What MVP Actually Means
MVP does not mean small version of the full product — it means the cheapest test of your single riskiest assumption.
🎯 By the end you'll understand
Why the conventional reading of MVP ("small but complete version of the product") produces over-built, under-tested experiments. The correct definition: the minimum test of the riskiest assumption. The five MVP archetypes — concierge, Wizard-of-Oz, smoke test, single-feature, and pilot — what each one tests and, critically, what each one fails to test. Why a prototype is not an MVP. And how Forge picks its MVP type given a specific constraint set and a specific riskiest assumption.
KEY INSIGHTAn MVP is an experiment, not a product
The word "minimum" in MVP does not describe the product — it describes the test. The question is never "what is the smallest product we can ship?" It is "what is the cheapest test that could prove our riskiest assumption wrong?"
The MVP concept was introduced to stop founders from spending six months building something no one wants. But the idea has been so badly corrupted in common usage that it now does the opposite: founders build a "small version of the full product," ship it with pride, learn almost nothing, and interpret the absence of catastrophic failure as validation. The problem is not the concept — it is the redefinition from experiment to small product.
Eric Ries defined an MVP as "the version of a new product which allows a team to collect the maximum amount of validated learning about customers with the least effort." The operative word is validated — meaning a customer decision under real conditions with real stakes. Not a demo. Not a prototype. Not a beta. A test that generates a verdict. If your MVP cannot produce a clear "this worked" or "this failed" signal, it is not an MVP — it is a more expensive prototype.
DEFINITIONThe correct definition — and why the wrong one is so sticky
Here is the correct operating definition: An MVP is the minimum set of activities — possibly involving no software at all — that tests whether a real customer will use and pay for your proposed value, under real conditions. Three parts matter.
Minimum — designed around a single assumption
Not minimum in the sense of "fewer features than we wanted." Minimum in the sense of "nothing beyond what is required to test this one specific assumption." If your riskiest assumption is "customers will pay before they see the product work," your MVP is a payment page, not a product. If your riskiest assumption is "our algorithm outperforms the incumbent," your MVP needs to demonstrate that comparison — not everything else. Minimum is scoped to the assumption, not to a product vision.
Viable — the customer experiences real stakes
Viable does not mean "works reliably." It means "the customer is in a real decision context with real stakes." A demo has no stakes; a pilot where the customer is betting production data on your output has real stakes. The stakes requirement is what turns a positive response into a signal. People are uniformly enthusiastic about things that cost them nothing and commit them to nothing. Viable is the filter that makes the signal trustworthy.
Product — something the customer actually uses
Not a pitch deck. Not a concept video. Not a mock UI. Something a customer uses to do something real — even if what enables that "use" is a team of humans operating behind a curtain. The distinction between "product" and "presentation" is the difference between revealed preference and stated preference. Stated preference is worth almost nothing. Revealed preference — a customer who comes back, who tells a colleague, who asks when the next version ships — is the real signal.
The wrong definition — "small version of the full product" — is sticky for a specific reason: it feels like shipping. Building a stripped-down version of your product activates the same dopamine loops as building the full product. You write code, you make design decisions, you have something to show. The correct definition of MVP often involves almost no code, which feels profoundly unsatisfying to a technical founder. The discomfort is the point. If building your MVP feels like building your product, you are probably building the wrong thing.
CORE CONCEPTFive MVP archetypes — and what each one actually tests
There is no single "MVP format." The right format depends on which assumption you are testing and what constraints you are operating under. Here are the five canonical archetypes, what they test, and — critically — what they do not test.
Concierge MVP
You deliver the promised value manually, without software, by doing it yourself for the customer. Dropbox could have manually backed up files before writing a line of code. Forge could manually review inspection images and deliver defect reports to one plant. What it tests: whether customers value the outcome enough to use it and return. Whether your understanding of the workflow is accurate. What it does not test: whether the automated version is technically feasible, scalable, or commercially viable at scale. The concierge is a workflow validator, not a product validator. Typical build time: days to 2 weeks. Failure signal: customers use the service once, then stop returning without explanation — or you discover the workflow you assumed was the bottleneck is not where the value actually comes from.
Wizard-of-Oz MVP
The product appears to work automatically — the customer experiences it as a real product — but humans are operating the machinery behind the interface. The technique — simulating system behavior with a hidden human operator — has been used since early HCI research in the 1980s, most notably formalized by John Kelley. For Forge: an edge device appears to run real-time CV inference; in reality, images are streamed to engineers who annotate defects manually and push alerts. What it tests: whether customers will use the interface as designed, respond correctly to the alerts, and integrate the output into their workflow. What it does not test: whether the automated inference is accurate enough, fast enough, or reliable enough for production. Typical build time: 1–3 weeks. Failure signal: customers receive alerts but do not act on them — the workflow integration assumption was wrong regardless of how well the detection works.
Smoke Test MVP
A landing page, sign-up form, or pre-order page that describes the product and asks customers to commit — sign up, pay a deposit, join a waitlist — before the product exists. Dropbox's explainer video is the canonical case: approximately 75,000 sign-ups before a line of product code was written. What it tests: whether customers who understand what you are building well enough to commit actually will. Demand signal under conditions of stated intent. What it does not test: whether customers will actually use the product once it exists, or pay full price when the commitment is real rather than a deposit. Stated intent has a conversion rate — the ratio of sign-ups to actual paying customers is almost always much lower than the initial signal suggests. Typical build time: hours to 3 days. Failure signal: sign-up rate below 5% of targeted traffic, or sign-ups that go cold when you follow up with a real sales conversation.
Single-Feature MVP
Build exactly one feature — the one that delivers the core value proposition — and nothing else. No onboarding flow. No settings. No dashboard. Just the thing. What it tests: whether the core value proposition, in its simplest form, is worth using. This is the closest to the conventional "small product" framing, but the key distinction is that it is scoped to one specific hypothesis, not a reduced-scope product roadmap. What it does not test: whether customers will stay (retention requires the full surrounding context: onboarding, notifications, support), or whether they will pay commercial pricing (free or low-priced single features have different demand curves than enterprise contracts). Typical build time: 2–6 weeks for software; longer for hardware. Failure signal: activation happens but D7/D30 retention collapses — customers try it, do not return, and the core value loop is not self-sustaining.
Paid Pilot MVP
A real customer pays real money for a defined, time-bounded deployment of your product — even if parts of it are manually operated. This is the highest-stakes MVP format and the most definitive test of commercial viability. What it tests: willingness to pay at a real price point under procurement conditions (budgets, approvals, procurement timelines, integration requirements). Whether the outcome produces enough value to justify renewal. What it does not test: whether the product can scale beyond this customer, or whether the onboarding and support overhead is sustainable. The pilot is not a product-market fit signal — it is a commercial viability signal for this specific customer at this specific price. Typical build time: weeks to months for the product; the sales cycle is often longer than the build. Failure signal: customer completes the pilot and declines to renew, or pays but never actually deploys the product in a meaningful workflow — "shelfware" signal.
| Archetype | Riskiest assumption tested | Build time | Cash required | Failure signal |
|---|---|---|---|---|
| Concierge | Customers value the outcome & return | Days–2 wks | Near zero | No return usage without explanation |
| Wizard-of-Oz | Customers will use the interface & act on output | 1–3 wks | Low (<$5K) | Alerts received but workflow unchanged |
| Smoke Test | Stated demand exists at scale | Hours–3 days | Minimal | Sign-up rate <5%; cold follow-ups |
| Single-Feature | Core value loop is usable | 2–6 wks | $2K–$20K | Activation without D7/D30 retention |
| Paid Pilot | Commercial viability at real price | Wks–months | $5K–$50K | Pilot completes; customer declines renewal |
SEE ITMVP Type Selector — find the right experiment for your constraint set
FORGEChoosing the right MVP for Forge's Stage 2 milestone
Forge's Stage 2 milestone is specific: three paid pilot contracts signed at $8K each, edge device plus dashboard MVP deployed at one manufacturing site, and first activation data collected. That milestone determines the MVP type. Let's work backward from the milestone to understand why.
The riskiest remaining assumption: will customers integrate and act on the alerts?
Stage 1 established that the problem is real and customers will conditionally commit to a pilot. The riskiest assumption that Stage 1 did not close is whether customers will actually change their workflow based on Forge's alerts — whether the ops manager who says "yes, this looks interesting" will, when the moment comes, require line workers to stop and investigate every alert the system raises. This is a change management assumption, not a technical assumption. The CV algorithm may work perfectly and the product may still fail here.
Why Forge uses a Wizard-of-Oz MVP first, then a paid pilot
The Wizard-of-Oz MVP lets Forge test the workflow integration assumption before spending months building the automated inference pipeline. Forge deploys a camera at one line, streams images to engineers who annotate defects, and pushes alerts to the ops dashboard. If line workers check the alerts and escalate to the QC team — workflow integration confirmed. If alerts go unacknowledged for eight hours — the change-management assumption is wrong and Forge needs a different activation design before the paid pilot. Running the Wizard-of-Oz MVP for 2–3 weeks costs almost nothing relative to the cost of a failed paid pilot.
The paid pilot is the commercial viability test — not a product launch
Once the workflow integration assumption is validated, Forge signs a paid pilot at $8K for 60 days. The pilot is designed to answer one question: will the customer renew, expand, and recommend? Not "does the product work?" — they already know that from the Wizard-of-Oz. The pilot generates real activation data (do customers complete onboarding? do they check the dashboard daily?), real economic signal (does the savings calculation they did in Stage 1 hold?), and a real renewal decision. Three pilots at $8K each is $24K in Stage 2 revenue — enough to demonstrate commercial viability to a seed investor and enough to fund Stage 3.
Forge's Stage 2 milestone — three paid pilots at $8K each, one site deployed, activation data collected — is not a product roadmap. It is an MVP specification. It defines exactly what assumptions need to be tested (commercial viability, workflow integration, onboarding completeness), the minimum commercial commitment that constitutes a test (real money at $8K), and the activation metric that serves as the proxy for value (defect detection improvement in 30 days). Every feature decision in Stage 2 is evaluated against one question: does building this feature improve the probability of passing this test? If not, it does not get built.
WEAKNESSThe prototype trap — confusing technical validation with customer validation
The most expensive mistake a technical founder makes is confusing a prototype with an MVP. They are testing opposite things and the distinction matters enormously.
| Prototype | MVP | |
|---|---|---|
| Tests | Technical feasibility — can we build it? | Customer viability — will they use it and pay for it? |
| Who uses it | Engineers, internal reviewers, demo audiences | Real customers in real work contexts |
| Stakes | None — it is a proof-of-concept | Real — customer is using it to do real work |
| Success signal | "It works as designed" | "Customer returns, pays, or expands" |
| What a positive result proves | The technical problem is solvable | The customer pays for the solution to the customer problem |
| What a positive result does NOT prove | That customers want it, will use it, or will pay for it | That the automated version is technically feasible at scale |
For Forge, a technical prototype — a CV model that achieves 98.7% precision on a benchmark dataset of defect images — proves that the technical problem is solvable. It does not prove that ops managers will change their QA workflow, that the product can be deployed at a real facility within an IT-approved security perimeter, that the maintenance team will support the hardware, or that the $8K price point clears the procurement process. These are all customer-side assumptions that no amount of technical prototyping can answer. Founders who conflate the two spend six months building a better prototype when they need to spend six weeks building a worse MVP.
A demo is not an MVP. A demo is a prototype on a presentation stage. The feedback you get from a demo — "this looks impressive," "I can see how this would be useful," "we should set up a follow-up call" — is systematically biased toward positive responses because the audience knows it is a demo, has no stakes, and is in a social context that rewards polite enthusiasm. Treating demo enthusiasm as customer validation is the single most common path to a product that works but nobody buys. The distinction to enforce: was the customer using it to do real work, on real data, with real consequences if it failed? If no, it was a demo.
WEAKNESSWhere this bites
Every MVP archetype is designed to test a specific class of assumption. Running a smoke test when your riskiest assumption is workflow integration generates irrelevant signal — stated demand does not tell you whether a customer will change their QA process. Running a Wizard-of-Oz when your riskiest assumption is pricing generates irrelevant signal — customers will use a manual service they are not paying for. The discipline is to name the single riskiest assumption first, then select the MVP archetype that tests it most directly. Founders who reverse this — pick an MVP type they are comfortable with, then retrofit an assumption — build the wrong experiment and either miss the real risk entirely or get a false positive that sends them in the wrong direction.
The absence of a clear failure signal is not a pass signal. A concierge MVP where customers used the service once and did not complain is not evidence that customers value it — it is a neutral result. A pilot where the customer received every deliverable but never integrated the output into a workflow is not evidence of value — it is a shelfware signal. MVP design must specify, in advance, what the pass signal looks like (not just the fail signal). "Customers did not leave" is not a pass signal. "8 of 10 customers returned for a second session without being invited" is a pass signal. Without a pre-specified pass criterion, every MVP result gets interpreted as a pass by the founder who needs to move forward.
One pilot customer who loved the product is not a pattern — it may be an outlier, a friendly design partner, or a customer whose problem profile is unusual in ways you do not yet understand. A smoke test that converted 12% of traffic on 40 visitors is not a 12% conversion rate — it has an enormous confidence interval. MVP design in a B2B context requires thinking carefully about minimum sample sizes. For a pilot: three to five paid customers is a minimum to start pattern-matching across rejection and acceptance behaviors. For a Wizard-of-Oz: at minimum two simultaneous sites to avoid mistaking one site's organizational quirks for universal behavior. For a smoke test: at minimum 200–300 targeted visitors before drawing any conclusion about conversion rate. Small samples that produce positive results are among the most expensive sources of false conviction in early-stage companies.
KEY INSIGHTHow it connects
- 1.2 · The Assumption Stack — your MVP archetype is determined by which assumption sits at the top of the stack. If you have not mapped and risk-ranked your full assumption set, you cannot know which archetype to run. The MVP selector widget in this lesson assumes you have already identified the single riskiest assumption — that work lives in Lesson 1.2.
- 1.9 · The Validation Verdict — Go, Pivot, or Kill — the go-signal from Stage 1 determines what Stage 2 inherits as validated. If the validation verdict was a conditional go (problem confirmed, solution hypothesis unconfirmed), the MVP archetype is almost certainly a Wizard-of-Oz or paid pilot, not a smoke test. Lesson 1.9's output feeds directly into the assumption you select in this lesson's widget.
- 2.1 · From Problem to Solution — The Gap Between Insight and Product — the prior lesson established that the validated problem does not uniquely determine the solution. The MVP archetype you choose is partly a function of how much solution-space uncertainty remains: if the solution concept is still wide open, a concierge MVP helps you discover the right mechanism before committing to a specific product form.
- 2.3 · Scoping the MVP — Ruthless Subtraction — once you have selected the right archetype, the next challenge is deciding what goes inside it. Lesson 2.3 gives you the subtraction framework: starting from the full solution concept and removing everything that does not directly test the riskiest assumption.
GLOSSARYKey terms introduced
- Minimum Viable Product (MVP)
- The minimum set of activities — possibly involving no software — that tests whether a real customer will use and pay for a proposed value, under real conditions. Defined by the assumption it tests, not by the features it contains. The correct scope question is "what is the least we need to build to get a verdict on our riskiest assumption?" not "what features can we remove from the roadmap?"
- Concierge MVP
- An MVP where the value is delivered manually by the founding team, without automation, to a real customer who experiences the outcome. Tests workflow fit and outcome value. Does not test scalability, technical feasibility of automation, or commercial pricing. Typical build time: days.
- Wizard-of-Oz MVP
- An MVP where the customer experiences the product as if it were automated, but humans are operating the mechanism behind the interface. Tests UI/UX fit, workflow integration, and response to outputs. Does not test the performance of the automated system or scalability. Often the right format when the customer-facing behavior needs to be validated before the automated version is built.
- Smoke Test MVP
- A landing page, pre-order, or sign-up form that tests stated demand before the product exists. Fast to build, weak signal: stated intent has a known gap from actual behavior. Useful for validating whether a large enough market segment will self-identify as interested; not useful for validating whether they will actually use or pay for the product at full price.
- Paid Pilot
- A real customer paying real money for a defined, time-bounded deployment. The highest-stakes MVP format and the most definitive test of commercial viability. Tests pricing, procurement process, activation, and renewal intent. The Forge Stage 2 milestone is structured as three paid pilots at $8K each.
- Prototype vs. MVP
- A prototype tests technical feasibility (can we build it?). An MVP tests customer viability (will they use and pay for it?). A prototype is used by engineers and demo audiences with no stakes. An MVP is used by real customers doing real work. Conflating these is the technical founder's most expensive mistake — building a better prototype when the riskiest open question is a customer assumption.
- Pass signal vs. fail signal
- An MVP must specify both in advance. The fail signal is the evidence that the tested assumption was wrong. The pass signal is the evidence that it was right. Without a pre-specified pass signal, founders interpret any non-failure as a pass — including neutral results, shelfware, and single-use without return. Both signals must be defined before the experiment runs.
- Workflow integration assumption
- The assumption that customers will change their existing workflow to incorporate the product's output. For B2B products, this is often the riskiest remaining assumption after problem validation. Confirmed only by observing that customers act differently because of the product — not by observing that they use the interface.
ACTIVE RECALLCheck your understanding
Answer in your own words first, then reveal.
1A technical co-founder says "our MVP is a stripped-down version of the product with just the core detection feature, no dashboard." What question should you ask to evaluate whether this is actually an MVP or a prototype?›
Ask: Which specific assumption does this test, and will a real customer use it under real conditions with real stakes? A stripped-down product with only the detection feature may still be a prototype if it is being shown to demo audiences, used in a lab setting, or evaluated by engineers without operational stakes. The key diagnostic is: will a real customer be using this to do real work — production data, real workflow, real consequences if detection fails — and are they in a position to give you a behavioral signal (returning, escalating alerts, requesting more, refusing to renew)? If the answer is "engineers will evaluate the detection accuracy," that is a prototype. If the answer is "ops managers will run QC decisions off the alerts for four weeks," that is an MVP. The physical product may be identical; the difference is entirely in the customer context and stake level.
2Why is the Wizard-of-Oz MVP a better first step for Forge than a paid pilot, even though the paid pilot ultimately produces stronger commercial validation?›
Because the Wizard-of-Oz tests a prior assumption — workflow integration — that the paid pilot cannot fix once the pilot has started. If Forge signs a $8K paid pilot and the customer receives alerts but never acts on them, Forge has learned that the workflow integration assumption was wrong, spent $15K–$30K building a real product, and spent 8 weeks of founder time on a sales and deployment cycle. If Forge instead runs a 2-week Wizard-of-Oz MVP first and discovers alerts go unacknowledged, it can redesign the alert delivery mechanism (push notification vs. dashboard, escalation vs. informational, frequency), validate the redesign in another 1-week experiment, and only then enter the paid pilot with confirmed workflow integration. The correct sequencing principle: test assumptions in order of risk, from cheapest to most expensive, from fastest to slowest. The paid pilot is the right instrument for testing commercial viability — but only after the pre-commercial assumptions have been closed.
3What is the specific failure mode of a smoke test MVP, and why can it produce a false positive that sends a startup in the wrong direction for months?›
The specific failure mode of a smoke test is stated-intent to actual-behavior gap: people who sign up for a waitlist or make a small deposit in a low-stakes context behave differently than people who are asked to pay full price for a real product that requires real integration effort. A smoke test that generates 2,000 sign-ups at 15% conversion produces a false positive if the product is a B2B enterprise deployment requiring IT approval, change management, and a six-month procurement cycle. The sign-up population is typically composed of the most enthusiastic early adopters who are furthest from the typical buyer profile. A startup that launches into fundraising off a 2,000-person waitlist, builds the product over six months, and then discovers that waitlist-to-paying-customer conversion is 0.3% (because the real customer is an operations manager with procurement requirements, not a researcher who clicked a link) has spent a fundraising round on a smoke-test false positive. The fix: smoke tests should be followed immediately by direct sales conversations with sign-ups, testing whether stated intent converts to conditional commitment at real pricing within 48 hours of the conversation.
4Forge's Stage 2 milestone requires "first activation data collected." What makes activation data different from usage data, and why does Forge's milestone specify it?›
Usage data tells you that a customer accessed the product. Activation data tells you that a customer experienced the core value the product promises. A customer who logs into the Forge dashboard every day but never acts on a single alert is a usage metric, not an activation. A customer who receives an alert, dispatches an inspector, finds the defect, and prevents a quality escape is an activation event — they experienced the value Forge promises. Forge's Stage 2 milestone specifies activation data because usage without activation is the diagnostic for a product that has deployed but not yet changed anything. The specific activation proxy Forge uses — defect detection improvement within 30 days — is deliberately outcome-anchored, not feature-anchored: it measures whether the product produced the business result, not whether the customer learned to use the UI. This matters for the seed round: an investor evaluating Forge's Stage 2 results needs activation data to distinguish "customers are politely using a product they are not sure about" from "customers are seeing a measurable improvement that justifies renewal." Only activation data closes that distinction.