Return on Investment

The first question is always: what do you mean by return?

— Douglas Hubbard, How to Measure Anything [1]

Where does ROI fit in the outcome-driven product development cycle?

The model at the heart of AI-Native SAFe is the outcome-driven product development cycle, which applies across the Portfolio, ART, and Teams (see Figure 1). Portfolio leaders and business owners use forecasted ROI to develop outcomes and prioritize initiatives to make economically attractive investments. Actual results generated by products and solutions generate ROI that can be used to evaluate past investments and guide future ones.

Why does ROI modeling matter more in the age of AI?

AI investments can be large, fast-moving, and difficult to evaluate. GPU infrastructure, foundation model licensing, data engineering, agent development, and the construction of a retrieval-augmented generation pipeline are examples of common AI-Native investments that require significant capital and operational outlays. Business cases stated in qualitative terms, such as "this will improve productivity" or "this will accelerate delivery,” are unverifiable claims, and unverifiable claims are poor means to assess portfolio investment decisions.

AI also enables organizations to build and ship features at a pace that was previously impractical. This is a financial risk if the economic impact of these features has not been modeled before development begins. Accelerating the delivery of low-ROI features does not improve product or portfolio performance. It compounds investment waste faster.

Additionally, token costs introduce a variable operating expense that scales directly with adoption. AI-Native features that rely on inference costs must include those costs in the economic model to avoid overstating ROI. The organizations that benefit most from AI will be those that apply rigorous ROI discipline to AI investment decisions, not those that simply spend the most.

How do you model ROI?