The six principles for how we run Data & Insight at Oda
We have formulated six principles capturing the heart, soul and aspirations for how we think about value creation from data and insights at Oda. In this post we introduce these and lay the ground for digging into each of these in following posts.
Why principles?
Oda is an online grocery player now scaling from its origins in Norway to Germany and Finland. We exist to build the world’s most effective retail system to create a society where people have more space for life. Data, insights and algorithms are at the core of our end to end retail system, but also ingrained in our processes and way of working.
Aligning on how to best approach value creation is easy when you are five data people sitting together. But since 2019, Data & Insight at Oda has grown from 3 to 75 people, spread across Oslo, Berlin and remote, and almost 30 teams. At our current scale we need a common vision of how to further develop processes and products with data. Therefore, we have formulated six principles to guide our decision making and our approach to working with data. This captures the heart and soul of our community and its ambitions to use data to create value for our customers, society and owners of Oda. It also makes it easier for people in other disciplines to understand how we work and make decisions.
Our six principles
Our six principles work together as a holistic strategic and cultural foundation for our way of working. They are equally important and ordered to make it easier to explain how they build upon each other. Agreeing on only six principles is not easy. We choose those that set us apart and that we need to focus and work on to achieve our goals.
Domain knowledge + Discipline Expertise = 🚀
In order to produce world class results we believe our people not only need world class data skills, but also deep knowledge of the problem domain. The best way to acquire such in-depth knowledge is to work in a cross functional domain team over time, solving real-world problems together with domain experts, Software Engineers, Product Managers and UX designers. We believe this increases our ability to solve the right problems in the right way, both in the long term and the short term.
We don’t believe in full decentralization nor full centralization, but rather a place in between depending on the data maturity and complexity of the domain that varies even within Oda. Picking operations at our fulfilment centers is a good example of high maturity where domain knowledge is crucial. Using data on areas such as recruitment is still lower maturity and can be supported by a central team on a case by case basis. Sense of affiliation to a domain and our discipline, relationships with like-minded peers, and professional development are all factors that ensure our people, discipline and organization thrive in the long term, and they also play a key part in our balancing act.
Distributed data ownership, shared data governance
Given our industry and wish for speed and flexibility, our data strategy is fundamentally offensive, that is, we value autonomy and self-service in using data. Distributed ownership of everything data (including pipelines, dashboards and algorithms) is key in order to succeed at these objectives at scale.
Our data is divided by domain, and each domain team is responsible for the whole data value chain, from data production, data pipelines, and products, all the way through to seeing and measuring impact.
Thanks to a shared set of tools, infrastructure and rules agreed between domain teams and Data & Insight Platform, our data is made interoperable so that everyone at Oda can easily use it. That is the core to our shared data governance. It also include concrete things like naming conventions, design rules for modelling and transformation of data and how to structure code and projects.
Data as a product
We strive to use a product mindset on our platforms, datasets and dashboards to reduce the cost of discovering, understanding, trusting, and ultimately using quality data. Domain teams consider data as their products, and Data & Insights Platform treat their capabilities as products too. When data is considered as a product it should have the following characteristics [1]:
- Discoverable
- Addressable
- Trustworthy
- Self-describing
- Interoperable
- Secure
Our colleagues in the rest of Oda are our customers and we work with them to understand the business impact of their needs, define what success means and work towards measuring the outcomes and impact of our products rather than the outputs.
Enablement over handovers
In order for everyone at Oda to work effectively on solving hard problems, we believe it’s better to enable and empower people to work autonomously and end-to-end rather than relying on others to do parts of it for them. We follow this philosophy both within and outside of the discipline. As an example a Data Engineer will build tools, infrastructure and guidelines that enable a Data Scientist to work as much as possible end to end without handover to get a model into production. In the same way, a Data Analyst will enable a Business Controller to be self served with finding the data and insights they need. By minimizing the number of handovers, we are maximizing the speed and agility in which teams are able to move to solve their problems, and optimizing for flow and scale.
Impact through exploration and experimentation
As we mature in how we use data to develop Oda and our products, it becomes less and less obvious which changes will have the largest impact. We are working on novel problems that arguably no-one has really solved in a very good way before. In Data & Insight, we are uniquely positioned to ensure that we as a company validate, test, measure and learn and act accordingly.
With data as a key enabler to build the world’s most efficient retail system, our way of working must also take into account the explorative and experimental nature of data work. In most cases we cannot know ahead of time if we will be able to collect the right data, that it will have sufficient signal and that we are able to build a model that improves the current state. The potential upsides here should though justify taking bets on such projects. Goals need to be set in such a way that they appropriately balance short-term and long-term returns. Our culture needs to support a steady process of learning by doing and being comfortable with failure and ambiguity.
Proactive attitude towards privacy and data ethics
We take a proactive and preventive attitude on ethics and privacy; which means we collect, process, store, use and delete data responsibly and consciously, and don’t build products that are adversarial to our customers.
As experts on data, analytics and algorithms, we have an extra responsibility to make sure that we:
- Incorporate privacy principles (privacy by design) into why and what we do
- Set metrics that balance customer’s long term goals, business outcomes and impact on society and the environment
- Build and apply algorithms that help all people live fulfilling and sustainable lives
We want these aspects to be considered from the start of all the work we do and not be considered as an afterthought when building data products.
Making sure the principles have an impact
Putting principles on paper is only the first step. Making sure we interpret them in a similar way and live by them is where the real challenge lies. We need colleagues in other disciplines to understand and agree to succeed with this way of working. Spending time on communicating them (including this blog post!) and putting them into practice in our daily work, is our current focus. We will also embark on a gap analysis on how we are living up to the principles this fall, and make a prioritized road map. Initiatives ranging from data literacy to technical capabilities are already ongoing. They are needed to support distributed ownership and enforcement of best practice guidelines.
We are already seeing that these principles, and our overall Data & Insight operating model, is guiding decisions and our work on a daily basis. Even the common process to formulate these principles gave a motivation boost to improve and a common direction for it! We would love to hear if you have experience on working this way and how you made it work.
Stay tuned for more
Our current principles are an iteration on those formulated in 2019 when we where less than ten people in the discipline. Since then we have grown and matured, and see development in best practices and technology. We therefore saw the need to update, also to create new goals to aim for. The formulation of the principles have been done by a big group of people at Oda, and is the synthesis of all these great minds and hearts. You will hear from some of them in the following posts where we dig even further into what the principles look like in practice.
You can learn more by listening to a podcast episode recorded in Sep 2022 where I talk about these principles with the Norwegian Data Management association. In addition we have written a more detailed post about distributed ownership, shared data governance and how we treat data as a product.
You can also read more about Data & Insight and our way of working at the Oda Product & Tech Medium blog. The work on modeling service time (part 1 and part 2) is an excellent example of applying the principle of proactive attitude towards privacy and data ethics in practice. You might also find the series on how we define our three analytical roles in Data & Insight interesting if you liked this one.
[1] Characteristics of data when considered as a product borrowed from Zhamak Dehghani’s article How to Move Beyond a Monolithic Data Lake to a Distributed Data Mesh
