How often does your organization conduct a diminishing returns analysis to prioritize analytics and data science projects, and what factors are considered?

Data and models are novel asset classes because they can be monetized multiple times. The artifacts from one project will be monetizable across multiple use cases. Data and AI product strategy should lead to prioritization based on the total opportunity size, which leads to the opposite of diminishing returns for the project’s initial artifacts.

However, diminishing returns comes into play with incremental improvements to each artifact. As more complex machine learning methods are applied, costs typically scale faster than returns. Connect business metrics to model metrics to run the diminishing returns analysis.

By that, I mean what KPIs or metrics benefit from incremental improvements in model accuracy? For large models, what additional use cases can be serviced (and what’s the ROI) by incremental improvements in model functionality?

Most organizations are figuring out the data use cases that are feasible and profitable by calculating ROI in some form. But diminishing returns analysis is not widely applied to data projects.