Search intent. Know what to check before a software MVP, choose the least expensive test that produces the necessary evidence, avoid false positives, and decide with criteria written in advance.
An MVP is often presented as a small version of the future product. This definition already encourage the construction. However, the first need is not always software: it is a reduction in uncertainty.
If the main uncertainty concerns acquisition, a landing page and a limited campaign may be sufficient. If it concerns delivery, a manual service learns more. If it concerns the business result, an analysis on historical data precedes the interface.
The good prototype is therefore not the most realistic. This is the least expensive one that could honestly invalidate the critical hypothesis.
1. Key figures: surviving is not enough, learning changes decisions
According to INSEE, 69 % non-micro-entrepreneur companies created in the first half of the year 2018 were still active five years later. The rate reached 70,6 % for companies and 63,3 % for classic sole proprietorships. The business was located at 63,9 %, against 73,8 % for specialized, scientific and technical activities.
This data does not say that a particular product will fail 31 %. They focus on companies, sectors and a cohort that has gone through the pandemic. They remind us that validation must distinguish legal survival, commercial traction and value creation.
A randomized experiment published in Management Science followed 116 Italian startups, with 16 measurement points over approximately one year. Both groups received training; the treated group learned to make predictions and test their hypotheses like scientists. The authors see better performance and more pivots, results consistent with better detection of false positives and false negatives.
In the pilot documented by the Innovation Growth Lab, after six months, the treated startups had activated 27 customers on average, against 4 in the control group. The size and context dictate that this report should not be generalized as a promise, but the discrepancy illustrates the interest of an explicit method.
A larger-scale replication published in 2024 groups 759 companies in four randomized trials. She observes a positive effect on stopping ideas and non-linear results on radical pivots: scientific discipline also improves the ability to give up, not just to confirm.
Finally, a field experiment on a platform sent invitations to 16 349 people. Statements about the expected future baseline influenced adoption more than announcing the current baseline; announcing a small number, current or expected, could reduce demand compared to no announcement. For a network effect product, the perceived value therefore also depends on expectations, not only on the functionalities available on day one.
2. What “validate” really means
A validation is a decision supported by proof within a scope. It does not transform a hypothesis into universal truth.
The full sentence looks like this: “For company operations managers from 50 to 250 employees recruited through this channel, 9 on 30 have accepted a paid pilot at 1 500 euros after have viewed the proposal and provided data, which exceeds our threshold of 20 %; we are continuing the delivery test. »
Segment, behavior, price, channel, size and threshold are visible. “Feedback is excellent” allows no arbitrage.
We gradually validate five objects: problem, segment, proposition, delivery and economy. A strong signal on one does not compensate for the absence of the others.
3. The register of hypotheses
Before interviews, the team writes down what they croit. Each line of the register contains:
- hypothesis and causal mechanism;
- population concerned;
- importance for the project;
- degree of uncertainty;
- observation which would support it;
- observation which would contradict it;
- test, cost and duration;
- go, pivot and stop threshold;
- result and decision.
The hypotheses are related. “Teams are wasting time” does not imply “they will buy our tool”. They may tolerate the problem, prefer a service, lack budget or refuse integration.
The mechanism is formulated as a chain: situation → friction → consequence → current solution → trigger → choice → value → payment → retention. The test targets the most fragile link.
4. The eight risks to separate
4.1. Reality of the problem
Does the problem actually occur, with what frequency and what consequence? Vague memories are supplemented with examples, documents, data and observation.
4.2. Priority
Even if real, is it important enough to displace time, budget or risk? Competing priorities matter.
4.3. Segment
Who feels the problem in a similar way, has the power to act and remains accessible through a sustainable channel? A broad persona hides differences.
4.4. Proposal
Is the promised outcome understood, credible and preferable to alternatives? The proposal is not limited to a list of functions.
4.5. Behavior
Do people move forward when action costs something: time, data, reputation, contract or money? A free compliment has little force.
4.6. Delivery
Can we produce the result with quality, deadline, security and compliance? An unserviceable sale is not validation.
4.7. Economy
Can price, margin, acquisition, support, sales cycle, churn and working capital form a viable model? The hypotheses are presented in scenarios.
4.8. Dependence
Is the project based on an uncontrolled platform, data, integration, supplier or network effect? This condition can invalidate superficial traction.
5. Build a scale of evidence
Not all observations are equal. They can be ordered from the weakest signal to the most engaging.
- general opinion;
- accurate account of past behavior;
- access to proof: invoice, export, procedure, calendar;
- time commitment: workshop, data, integration;
- reputation commitment: introduction, sponsor, conditional letter;
- contractual or financial commitment;
- repeated use;
- renewal or recommendation with verified result.
This scale does not say that you have to wait for a renewal before starting. It avoids presenting fifteen positive reactions as the equivalent of five purchases.
The same test can produce several levels. A person agrees to an interview, describes an incident, shows the affected file, then declines the driver. The problem is better supported than the business proposition.
6. Interviews that really reduce uncertainty
The interview first focuses on a past episode. “Tell about the last time” is better than “would you use?” ".
The guide explores trigger, steps, people, tools, deadline, errors, cost, consequence and current solution. He asks for exceptions. The terms used by the participant then enrich the message.
The recruiter only avoids his close network. It documents criteria, source and refusal. Friends, early innovators, and experts may overvalue curiosity relative to the accessible market.
The number of interviews does not follow a magic number. We continue until the priority profiles produce stable patterns and the disagreements are understood. However, qualitative saturation does not measure the proportion of the market; the behavioral test takes over.
After each series, the team fills out a matrix: fact, interpretation, confidence, counterexample and decision. Citations are not a substitute for counting cases when the team claims a frequency.
7. Test the request before building
7.1. Landing page
It presents a credible proposition to qualified traffic. The test measures progress: visit, understanding, action, qualification and appointment. The overall rate without source quality is misleading.
The future product is described honestly. A page can offer a waiting list, a diagnosis or a pilot; it does not simulate non-existent availability after payment.
7.2. Fake door framed
In an existing product, an entry announces an upcoming feature, then garners interest or registration. The user must understand before providing a significant commitment. The experience does not block its work and the data is minimized.
7.3. Pre-order or deposit
Money increases the strength of the signal. Conditions, refund, deadline and product status are explicit. The amount chosen must be significant enough to test a decision, without exploiting uncertainty.
7.4. Letter of Intent
In B2B, it documents sponsor, scope, conditions, schedule and sometimes budget. A non-binding letter remains less strong than a contract; its precision and the power of the signatory determine its value.
7.5. Concierge
The team manually produces the result behind a simple interface or service. It learns edge cases and value before automating.
7.6. Wizard of Oz
The user croit interagir with a system that is more automated than it is, within ethical limits. For sensitive decisions, transparency, supervision and security prohibit certain forms of simulation.
7.7. Clickable prototype
It tests understanding and journey, not willingness to pay or technical feasibility. Participants know that the simulation is not the product.
7.8. Paid driver
It brings together demand, delivery and initial economics. The scope, results, responsibilities, access, security, support, price and success criteria are written.
8. Set thresholds before seeing results
A threshold protects against retrospective narration. It comes from economics and decision-making, not from a generic benchmark.
Let’s imagine a B2B offer. On 100 targeted accounts, the company estimates it can contact 60 decision-makers, obtain 18 appointments, propose 10 drivers and sign 3. With a test cost of 12 000 euros, an initial margin of 4 000 euros per driver and an expected conversion to the 50 % subscription, the scenario remains too fragile; the driver does not cover acquisition alone.
The thresholds could be:
- go : at least 3 paying drivers, 2 sponsors provide the data on time and the delivery cost remains below 55 % the price;
- pivot : commercial interest achieved, but recurring objection on integration, segment or price unit;
- stop : less than one driver after 100 eligible accounts with controlled message and channel, or inability to produce the result without processing prohibited data.
The threshold includes guardrails. An increase in registrations accompanied by refunds, incidents or unserved leads does not trigger a go.
The sample size depends on the difference that must be distinguished. For a high investment decision, five conversions do not allow a precise estimate of a rate. The uncertainty is shown, and the next test enlarges the sample.
9. The metrics of each floor
Problem. Frequency, duration, loss, incident, circumvention and current budget.
Acquisition. Eligible audience, response, appointment, cost and time per channel.
Proposal. Understanding, objections, progression and qualification.
Activation. Time to first value, prerequisites, abort, and support.
Result. Business gain, precision, delay, error, sponsor and user satisfaction.
Retention. Repeat use, cohorts, renewal, expansion and reason for leaving.
Economy. Recognized revenue, contribution margin, cost of service, acquisition, support, cycle and collection.
Metrics are not added into an opaque score. A strong activation with a negative margin remains visible as a contradiction.
10. The MVP as a measuring instrument
An MVP is built when the code or product is the least expensive way to test the remaining hypothesis. Its scope comes from the protocol.
The main scenario is vertical: input, processing, result and proof. Administrative functions necessary for security or operation are not sacrificed because they are invisible.
Logging measures steps, errors, delays and manual interventions. The team distinguishes what the user croit automatically from what actually is. The hidden cost of manual labor is accounted for.
Sensitive data, rights, backups and deletion have a proportionate level from the pilot. “It’s only an MVP” does not justify any leaks, illegal processing, or uncontrolled automated decisions.
One long sentence summarizes the boundary: if the test can be run manually with ten clients without changing the value received, then developing for four months a multi-tenant architecture, a rules engine, an administration panel, six integrations, a mobile application and a billing system does not produce six times more proof on demand; he makes six technical bets before having resolved the commercial bet.
11. Avoid false positives
A false positive causes a weak idea to be pursued. It appears when the traffic is not representative, the free service attracts another audience, the founder sells through a personal relationship, the measurement ignores abandonments or the novelty increases usage.
The protections are: documented segment, real channel, price or behavioral cost, cohort, sufficient period, business result and comparison group when possible.
The “concierge” effect must be quantified. Exceptional delivery by founders can produce satisfaction that is impossible to replicate. The driver measures hours, interruptions and dependencies.
Stated demand is compared to behavior. “I would pay 100 euros” becomes a payment test, a deposit or a choice between options.
12. Avoid false negatives
A good idea can fail because of an incomprehensible message, a bad channel, an inconsistent price or a slow product. Stopping after a single test confuses hypothesis and execution.
The protocol diagnoses the chain. If no one visits, the proposal has not yet been tested. If people understand but do not progress, we explore priority, confidence and alternatives. If they buy but don't activate, delivery friction dominates.
A pivot changes a structuring hypothesis — segment, problem, proposition, channel, model or solution — and preserves the learnings. Changing the color of a button is not a button.
The number of pivots is neither a virtue nor a failure. Replication on 759 companies suggests a non-linear effect: the scientific method does not push you to pivot endlessly, it improves searching and stopping.
13. The decision-making committee
At the end of a cycle, the team presents the ledger without rewriting the hypotheses. It separates results, limits, interpretation and decision.
The sponsor knows the sunk costs but does not give them a vote. Time already spent does not increase the future probability of success.
Three decisions exist.
Go. The threshold has been reached; we test the next risk or we gradually increase the investment.
Pivot. The signal supports the problem or request, but a structuring assumption must change. The new register is written.
Stop. The stopping threshold or an unacceptable risk is reached. The team documents the learning, informs participants, returns or deletes data and closes cleanly.
“Keep going a little longer” is not a fourth decision. A new hypothesis and an expected amount of information are needed to justify the cycle.
14. Practical case: a commercial preparation co-pilot
One team believes that salespeople waste two hours a week preparing for meetings. Fifteen interviews show a dispersion of 20 minutes to three hours, but only major accounts document a consequence.
The segment becomes account-based teams of more than twenty salespeople. Six companies provide a sample account. The concierge service prepares files manually using authorized sources.
Four sponsors accept a paid driver. Only two connect the CRM within the deadline; the others cite safety and quality. The user result is good, but the activation fails.
The decision is neither “validated market” nor “rejected idea”. The problem, the segment and a willingness to pay are supported. Integration and governance become the critical risk. The next cycle tests a version without CRM access, then compares value and cost.
15. Frequently Asked Questions
15.1. How many interviews do you need before an MVP?
Enough to understand the motives and counterexamples of the segment, then a behavioral test. The number depends on heterogeneity and risk; there is no universal quota.
15.2. Does a waiting list validate the market?
It proves interest under a message and a channel. Qualification, payment, use and retention remain to be tested.
15.3. Should we charge the pilot?
Often yes to test engagement, particularly in B2B. A free driver may be justified for learning or access, but its commercial signal is weaker and should be named as such.
15.4. When to write code?
When the behavior of the system, its feasibility or its ability to deliver value constitutes the priority uncertainty and a less expensive test is no longer sufficient.
15.5. How to set a threshold without historical data?
Start with the minimum saving and investment decision, then use a range. The first cycle calibrates; it should not be presented as an accurate estimate of the entire market.
15.6. Does quitting mean the idea was bad?
No. It may be inappropriate for this segment, this moment, this channel or these constraints. Explicit stopping preserves some capital and can reveal a better direction.
16. What Logiks recommends
First write down the risk that could kill the project, the behavior that would test it and the decision thresholds. Look for real commitment before heavy construction, then measure delivery and economics with equal rigor. An MVP is not a small promise: it is an instrument of proof designed to make the next decision less uncertain.
17. Main sources
- Insee, Companies created in 2018: 69 % are still active five years after their creation, 2025 : https://www.insee.fr/fr/statistiques/8634190
- Camuffo et al., A scientific approach to entrepreneurial decision making, Management Science : https://doi.org/10.1287/mnsc.2018.3249
- Bocconi University, archive and summary of the randomized trial: https://iris.unibocconi.it/handle/11565/4013977
- Innovation Growth Lab, A scientific approach to entrepreneurial experimentation : https://www.innovationgrowthlab.org/projects/a-scientific-approach-to-entrepreneurial-experimentation-evidence-from-a-randomised-controlled-trial
- Camuffo et al., replication of four trials and 759 companies, Strategic Management Journal, 2024 : https://doi.org/10.1002/smj.3580
- NBER, Promoting Platform Takeoff and Self-Fulfilling Expectations, experience on 16 349 invitations: https://www.nber.org/papers/w28325
- OECD, Cross-country evidence on start-up dynamics : https://www.oecd.org/en/publications/cross-country-evidence-on-start-up-dynamics_5jrxtkb9mxtb-en.html