Back to all tech blogs

Evolving from a Data Platform to a Data Products Platform

The Data Journey at Adevinta Spain

Here at Adevinta, a global classifieds specialist, we have transitioned from a platform aimed at users to a platform focused on data products.

We started with a data infrastructure centred on providing services to users (analysts, data scientists, data engineers, business, etc.) for ingesting, transforming and exploring data. Now we have a data infrastructure focused on the creation, management and discovery of data products by users.

But what do we mean by data products? Using the definition formulated by J. Majchrzak, a data product is an autonomous, read-optimised, standardised data unit containing at least one dataset, created to satisfy user needs.

So how did we do it?

Facing reality and focusing on what’s important

Let’s start by addressing the realities of the journey towards a Data Products Platform.

Moving from a Data Platform to a Data Products Platform requires a deep dive into what lies beneath the iceberg of a data-driven culture. It’s essential to explore the shadows, those elements that are not immediately evident in the final delivery, but which are fundamental for creating robust and reliable data products.

Because the easy part is to see the final product, for example CHAT GPT!

Under the iceberg

Undoubtedly, generative AI models, natural language processing, video and image generation, code generation, action-driving dashboards and recommendation models are the visible and attractive face of a Data Platform. These innovative solutions captivate users and have a significant impact on their experiences. However, it’s crucial to recognise that behind each of these brilliant, exciting products, there’s hard work and a solid foundation of data and processes that enable their success.

We must be aware of the realities that exist beneath the surface.

Factors like “garbage in, garbage out” become challenges to overcome, as low-quality, incomplete or biased data can lead to erroneous results and disastrous predictions.

Meme

A lack of ownership creates a lack of responsibility where no one cares about the quality and traceability of the end products. Imagine a dashboard used for high-level decision-making that was created by someone who is no longer with the company. One day, it fails or a data point is questioned, and we discover that there’s no one to provide fixes, let alone ensure that it’s being updated correctly.

Meme

Discovering this array of situations, shows us the processes that build these data products. And the organisation and reuse of transformations becomes crucial to ensuring process efficiency, as well as the quality and consistency of results.

Meme

From user-centred data to data-centred products

Being aware of complexities, the need for ownership and quality of process and data is just the starting point. There’s another important factor on this journey: thinking with a product mindset instead of just a platform mindset.

On the Adevinta journey, our Data Platform evolved from being merely infrastructure that provides services to users. Now it is the bridge between our company strategy and technological execution. It sits at the core of our data-driven culture that focuses on “building valuable data products for the company’s development.” This approach goes beyond improving the user experience; it’s about defining a shared common goal between the data platform and all of its users.

Product Vision

To change successfully you must treat data as a product and customers as consumers of the product. A valuable product solves a need, is feasible to build, is easily usable by its customers, and most importantly, has a team behind it that cares for it and nurtures it.

To achieve this you will need to embrace product thinking:

  • Considering how data should be-modelled and shared
  • Eliminating friction in its use
  • Enhancing the user experience

It’s an approach that places the value of data at its centre, allowing users to contribute to the growth and development of the product and maximise its value in decision-making and actions within the company.

Data product as a product of products

In this sense, the definition of a data product expands considerably. You have the final deliverable, and intermediate products and transformations that allow data reuse and prevent unnecessary duplication of efforts. These intermediate data sets are important for reducing complexity and maximising the scalability of the platform’s data.

Data value chain

What did we learn?

A data product platform requires a complete change of mindset within the company. Stop thinking about a service platform to facilitate users’ lives (democratising data, analytics). Start thinking about one which generates intermediate reusable and reliable data products that bring value to the company and users for decision-making and the construction of impactful products.

The change is achievable because the responsibility isn’t solely on the platform to provide service; it’s a shared responsibility between users and the platform, with a common goal of generating valuable data products for the company.

For this evolution to succeed, we must:

  • Understand the technical and cultural complexity behind a product
  • Understand the importance of data quality and process ownership
  • Ultimately, focus on a product mindset rather than a platform mindset.
  • Regularly review our goals and progress towards them.

Our Learnings

  • The data product is no longer just the end result
  • Datasets must be reused. Reusability scales and accelerates new opportunities (if it also helps to bring cloud computing costs back under control).
  • An owner for each dataset is crucial, someone who evolves them and, above all, provides transparency and confidence in their use.

Related techblogs

Discover all techblogs

Oops, I forgot to –publish! How can I connect to the container then?

Read more about Oops, I forgot to –publish! How can I connect to the container then?
Oops, I forgot to --publish! How can I connect to the container then?

Centralisation and distribution: When one node is enough

Read more about Centralisation and distribution: When one node is enough
When to consider single node alternatives to distributed frameworks

Trial by Fire: Tales from the SRE Frontlines — Ep2: The Scary ApplicationSet

Read more about Trial by Fire: Tales from the SRE Frontlines — Ep2: The Scary ApplicationSet
Understand the safeguard configuration of the ArgoCD’s ApplicationSet through the experience of our SRE who learned from an incident