The Wizard of Oz MVP
Test whether people want an automated product — before you build the automation. A Wizard of Oz MVP looks fully automated, but humans do the work behind the curtain. Here's how to run one, what to measure, and when to use it instead of building the real tech.
A classic pretotyping method · Built at Google, taught at Stanford
Banked
$68M+
Fund the best, save on the rest.
What is a Wizard of Oz MVP?
A Wizard of Oz MVP — also known as Wizard of Oz testing or a Wizard of Oz experiment — is a pretotyping method where users interact with a product that looks fully automated, but humans secretly do the work behind the scenes. Like the man pulling levers behind the curtain in the 1939 film, you build only the interface — then fulfil every request by hand. It lets you test whether people value the experience before you invest months building the real technology. It’s one of the fastest ways to validate an idea that depends on AI, automation, or a complex back-end. It fits into a broader plan for how to validate a business idea, and often follows a fake door test once you’ve confirmed there’s real demand.
$50M
Aardvark validated demand with humans routing questions, then sold to Google
Aardvark / Google
5–20
pilot users is enough to learn from a Wizard of Oz test
Pretotyping playbook
90%
of new products fail — most never tested whether people wanted them
Harvard Business School / Christensen
How to run a Wizard of Oz test in 6 steps
Build the front, fake the engine, and match the timing the real product would deliver — so the signal stays honest. See all 10 pretotyping methods →
Build only the interface
Create a functional-looking front end — forms, buttons, outputs — with no real back-end logic. The user should believe it's automated.
Set up the curtain
Wire a private channel (email, Slack, a simple dashboard) where a human receives each request and returns the result.
Recruit 5–20 pilot users
Enough to learn from, few enough that your manual process can keep up. Use real target users, not friends being polite.
Operate manually — match the real timing
Fulfil requests by hand, but deliver at the speed the automated version would. Response time changes how people judge the value.
Measure usage and willingness to pay
Track how often people return, how satisfied they are, and — critically — whether they'll pay. Behaviour beats stated intent.
Decide: automate or pivot
If people love the output, build the real technology with confidence. If they don't, you've saved months of engineering on the wrong thing.
Same manual work — different curtain
Both methods have humans deliver the value by hand. The difference is whether the customer knows. Pick the one that matches the assumption you're testing. See the concierge MVP method →
| Dimension | Wizard of Oz MVP | Concierge MVP |
|---|---|---|
| Do users know humans are involved? | No — the work appears fully automated | Yes — the manual, high-touch service is explicit |
| Best for testing | Whether the solution / output is valued | Deep problem discovery — what customers actually need |
| What you build | A polished interface; humans fake the engine | Often nothing — you deliver the value in person |
| Data you get | Behavioural signal at a small, automated-feeling scale | Rich qualitative insight from watching people up close |
| Reach for it when | The risk is in expensive tech (AI, ML, automation) | The risk is in not understanding the customer yet |
Wizard of Oz MVPs, answered
What is a Wizard of Oz MVP?
A Wizard of Oz MVP is a pretotyping method where users interact with what looks like an automated product, but humans secretly do the work behind the scenes. You build only the interface and fulfil each request by hand. It tests whether people value the experience before you build the real technology — ideal for AI, automation, or complex back-end ideas.
What's the difference between a Wizard of Oz MVP and a Concierge MVP?
Both have humans doing the work manually, but the difference is visibility. In a Wizard of Oz test the human is hidden — users think it's automated. In a Concierge MVP the human help is explicit — you openly deliver the value as a high-touch service. Wizard of Oz tests whether the solution is valued; Concierge tests what customers actually need.
Is Wizard of Oz testing ethical?
Yes, when done responsibly. Ensure the output quality matches what the automated version would deliver, never collect sensitive personal data under false pretences, and comply with privacy rules. Many teams disclose that the beta is 'powered by our team'. The goal is to learn whether people value the experience, not to deceive.
What's the difference between a Wizard of Oz MVP and a fake door test?
A fake door test measures demand before anything works — it counts clicks or sign-ups for a product that doesn't exist. A Wizard of Oz MVP goes a step further: the product appears to work and actually delivers value, but a human powers it. Fake door tests raw interest; Wizard of Oz tests whether the working experience is good enough to keep people coming back.
Is a Wizard of Oz MVP the same as Wizard of Oz testing?
Yes — "Wizard of Oz testing" and "Wizard of Oz experiment" are just other names for running a Wizard of Oz MVP: putting an automated-looking interface in front of users while a human quietly powers it behind the curtain. The label doesn't change the method — the point is to test whether people value the experience before you build the real automation.
What are real examples of Wizard of Oz MVPs?
Aardvark, a social search engine, used interns to manually route questions to experts while users thought it was automated — it validated demand and Google acquired it for $50M. CardMunch claimed 'OCR technology' to digitise business cards, but humans on Mechanical Turk transcribed them; LinkedIn later acquired it. Both validated demand before building the real technology.
Want help faking the engine before you build it?
Twenty minutes with Leslie Barry. We look at one idea on your roadmap, design a Wizard of Oz test, decide what to measure, and tell you whether pretotyping is the right next step. No pitch.