Data and AI
Paula’s Choice Skincare is a leading, direct-to-consumer (DTC) global, mission-based brand rooted in truth and transparency. Founded in Seattle, Washington, in 1995 by American radio talk show host and “cosmetics cop” Paula Begoun, the pioneering brand is famous for its “industry-leading innovation, accessible jargon-free science, high-performing ingredients, and cruelty-free products.” Paula’s Choice offers powerful content and digital tools to demystify the science behind skin care, including an extensive “Ingredient Dictionary” that breaks down the research behind nearly 4,000 ingredients and Expert Advice, a curated online hub of skin care and ingredient knowledge. The brand is distributed through global DTC and select prestige retailers in North America, Europe, and Asia. It was acquired by the multinational consumer goods company Unilever in 2021 and continues to operate independently. Its main headquarters are in the United States, and its European offices are in the Netherlands.
Defining the Future
As a leading, direct-to-consumer global brand, Paula’s Choice is and has always been focused on eCommerce. But that didn’t necessarily mean it was data-savvy as a company. “Although digital is in our hearts and is our focus, we didn't have anyone taking care of the data — no CRM, no data warehouse, no one taking accountability,” explained Roeland Euser, the head of marketing and data in the Netherlands. “So, I raised my hand and said, I’ll take care of the data for now.” But once he took that responsibility, he realized no one in the organization knew where he should start, so Euser sought help outside Paula’s Choice.
The entire company understood that it needed to become data-driven, but no one knew where to start. After some initial research online, Euser made contact with Xebia. “We weren’t quite ready to start working with Xebia right away, but then all the stars and planets aligned and we engaged them for some strategic coaching. It was like getting a customized guide for our journey,” Euser said.
Since Paula’s Choice had recently been acquired by Unilever, Euser and his team were tasked with presenting the independently operating skincare company’s data strategy. Fortunately, the strategic coaching was already well underway, and Xebia's consultant Steven Nooijen gave Euser’s executive team a crash course on what it needed to know: What’s the cloud? What's a data platform? What's a customer data platform? “Steven covered the entire topic and helped us get all the lingo right. He even told us what we should focus on in the first year,” recalled Euser.
Through strategic coaching, Xebia helped Paula’s Choice define a concrete and pragmatic strategy that could be executed immediately.
To begin with, it was all about marketing — how could the company use its customer data to serve them better? Nooijen and Euser defined the first use case as creating a product recommender. “Initially, we wanted to have a machine learning model that integrated session data with product attributes to predict the specific products a particular user wanted,” explained Euser.
Through Xebia's strategic coaching, they decided to start with a more straightforward benchmark instead — a rule-based model, which was easier to create and interpret. They could turn their domain knowledge into a model, put it into production, industrialize it, see the results and improve from there.
“We were already working in an Agile way, but Xebia helped us with fine-tuning,” said Euser. “Steven sat in those meetings giving pointers and suggestions on how to improve and what we should prioritize. He coached me in those sessions. Make sure to address this, this, this, this, this. So he essentially prepped me on how to be a good data manager," he added.
The strategic coaching helped align roles and responsibilities across all stakeholders to make analytics work, including IT and HQ in the USA, plus the web, data, and marketing teams.
The one-year goal to achieve by the end of 2022 was to have the first two to three data products live and in a continuous improvement cycle. They also set a three-year “audacious goal” goal: to become the leading data-informed skincare company.
Once the goals and strategy were determined and the preconditions for successful development were prepared, Vadim Nelidov, a data scientist from Xebia, joined their team and began building the data science algorithm defined by Nooijen and Euser. Simultaneously he began training a junior data scientist at Paula's Choice and introduced new ways of working with data products. Testing, code quality, documentation, scalability, version control, and interpretability became new core principles of the team. This way of working empowered the team to be able to deliver future-proof solutions on their own.
At the same time, the first results from the newly implemented rule-based model were encouraging as well — predictions matched actual historic consumer preferences in at least 70% of the cases.
Aligning Goals through Teamwork and Coaching
From the beginning, it was clear Paula’s Choice would focus on bringing its data architecture up to date as efficiently as possible, and the fact that the company was already working in an Agile way proved helpful in scaling.
To further ensure the company and its employees were all on the same page, strategic coaching and training started immediately, with a focus on the data, web, marketing and IT team along with US headquarters to align roles and responsibilities. Making analytics functional and working across all teams involved training leadership and data teams in data-driven methods. Teams had weekly homework and worked together to create a solid base.
An Azure Databricks stack was used for its data platform. The stack combines data and machine learning solutions with Paula’s Choice's own recommendation models. Further, it allowed easy integration with Bitbucket, used for code repositories and version control. This resulted in being able to automate reports more often and save plenty of previous manual work while only raising the quality bar.
The whole process took much of the guesswork out while giving more freedom instead of waiting on feedback and directions. The data team’s operational backlog was refined on a weekly basis. The team could take the driver's seat and have more autonomy to accomplish a wide range of goals to make Paula’s Choice the skincare line of the future.
Trailblazing for the Entire Company
Working with Xebia created a paradigm shift for Euser and his team.
“We're now leading the global company in leveraging our data platform,” he remarked, "Before, we were a bit more cautious, waiting for directions on what we were going to do and hoping it would be something that we could use. Now, it changed into us being in the driver's seat."
The one-year goal was to have three use cases up on a continuous improvement cycle, and Euser expects to reach that goal. The company has a fully functioning data lake with Databricks. They're using the machine learning functions of Databricks as well, and they've built an API between Databricks and Salesforce to industrialize their machine learning models and product recommendation models. "This is something that no one else, not even the global office, is doing yet," smiled Euser.
With the help of one Xebia data scientist, Paula’s Choice was able to develop a new product recommender that showed a 8.5% increase in click-through rate and provided 2,000 unique product combination recommendations during the test period, compared to 300 unique combinations from the previous standard recommender that they had before.
"We’re setting the tone for how to actually use the platform we were told we had to use. And we’re like, okay, great! We're gonna exploit it because of the work we're doing with Xebia."