Startups as a science: the five steps

This is a step-by-step process for launching a business without relying on gut feelings. It grounds each decision in a hypothesis, a test and data, so you can find product-market fit with confidence before you commit real money or time to building.

Step 1: Start with a hypothesis

Every business starts with one core idea. That's your starting point.

1. Observe. Look at an existing trend, behavior or system in the world.
2. Find the flaw. Figure out how you could make it fundamentally better, more efficient or more enjoyable.
3. Find the value. Work out how you could get people to pay you for making it better. What tangible value are you creating?

That first idea is Hypothesis 1.

How to form a good hypothesis. A strong hypothesis isn't just an idea. It's a statement that can be proven wrong. "People want better nightlife" is a bad hypothesis. "People who organize social outings for their friends will pay a monthly fee for a tool that guarantees access and makes hosting easier" is a good one, because it gives you something concrete to test.

Step 2: Choose your path

The time and money you have decide your next move.

Path A: you have money but not time. Launch a minimum viable product based on Hypothesis 1 right away. This path is risky. The hard part isn't building the product. It's your own bias. Early customer feedback is noisy and can mislead you, and founders are often too busy building to listen. It takes discipline. Talk to customers every day, and be ready to change course or cut features based on what they tell you, not on your assumptions. If you can keep that up, continue with the next steps using that feedback.

Path B: you have time but not money. This is the path I recommend. Don't build anything yet. Use your time to reduce the risk by writing four more hypotheses. You should end up with five competing ones. Each is a clear variation of the original idea, with its own business model, features, value proposition and target person.

If Hypothesis 1 is a VIP pass for nightlife, the others could be:

• Hypothesis 2: a cheaper, solo-only VIP pass.
• Hypothesis 3: a pass focused on discovering new events, not just getting in.
• Hypothesis 4: a platform for venues to run their own memberships.
• Hypothesis 5: a pass focused on daytime experiences like restaurants and cafes.

AI is good at brainstorming variations like these quickly.

Step 3: Build a question bank

Now you design the test for your five hypotheses.

1. Write a question bank. For each hypothesis, write 10 targeted questions. That's 50 in all.
2. Mix the formats. Use forced-choice, multiple-choice and ranking questions to get richer data.
3. Design each set for yes. Write each set of 10 so that the person the hypothesis is aimed at would naturally choose the answers that support it. For the solo pass, a key question might be: "Which would you prefer: a cheaper pass that only works for you, or a slightly more expensive one that lets you bring friends?" The person the solo pass is for should, in theory, pick the cheaper one.

Write questions that make people choose. The best questions force a trade-off. Avoid "Do you want this?" because the answer is always yes. Ask "Which of these five features is most valuable to you? Split 100 points among them." That makes people show what they truly value, not just what sounds nice. Ranking questions work for the same reason.

Step 4: Run the survey

This is where your research becomes one tool.

1. Pick the key questions. From the 50, choose the two most important for each hypothesis. These should be the clearest, highest-signal questions, the ones that point directly to someone wanting that version.
2. Build the final survey. You now have a short survey of 10 questions.
3. Send it out and collect answers. Send it to a mix of the people your five hypotheses are aimed at.

Find the right audience. Don't just survey your friends. They'll lie to protect your feelings. Use Facebook or Reddit to find niche communities that match the people you're aiming at. For a small budget, you can run targeted ads to a page with the survey. The quality of your data depends on the quality of your audience.

Step 5: Read the results

Once you have a statistically significant number of responses, the data gives you a direction. You're no longer guessing. The answers give you a clear mandate for:

• the business model that wins,
• the core set of features,
• the customer you're building for.

You've gone from a collection of ideas to one tested plan, ready to execute.