Startups

Validating a startup idea before you spend: problem, customer, MVP and metrics

By Martín Díaz Cuadrado · 7 September 2026 · 9 min read

The most expensive way to find out an idea does not work is to build all of it. Validation means reducing the project's most dangerous uncertainty, at the lowest possible cost, before you commit money, months of work or a company with partners. It is not a formality and it does not guarantee success: it is a method for being wrong cheaply.

What validation means (and what it does not)

Validation is not asking your friends whether they like the idea, or posting a survey and counting the yeses. Nor is it launching a first version and waiting to see what happens. Validation means writing concrete hypotheses, designing tests capable of proving them false, and deciding in advance which result would make you continue, change course or stop.

Almost every idea rests on four linked hypotheses, and it pays to test them in this order:

1. Problem: there is a real, frequent or painful problem for a specific group of people or companies.
2. Customer: you know who suffers it most, how they solve it today and how to reach them.
3. Solution: your proposal solves it better than the current alternative, enough to be worth changing habits.
4. Business: someone will pay enough, and acquiring and keeping each customer costs less than they bring in.

The classic mistake is skipping steps: building the solution before confirming that the problem exists and matters to someone.

1. Write the problem as a hypothesis that can fail

A useful hypothesis is specific and measurable. Compare these two:

• Weak: "Small businesses struggle with admin".
• Useful: "Renovation companies with one to five employees spend several hours a week preparing quotes and consider it one of their three biggest admin problems".

The second tells you who to ask, what to look for and which answer would disprove it. Also write down the assumptions you are making without noticing: that the person using the product is the one who decides to buy it, that they have a budget for it, that you can reach them at a reasonable cost.

A practical exercise: list all your hypotheses and score each on two criteria, how much evidence you have and how much damage it would do to the project if it turned out to be false. Start testing the ones that combine little evidence with high impact.

2. Pick a specific customer, not a market

"Everyone who owns a car" is not a customer. To validate, you need a segment narrow enough that you can find dozens of similar people, talk to them and recognise patterns. Define:

• who uses the solution and who decides and pays, which in business-to-business sales are often different people;
• in what situation the problem arises and how often;
• how they solve it today: with another tool, with a spreadsheet, by hiring someone or simply putting up with it;
• where they gather, what they read and who they listen to, because that will be your acquisition channel.

Look especially for people who have already tried to solve it themselves. Someone who has cobbled together a workaround with spreadsheets and emails is showing you that the problem matters enough to spend time on it.

3. Problem interviews: ask about the past, not the future

Interviews are the cheapest tool to start with, but they only work if you stop people telling you what you want to hear. Almost everyone is kind about someone else's idea; very few pay for it. Some rules that help:

• Do not pitch your idea at the start. Talk about the problem and their day-to-day work or business.
• Ask about past events. "When was the last time this happened? What did you do? How much time or money did it cost you?" Hypothetical answers such as "yes, I would use it" are worth very little.
• Find out what they already do and already pay for. If they have never looked for a solution or spent anything on it, perhaps the problem does not hurt that much.
• Listen more than you talk and take verbatim notes. Interpretation comes later.
• End by asking for something: another conversation, an introduction to a colleague or a prototype trial. What someone is willing to give is a more reliable signal than what they think.

There is no magic number of interviews. A practical rule of thumb is to keep going until the answers stop surprising you and the same patterns repeat among people who do not know each other. If after many conversations each person describes a different problem, your segment is probably too broad.

If you store data about the people you interview or about a waiting list, data protection law applies: explain what you will use the data for and ask for consent where required.

4. Measure commitment, not opinions

There is a huge gap between "that sounds interesting" and "here is my card". Rank the signals from least to most valuable:

1. A favourable opinion in a conversation.
2. Time: agreeing to a second meeting, a demo or a prototype trial.
3. Reputation: introducing you to their manager or to other potential customers, or recommending you publicly.
4. Data or work: sharing real information about their business or spending hours setting up a trial.
5. Money: a pre-sale, a deposit, a letter of intent with a price or a paid pilot.

Design your tests to climb that ladder as quickly as possible. An email waiting list is a weak signal; a list of people who have left a refundable deposit is much stronger.

5. The MVP: the smallest experiment that answers your question

Minimum viable product does not mean "a cut-down version 1.0". It is the cheapest experiment capable of confirming or disproving the hypothesis you are testing at that moment. Depending on the question, it can take very different forms:

• Landing page with a call to action. You describe the offer and the price, and measure how many visitors from the right segment sign up, reserve or request a demo. Good for testing message and price, but it requires attracting qualified traffic.
• Concierge MVP. You deliver the service by hand, one by one, to the first customers. It does not scale, but it shows you exactly what they value and which steps are unnecessary.
• "Wizard of Oz". The customer sees an interface that looks automated, but people are doing the work behind it. You validate the experience before investing in technology. If you handle sensitive data, be transparent about what happens behind the scenes.
• No-code prototype. Forms, spreadsheets, automations and page builders let you set up real workflows in a few days.
• Paid pilot. In business-to-business sales, an agreement limited in time and scope, with a price and success criteria in writing, is worth more than any survey.

Before launching each experiment, write down three things: the hypothesis it tests, the metric you will look at and the threshold that will decide the outcome. Setting the threshold after seeing the data is the easiest way to fool yourself.

If you charge for pre-sales or pilots, you are already carrying on a business activity and must register and invoice properly. I go through the options in the guide to starting a business in Spain.

6. The metrics that matter (and the ones that distract)

Visits, followers, downloads or sign-ups are easy to inflate and say little about whether there is a business. They are vanity metrics. The ones that genuinely help you decide usually fall into five families:

Activation

The percentage of sign-ups who reach the moment when they get the promised value: they send their first quote, complete their first booking or receive their first report. Define that moment precisely. If many people sign up and hardly anyone activates, the problem lies in the offer or in the first-use experience.

Cohort retention

Group users by the week or month they started and measure how many are still using the product over time. If the curve falls to zero, there is no fit; if it flattens at some level, there is a core of users for whom the product is useful. For products used repeatedly, it is probably the most honest signal.

Conversion to paid and willingness to pay

How many move from trying to paying, and at what price. Test real prices early: giving the product away for months tells you whether people like it when it is free, not whether it is a business.

Customer acquisition cost (CAC)

Everything you spend on marketing and sales over a period, divided by the new customers you win in that same period. During validation it is only indicative, but it warns you if the only channel that works is too expensive.

Customer lifetime value (LTV) and margin

The gross margin a customer generates over the whole time they stay with you. The relationship between what it costs to win a customer and what they bring in determines whether growth generates cash or burns it. With only a few months of data it is an estimate, and should be treated as one.

Choose one primary metric for each stage and, at most, two or three supporting ones. If you measure everything, you decide nothing.

7. Decide: persevere, pivot or stop

At the end of each cycle of experiments, compare the result with the threshold you set. There are three reasonable outcomes:

• Persevere: the hypothesis holds and you move on to the next one on the list.
• Pivot: one part works and another does not. You change a single element (segment, problem, channel or pricing model) and test again.
• Stop: after several honest attempts, no signals of commitment appear. Stopping in time is not failure; it is exactly what validation is for.

Put a time and money limit on the validation phase before you start. Without that limit, any ambiguous data point becomes a reason to carry on for another month.

A six-week plan as a starting point

This is a template, not a recipe. Adjust the timings to your sector and availability:

1. Week 1: prioritised list of hypotheses, segment definition and interview script.
2. Weeks 2 and 3: problem interviews. At the end, rewrite the hypotheses based on what you learned.
3. Week 4: a commitment test, such as a page with a deposit, a pilot proposal or a letter of intent.
4. Week 5: concierge MVP or no-code prototype with the first committed users.
5. Week 6: review metrics against thresholds and write down the decision.

Mistakes to avoid

• Falling in love with the solution and only looking for data that confirms it.
• Asking only people you know, who tend to be kind and are rarely your real customer.
• Mistaking interest for demand: lots of likes and no deposits.
• Changing several things at once and not knowing what caused the improvement or the decline.
• Building technology for a problem you have not yet seen anyone pay to solve.

When to start spending for real

It makes sense to invest in development, incorporate the company or look for funding once you have evidence of three things: that a specific segment suffers the problem, that your solution generates real commitment (ideally, money) and that at least one channel for reaching those customers works at an affordable cost. That evidence is also the first thing any serious investor will ask about. In the article on investing in startups I explain how the people putting in the money look at it.