This section just covers the management and delivery when data models are a component of a product.

<aside> ℹ️

Note: Not all solutions with AI today involve data models initially due to the robust LLMs and GPTs on the market today. This section just covers Data Modelling specifically.

</aside>

Agile in Data Model Management

Agility is still relevant. There will still be an appetite for incremental progress, quick adaptations and retrospective reflections to improve the processes. There will still be the desire to change as the market changes and have processes to listen to that market for feedback.

However, common Agile processes that were created to serve a more linear development cycle are not going to apply well to managing data teams. Since model development relies on creativity, failure, grinding work and analysis, it’s important that the team doesn’t work in linear methods provided by the Scrum or Kanban Model initially.

However however, Data models are likely a component in an ecosystem of technology. Other teams or problems can (and should!) be structured as more traditional Agile teams to ensure transparency in progress and delivery to the business.

CRISP-DM

CRISP-DM is the current going model that reflects building modelling. It sits counter to the Software Development Lifecycle (SDLC) in that it does not have a linear path to delivery.

Essentially, it allows for logical checkpoints with the Business Understanding through the model development process, while allowing times for trying and failing.

Business Understanding

It’s important for a Product Manager to cultivate a strong connection to the stakeholders and customers that will be using this model throughout the entire process. While, this is a similar task to traditional software product management, here are some of the ways that it might differ:

• Involved risk and compliance stakeholders - this technology still has enough variance (80-90% success rates on good days) that Risk and Compliance colleagues are going to be important people to keep in the loop at all stages of the process. They will be important allies if problems arise, but they need to be brought into the conversation early and often.
• More proactive education - this is still a relatively new way to do technology. It’s important to not only inform other members of the company, but also to tell really good stories that help your colleagues and customers understand what you are trying to build and why.
• 10x Concepts - this technology is expensive to build, and ROI might not be certain for the first iteration of the product. It’s important to keep the 10X idea in focus, since building a data model will likely (rightfully) get questioned as the process wears on.
  • If the problem to be solved may have an easier path to ROI with a simple heuristic equation, maybe that would be an option to avoid building a data model.

Data Understanding