What is a minimum viable product?
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
A minimum viable product, usually shortened to MVP, is a deliberately limited first version of a product used to learn whether the proposed value is real. It should support enough of a meaningful task for people to try it. “Minimum” concerns scope, not permission to ignore basic responsibilities.
Consider a tool for organizing reading notes. An MVP might allow a person to add a note, find it later and return to the source. It may omit shared workspaces, elaborate formatting and automatic recommendations. Those exclusions keep the first version small while preserving a complete useful activity.
“Minimum valuable” is a useful way to remember the point: the smallest complete version that lets us test whether it helps someone. It is an emphasis, not a separate technical definition of MVP. A product with fewer buttons is not necessarily more minimal in the useful sense if it leaves the central task unfinished.
An MVP differs from a prototype. A prototype is an artifact built to examine an idea and may simulate important functions. An MVP is intended to test whether a limited product actually provides value. The labels are used loosely in practice, so it helps to say exactly what is real, what is simulated and who is expected to use it.
It also differs from a pilot, which is a limited trial in an intended setting with specified participants and responsibilities. An MVP may be the version used in a pilot, but the product and the trial are different things.
For AI-enabled products, reducing scope can make evaluation clearer. A tool that summarizes one document format can be tested more directly than a general assistant claiming to handle every research task. The model—the trained component producing responses—still needs appropriate information, and the complete product needs a plan for errors and uncertainty.
A limited first product inspired by the dental-care discussion might let one team identify a patient awaiting follow-up, assign the next action and record whether it was completed. That is a small but complete operational loop. It need not include automated calls, predictive risk scores or a redesign of the clinical record. The next question is whether staff use that loop and whether it helps resolve work that otherwise remains open. Those findings should determine expansion, revision or stopping—not the availability of more features to build.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy