Wizard of Oz MVP

A product that looks automated to users while people do the work behind the scenes.

Also known as: wizard of oz MVP, flintstoning, manual backend MVP

A Wizard of Oz MVP looks like a working product to users while humans operate it behind the scenes: the “AI” is a founder with a spreadsheet, the “instant” report is typed overnight. Unlike pure concierge service, the interface is real — only the machinery is manual — so you test the actual product experience, not just the outcome.

user clicks "Analyze" → loading spinner → founder does the analysis by hand → result appears
user experience: real product. backend: a human. learning: what the product must do.

It answers the questions code cannot cheaply answer: will users go through this flow? What do they expect at each step? Where do they get confused? The interface gets validated while the expensive backend stays unwritten.

The classic mistakes:

Scale pretensions. Oz works for tens of users, not thousands. Cap usage (waitlist, invite-only) so the humans behind the curtain survive the test period.
Slow humans breaking the illusion. An “instant” feature that takes six hours teaches you about patience, not the feature. Staff the manual operation to hit the experience bar, or narrow the scope until you can.
Never building the backend. Comfortable manual operations metastasize into headcount instead of code. Define the automation trigger (volume, repeatability, funding) before starting.
Deceiving users about data. A human reading every input raises privacy stakes. Say humans review submissions, limit who sees what, and treat the data accordingly.

The ladder: fake door proves interest, concierge proves value, Oz proves the experience — then code proves it scales. Each step costs more and teaches more; skip none for a novel interaction.