Ìý
is the world’s largest online classifieds group, spanning 10 countries with over 25 marketplaces including Leboncoin, Kleinanzeigen, Milanuncios, Subito and Marktplaats. With a reach extending to over 1 billion users worldwide and annual revenue exceeding €1.5 billion, Adevinta seamlessly connects buyers and sellers, enabling transactions across various sectors from job postings to real estate, automobiles, consumer goods and beyond.
Ìý
Elevating business insights
Ìý
Adevinta's analytics team developed a groundbreaking Customer Lifetime Value (CLV) Machine Learning (ML) model. This innovative tool analyzes specific data subsets to forecast CLV, a vital metric indicating a customer's total value to a business throughout their relationship. Understanding CLV empowers businesses to optimize marketing, retention and acquisition strategies, ensuring resource allocation efficiency. Adevinta's investment in this ML model reflects its commitment to precise long-term profitability assessment amidst the inherent uncertainty of customer behavior predictions.
Ìý
The issue
Ìý
Previously, retraining Adevinta's CLV Machine Learning (ML) model with varying data, filters and parameters was a manual process handled by a specific member of the analytics team, leading to dependencies and delays.
Ìý
In response, Adevinta initiated an MVP project aiming to empower non-technical team members, specifically the Global Marketing team, to autonomously conduct these tasks. The objective? To develop a user-friendly frontend interface enabling users to select data, apply filters, train models and explore results effortlessly with tailored metrics and plots.
Ìý
ºÚÁÏÃÅ' transformative solution for Adevinta's CLV model
Ìý
In a concise three-month collaboration, ºÚÁÏÃÅ spearheaded the development and delivery of the MVP for Adevinta. Crafting an end-to-end solution, we addressed critical needs with precision and innovation.
User-friendly interface
Our solution boasted a straightforward UI interface, catering to both technical and non-technical users. Empowering users to effortlessly select filters and execute the model, the interface also provided real-time job status updates. Upon completion, users could visualize results across various dimensions and download queries and outcomes.
Model productionization
Ìý
Transitioning from a notebook setup, we streamlined the model by deploying it onto a Kubeflow pipeline, scalable to handle vast datasets—a crucial requirement for Adevinta. This integration with the warehouse facilitated seamless data access and processing.
Ìý
Robust infrastructure
Ìý
Ensuring security and reliability, we hosted the UI application within Adevinta's AWS environment, meticulously constructing the infrastructure within its VPN for enhanced protection.
Our MVP garnered enthusiastic feedback from stakeholders, showcasing its seamless end-to-end functionality within the stipulated time frame, even amidst uncertainties.
Ìý
Overcoming difficulties in Adevinta's CLV model implementation
Ìý
Embarking on uncharted territory, we encountered several difficulties during the implementation phase, exacerbated by the novel introduction of the cluster within the client's infrastructure.
Remote model triggering
Ìý
We tackled the challenge of remotely triggering models with custom parameters from outside the cluster. This necessitated extensive experimentation and trial runs to identify viable solutions.
Integration complexities
Ìý
Navigating through processes and uncertainties, we grappled with integrating the system with diverse platforms such as the data warehouse and marketing data lake, each presenting its unique set of challenges.
Streamlined handoverÌý
Ìý
With a deadline looming, we opted for a simplified tech stack to ensure a smooth handover at the end of the three-month engagement. Emphasizing clear setup documentation and strategic decision-making, we aimed for seamless transition and continuity beyond our tenure.
Ìý
Empowering marketing excellence
Ìý
Our solution empowers Adevinta's marketing teams—both global and from diverse marketplaces—to conduct CLV experiments seamlessly, integrating insights into their campaigns for maximum impact.
The UI allows analysts and marketers with low-tech skills to tune their CLV calculation to the specific cohort and make it available in the marketing platform to be used along with other useful information. Also, it will allow in the future to easily extend the UI to do the same for other interested models.
By harnessing Lifetime Value (LTV) as a central metric, our initiatives have helped to revolutionize Adevinta’s marketing strategy, steering away from short-term metrics towards sustainable growth. This shift aligns attribution models, real-time optimizations, audience segmentation, and ad serving, enhancing customer acquisition, retention, and re-engagement.
Ìý
Investments in data analytics, infrastructure, experimentation, and loyalty programs underscore the commitment to driving sustainable growth and maximizing customer lifetime value. These endeavors optimize resource allocation, foster innovation, and nurture enduring relationships with high-value users.
Ìý
The collective impact of these initiatives is estimated at €2,750,000, encompassing software development, infrastructure upgrades, personnel training, data acquisition, and ongoing maintenance. Each initiative contributes to the overall goal of optimizing marketing effectiveness, enriching customer engagement, and unlocking long-term value for Adevinta.
Ìý
We now have a solution that can help to prove ROI for our marketing spend, predict churn for our users (and target them with dedicated actions), optimize campaigns based on CLV and target similar high-value users with third parties. CLV opens a new world to marketing at Adevinta.
