Viability is No Longer Enough

We are taught to build the Minimum Viable Product (MVP). But in the era of Artificial Intelligence, a technically viable product that is discriminatory, unexplainable, or dangerous is simply a liability. The new standard for launch isn’t Viability; it’s Acceptability.

I propose that every Product Manager launching an AI-powered solution must now aim for the Minimum Acceptable Viable Product (MAVP).

The MAVP is the smallest set of features that delivers customer value while adhering to ethical standards, governance requirements, and user expectations of fairness.

The MAVP Mandate: Three Conclusive Steps

1. The Governance Blueprint: Before writing a single line of code, you must define the Human-in-the-Loop strategy. For which critical decisions (e.g., denial of credit, medical diagnosis) will the AI act as an assistant, and for which will a human always have the final say? This defines the acceptable level of autonomy and risk for your product.
2. The Fairness Test: Your acceptance criteria must now include fairness metrics. Instead of just maximizing overall accuracy, you must test the model’s accuracy and performance across all defined demographic segments. If performance drops for any minority group, the product is not ready for launch.
3. The User Consent Contract: Beyond standard legal terms, MAVP requires transparent user communication. Users must understand how their data is being used to train the model, how the AI’s decision was reached (where possible), and how they can appeal or provide feedback on an automated decision. Trust is built on clarity, not concealment.

The Responsible Scaling of Innovation

The greatest Product challenge of this decade is not how fast we can build AI, but how responsibly we can launch it.

By adopting the MAVP standard, you, the Product Leader, transform from a risk-taker into a Strategic Steward of Innovation. You move your business away from the “Algorithm Cliff” and toward sustainable, ethical, and profitable growth.

The future of Product Management is responsible AI. Are you building it?