Consumer Data Products (Codapro) at Delivery Hero is a Tech org of about 120 engineers and scientists. Our mission is to optimize the entire customer journey. One of our important products for achieving this mission is the Customer Data Platform (CDP).
What does the CDP do?
The CDP computes customer attributes based on near real-time global streaming and global batch data. These customer attributes range from simple behavioural counts and aggregates to outputs of ML models. The attributes are served through low-latency APIs so that our brands in over 70 countries can use them for in-app experiences. The majority of the CDP use cases are in personalization, fraud detection, advertisement optimization, and search and recommendation. However, increasingly stakeholders approach us with unexpected yet impactful use cases.
What is the impact?
In the food delivery industry, customer experience is vital. It’s central to everything we do at Delivery Hero. We believe in a simple equation: the more personalized and smart the app experience, the better our long-term customer relationship and our competitive position in the market.
Our innovation strategy reflects this belief. It’s all about enhancing the customer experience. Our Customer Data Platform (CDP) is a critical tool in this strategy.
The CDP’s strengths lie in both speed and comprehensiveness. It offers fast access to fresh customer data via APIs, accelerating feature development and expediting implementation, resulting in improved outcomes. Moreover, its user-friendly self-service approach empowers teams across Delivery Hero to contribute their own code and insights to the CDP. This decentralized approach fosters collaboration and innovation, resulting in a comprehensive set of user attributes that sets us apart within the company.
What are the elements of the CDP?
Identity
Underneath the CDP sits the Identity Resolution System (IRS). The IRS identifies which accounts belong to each customer, allowing for account deduplication and fraud detection. Notably, the IRS manages these tasks with impressively low latency, especially when considering the vast number of customer accounts.
Utilizing the IRS, the CDP calculates customer attributes and makes predictions at both individual and aggregated customer levels. For example, if we use machine learning models to predict Customer Lifetime Value (CLV), using IRS will allow us to factor in all relevant data from known customer accounts (and not just the one the customer is currently logged into).
Computation
CDP computes data in stream and batch. We chose Apache Flink as our stream processor. We implement Flink jobs using Java, Python, or SQL, tailored to specific needs. What’s more, we open the stream job platform to other teams to contribute their own Flink job code, allowing a self-service mode.
For batch processing, we harness the power of BigQuery, Airflow, and dbt. The standout feature of dbt is its self-service nature, allowing teams to independently manage their data transformations.
Audiences
The CDP also features an Audience Engine that builds audience definitions from the customer attributes in the CDP. This function enhances personalization efforts and supports retention marketing strategies.
Openness
Designed for extensibility, the CDP allows Delivery Hero brands to compute customer attributes using their own code. This decentralized approach eliminates bottlenecks and promotes global sharing of innovative attributes and audience definitions.
Serving
The technical requirements of the CDP are demanding. Near real-time attributes must be processed rapidly to meet the low latency demands of various use cases, a significant challenge considering Delivery Hero’s presence in over 70 countries.
Additionally, for batch-based use cases, the CDP facilitates seamless integration with other systems, allowing for the sharing of customer attributes to drive further actions and enable a wider range of use cases.
Clouds
The CDP system operates across both AWS and GCP clouds. While the main components of the CDP are hosted on AWS, we have a significant presence on GCP as well, including SQL-based and batch-based computational components. This strategic distribution also allows us to accommodate stakeholders who run their systems on GCP and require low latency. Moreover, for efficient data serving, we utilize a combination of non-SQL databases such as DynamoDB or BigTable, supported by globally distributed APIs written in Go.
The header image is generated by Midjourney. The prompt is “customer data platform cyber drawing –ar 8:3”.
If you like what you’ve read and you’re someone who wants to work on open, interesting projects in a caring environment, the CDP team is on the lookout for Content Data Analysts, Senior Data Analysts, Senior Data Engineers and Staff Software Engineers (codapro). Check out our full list of open roles here – from Backend to Frontend and everything in between. We’d love to have you on board for an amazing journey ahead.
