Evolving from a place where collectors catalogue their items to a modern auction house for unusual, rare and exceptional objects. Catawiki connects buyers with sellers, delivering marketing services to help sellers get the best offer for their goods.
Data Science at Catawiki
With Domino Data Lab, Catawiki is arming business users with insights that help them make smarter, faster decisions and embedding machine learning to optimize its recommendation engine, product pricing predictions, marketing promotions, lead scoring and cancellation predictions.
During its pilot with the data science platform, “Domino demonstrated its ability to fundamentally shift our productivity,” said Peter Tegelaar, chief data scientist at Catawiki. “As we roll it out, it is allowing us to tackle a whole new class of problems, and it’s making what we were doing before much faster and more efficient, living in one place.”
Catawiki’s growth trajectory has historically been dependent on the size and capacity of its auctioneer team. The group of 200 individuals—possessing a skillset almost as rare as the items they auction—spent considerable time screening, curating and evaluating prices for new lots. They lacked the bandwidth to personalize the customer experience.
Likewise, Catawiki risked losing prospective customers because its Marketing team and recommendation engine struggled to precisely target the right audiences with the right products at the right time.
Catawiki had rich data that could be leveraged in predictive models to help auctioneers scale, improve customer engagement and increase sales, but data scientists were hamstrung by two key infrastructure challenges.
Domino demonstrated its ability to fundamentally shift our productivity... making what we were doing before much faster and more efficient, living in one place.
As a result, projects were delayed and data science team lost credibility with the business.
Limited compute resources constrained model development. The engineering team misunderstood the data science team’s uneven hardware usage and reduced their computing power to run experiments because average utilization was low. This constrained the team’s ability to deliver business value.
Differences between the engineering and data science teams’ processes slowed deployment of models into production. The data science team worked in languages like Python and R, while Catawiki’s production environment was in Ruby. Getting results into production either meant dumbing results down or delays as they re-translated the models. Instead, data scientists resorted to making batch predictions offline, preventing Catawiki from approaching real-time use cases where the customer experience could be changed mid-flow.
“This is exactly why data science fails in so many companies,” said Tegelaar. “You have a data scientist, but unless they also happen to be a stellar software engineer who’s good at building bridges, like with DevOps, it is very difficult to be effective.”
On its mission to empower auctioneers, marketing users and other lines of business with machine learning, Catawiki is implementing a Domino data science platform. Domino makes it easy for data scientists to:
Provision compute resources as needed during the model development process.
Collaborate with business stakeholders as they test, validate and tweak models.
Deploy models into production without Engineering support.
The main selling point of Domino was its ability to empower data scientists to experiment and deploy models independently. Catawiki also appreciates its collaborative features and reproducibility, protecting business continuity and mitigating against key man risk.
The auction house values the partnership it has built with the Domino team, both regarding their rapid resolution of technical challenges and communications around enhancements to the product.
The Domino Effect
By implementing the Domino data science platform, employees across the organization can work faster and make a bigger impact. For instance:
Auctioneers that use Domino do less manual research to calculate pricing, delivering time savings of 30% while improving the accuracy of their predictions.
The team can more quickly:
- Determine whether a lot should be approved or not based on its predicted value
- Provide faster response to sellers
- Get approved lots posted faster than before
This ultimately translates into more auctions taking place on the Catawiki platform, and greater revenues for the business.
Marketing directly impacts revenue by:
- Predicting customer lifetime value
- Matching supply and demand at a granular level
- Making smarter decisions about which products to promote
For example, the value of the commissions Catawiki earns is dependent on the highest bid of each product sold; Catawiki does not earn commissions for products that don’t get bids or whose reserve price is not hit. With the pricing predictions built in Domino, Catawiki can make much smarter decisions around which items it should promote, and to whom, in order to maximize bid values which will reap the highest commissions.
“We didn’t even take that on before Domino,” said Tegelaar. “The nice thing about machine learning is that you can be much more granular, rather than applying a sweeping rule that would have us, for example, spend marketing dollars promoting items that will hit their reserve price no matter what.”
Engineers and DevOps spend less time supporting the dynamic needs of data scientists, and more time driving strategic investments in their customer-facing platform and internal CRM.
Data scientists are self-sufficient and empowered to do more. They spin up workstations that meet individual project needs and publish API endpoints to push models from prototype to production with response times of less tha 200 milliseconds.
“In a company that’s growing as fast as Catawiki, speed is the name of the game,” summarized Tegelaar. “By helping us move very fast, Domino is making it much easier for data science to show its value.”