Data contracts at Nandos

Ben Watson at Nandos, a South African restaurant chain operating in 30 countries, has a great write up on their modern data platform with data mesh and data contracts.

It’s interesting how they started with a centralised data team ingesting data as raw as possible, and found they could deliver some value quickly:

This model enabled us to quickly pull together Nando’s core datasets such as sales, restaurant metadata and opening hours, and expose them in Looker for downstream consumption.

They then found the limitations of this model:

Of course we quickly found that this model has its limitations. It’s not easy for one team to take on the responsibility of understanding and answering questions about all data in a company — from stock to sales, labour to forecasts, smart fridge sensors to customer loyalty.

So they changed to a more decentralised approach and adopted a strategy based on data mesh where domain teams own their own data:

With data under their control they could build data-driven applications from the start, have more control over their platform, and the data would stay in the hands of the people who best understand it.

The problem they solved with data contracts is this:

We often get questions like “which tables do we share with external company X?” or “if I change this table then who will be impacted?”. Being able to quickly answer these questions and have confidence that this information is up-to-date is important for maintaining a stable data platform that can iterate quickly without constantly breaking hidden dependencies on other teams.

That’s a typical problem which data contracts can help any organisation solve.

They then discuss in detail how they implemented this on Google Cloud (Nandos prefer to build their own tools), so well worth reading the article if you’re interested in that.

They finish the article discussing data culture, and I loved this sentence:

Running a successful data platform requires more than just technical work and as such much of our effort is culture-focused.

That’s then followed with some specific actions they took to improve the data culture.

Check out the full article from on Medium.