Beyond the lifecycle: Why your company needs data product managers to succeed with analytics
One of the biggest issues with analytics model development is often seen as the final stage of the analytics lifecycle: deployment into production. Many models never make it this far, even if the prototype does exactly what was required. Organizations have generally found that the answer to this problem is to think about deployment during development.
In other words, even as the model is developed, the data scientists, business team, and IT team need to be thinking together how the model will be deployed. This will include consideration of the systems required to run it, whether it will need to be ‘translated’ to run on those systems, and how it will be monitored. This makes it easier to deploy the model and keep it running in the future.
Recently, however, companies that are more mature and higher performing in analytical terms have started taking this a step further. They have begun thinking beyond the analytics lifecycle to create data or analytics assets that can be used over and over again for different purposes. The key to this is data product management, and the role of the data product manager.
Introducing data product management
Product management is a fairly standard discipline in many tech companies. It has been described as being at the interface between user experience, business and technology teams. It is broadly about delivering great products that meet customer needs. It is particularly important within agile product development, where the product manager is responsible for keeping the process on track and consistent with customer needs.
In organizations using data and analytics, the role is similar. In a recent article in Harvard Business Review, Thomas Davenport, a co-founder of the International Institute for Analytics (IIA), suggested that data product managers are responsible for two types of products: data products, or reusable datasets, and analytics products, or models that can be reused or repurposed. Data managers are responsible for managing the product development process with an eye to future use.
They therefore work closely with both data scientists and business teams, coordinating the process and communicating and translating between the two. Ultimately, they have the responsibility for ensuring that models and datasets are deployed into production. This means making sure that they are usable and valued by customers. Data product managers therefore need a very good understanding of user needs, and must be able to speak the language of both the business teams and the data scientists. They need to understand data, but also the business.
After deployment, data product managers are then also responsible for managing the ongoing life of those products. This means monitoring their use and deployment over time, to ensure the ongoing value of data and analytics assets.
Finding the balance: data product management in practice
What kinds of people make good data product managers? The answer to that, interestingly, is very much NOT data scientists. They are usually far too interested in the details of the model, and refining it to make it fit for purpose. Interestingly, data scientists often want to take a model further than is strictly ideal for the product manager, who is likely to favour a more generalist and therefore reusable model. This points to an interesting tension at the heart of any data product development process. In practice, organizations often find that there are three roles that need to be balanced: data scientist, engineer and data product manager. These three need to be jointly responsible for negotiating the product development process, and compromising on their individual priorities.
A new name for an old problem?
Interestingly, the need for this role is not new. We have been talking for some time about how organizations need someone who can translate between data science, IT and business teams. In the past, we have often seen this as the role of data scientists. We have discussed how they can and should develop a better understanding of the business. We have also supported the trend for business users to develop their analytics and data skills, and become citizen data scientists to bridge the gap.
It therefore makes sense that high performing analytics organizations have formalized this role of data product manager. The only question is how long it will take before this is seen as standard for success with analytics. There is still an early-mover advantage to be had here, and wise organizations are likely to adopt this product mindset rapidly.
