Learning-First MVP
Launch the smallest experience that resolves the next uncertainty
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 98%
A Learning-First MVP is the smallest launchable experience that answers the next material product or market question. Instead of treating “minimum” as permission to ship a cheap miniature of the final product, define the behavior that would count as evidence—such as a signup, reply, usage event, procurement effort, or payment. Build only what is necessary to expose that behavior, then watch what prospective customers work to obtain. Use those signals to identify the specific value or capability they are buying. Deliver enough to serve the earliest adopters, learn from them, and only then streamline, automate, and add depth. The mechanism limits wasted development by sequencing investment after evidence. Ernsten narrows the advice to products with low technical risk and explicitly distinguishes it from scientific products where whether the underlying intervention works is itself the critical question.
Origin
Extracted from The Foundr Podcast
Core principles
- 01An MVP exists to generate learning
- 02Demand evidence should precede broad technical investment
- 03Build only enough to deliver the promised value
- 04Early customers reveal which capability they actually buy
- 05Strong customer effort can signal an acute problem
How to run it
- 1
Define the learning goal
State the customer or market uncertainty the MVP must resolve. Do this before selecting features or technology.
Watch out Do not define success as merely launching the artifact.
- 2
Select a behavioral signal
Choose an action that would provide evidence of interest or need, such as signing up, replying, using the offer, or paying. Decide what you will observe before launch.
Pro tip Use stronger signals as soon as they are practical.
Watch out Compliments alone do not establish that someone will use or buy.
- 3
Build the minimum test surface
Create only enough of the experience to let relevant people take the target action. A landing page or outreach message may precede a functioning product.
Watch out A low-quality miniature packed with partial features may teach less than one complete interaction.
- 4
Find acute early demand
Look for users willing to overcome friction because the problem matters to them. Invite them into the learning process rather than hiding every limitation.
Pro tip Notice extraordinary effort, such as navigating procurement or tolerating an unfinished workflow.
Watch out Do not infer a broad market from one determined buyer without further tests.
- 5
Build the purchased value
Determine which capability customers are actually pursuing and implement enough to deliver it well. Delay unrelated features.
Pro tip One decisive capability can matter more than five speculative ones.
- 6
Iterate and industrialize
Learn with early customers, then streamline, automate, and expand the proven workflow. Let repeated evidence justify each increase in scope.
Watch out Do not automate a value proposition that remains unproven.
In the wild
Before knowing which buyer role and message would resonate, Alpha bought role- and industry-specific email lists and tested value propositions through outreach. Ernsten says Citibank, AT&T, and Pfizer became its first three clients through somewhat different messages, after which the company could learn what those buyers needed it to deliver.
→ Customer response and purchasing effort guided what Alpha built rather than a full speculative feature set.
Common mistakes
Building the cheap final-product miniature
An MVP chosen for low cost rather than a learning objective may deliver poor value and produce ambiguous evidence.
Shipping every imagined feature
Customers may buy for one capability. Building five before identifying that one delays learning and consumes resources.
Ignoring the type of technical risk
The approach assumes feasibility is not the main unknown. Scientific and safety-critical work requires evidence appropriate to those risks.
Is it for you?
Best for
Software, app, and service concepts where technical feasibility is known but demand is uncertain.
Not ideal for
Biotech, drug development, or other products whose central uncertainty is scientific or safety feasibility.
From the transcript
“what is the thing we can launch to learn something”
“you have to build very very very very little”
“you only have to build one instead of five”
From the episode
336: Starting a Business During a Crisis with the founder of Alpha and Strata, Thor Ernstsson