for the team as the analytics operations have grown.
If you use a visual recipe, it’s much easier to explain to the business side what’s happening to the data. ‘This is what’s being filtered. This is how it looks.’ You can take them through every step of the path, giving them confidence in the results.Dawid Kirsten BI Consultant for Insights & Analytics at Action
As the team’s ability to build and deploy models increased, so too did the need for a reliable means of monitoring the processes from start to finish. “Of course, the first step is to create the algorithm,” de Heus said, “but the next step is to make sure that you can run it every day, multiple times a day.” An operation of that caliber has many stakeholders, from the data scientists building the models to the business analysts deriving insights from them. At every point of contact, the data analytics team needed to have visibility on how the model is being built, changed, realigned, and used.
With Dataiku, Action has begun to see the benefits of a robust AI Governance system. The platform allows user-owners to view and control access to the team’s models and processes, from inception to deployment to feedback-aligned redeployment.
I think that’s the added value of Dataiku: it’s not only the ability to build a solution or come up with an algorithm, but also the ability to deploy, maintain, and improve it.Randy de Heus Manager of Business Insights & Data Analytics at Action
Throughout our conversation, de Heus and Kirsten stressed that, despite the successes their team have already enjoyed, they are only just getting started. Looking ahead, they would like to continue building use cases that further prove the business value of data and analytics operations to the company’s line-of-business teams.
Their first three use cases (sales forecasting plus DC outbound forecasting and route optimization for the company’s trucks) are already integral to business decisions at the company.
They also help to promote the self-service data discovery that many business analysts have begun doing thanks to Dataiku. “We get involved when there are operational requirements, or if there is a specific request,” Kirsten said, but otherwise the team hopes that self-discovery can take off independently while the data scientists focus on playing with the data to find potential use cases, and on seeing whether they might integrate one of Dataiku’s out-of-the-box business solutions, like Market Basket Analysis, to support core business operations.
You want to give your business side the flexibility to start deploying a use case, something which is much easier to do with a tool like Dataiku.Randy de Heus Manager of Business Insights & Data Analytics at Action
As they plan for the months and years ahead, de Heus and Kirsten aim to select the use cases to develop according to the principles that guide the growth of their team. “In the future, data science can be the difference between being successful and not,” de Heus said. “That, at the end, is what you’re trying to prove to the company. We’re currently doing well. But we always need to focus on how we can improve.”
Working with their colleagues at Action to adopt this data-driven mindset has been the bright thread for de Heus and Kirsten. As Action continues to grow and move into new markets, they hope to be a leading force behind the business’s strategic decisions. “It’s not easy to focus on the things that give you that extra 1% or 2%,” de Heus said, “but that’s what data can give you”: seemingly small differences with immense impacts in the long run.
Though the savings generated by the express shipping recommendation model will only fully materialize over time, the tool when globally implemented is estimated to reduce express shipment costs by 11-36%.
Read moreEveryone's talking about MLOps, but what value does it provide in practice? And how can Dataiku help? We sat down with Subhadip Roy, Head of Machine Learning Engineering, AI, and Data at Deloitte to hear about his experience in the field.
Learn MoreOrange has accelerated their data science practice with Dataiku, building a call load detection and triage model in just one month, with more use cases on the horizon.
Learn MoreLG Chem noticed that their employees were spending a lot of time searching for safety regulations and guidelines so, with the help of Generative AI and Dataiku, they provided an AI service that helps them find that information quickly and accurately.
Learn MoreHear how Fanalists — a data and marketing agency working for the sports, media, and entertainment industries — has completely transformed their business with Dataiku.
Learn More
