Customer development: Steve Blank's four steps, and when to move on

Customer development is Steve Blank's four-step method for building a startup around evidence from customers: customer discovery, customer validation, customer creation and company building. The first two steps search for a business model that works, and the last two scale it once you have found one. Discovery tends to get most of the attention, so this post covers the whole model and, above all, how to tell when you are ready to move from one step to the next.

Blank's own summary of the idea is that a startup should test its guesses in front of customers instead of in planning meetings and requirements documents, as he puts it in part 5 of his Customer Development Manifesto. The four steps give that testing an order.

1. Customer development came first, and lean startup grew out of it

Blank laid out the model in The Four Steps to the Epiphany, first published in 2005, according to Wikipedia's lean startup article. The same article says Blank invested in IMVU, the company Eric Ries cofounded, and had its executives audit his entrepreneurship class at UC Berkeley.

Ries then combined customer development with ideas from lean software development. He coined the term "lean startup" on his blog in September 2008 and published The Lean Startup in September 2011, per the same article. Ries's build, measure, learn loop and the minimum viable product are tools for running the experiments; customer development is the larger map that says which experiments matter at which stage.

Blank later paired the model with the business model canvas. In a 2010 post he describes each box on the canvas as a set of discovery hypotheses to test, and the canvas itself as a scorecard for tracking changes. Use the canvas as your list of guesses and customer development as the order you test them in.

2. Treat the four steps as two searches and two scale-ups

Here is how the steps fit together, using Blank's descriptions from the manifesto post and his 2015 post on the startup lifecycle:

The order matters because each step spends more money than the one before. Discovery costs founder time. Creation costs a marketing budget. Spending the budget before discovery is done is how a company ends up buying ads for a problem nobody has.

Short version: search twice, then spend.

3. Leave discovery when interviews stop surprising you

Discovery has its own post: customer discovery covers writing your beliefs down and testing them, and the interview script covers what to ask in each call. The part that matters here is the exit.

In a reply under the manifesto post, Blank says you are done with discovery when customer answers converge and further interviews add little new information. That is a test you can run: after each interview, write down which of your beliefs it changed. When three or four interviews in a row change nothing, stop and move to validation. How many customer interviews you need goes into the counting.

Here is a made-up example that runs through the rest of this post. Dana is building a scheduling tool for independent veterinary clinics with two to five vets. Her discovery interviews kept describing the same thing: the front desk loses about an hour every Monday rebuilding the week after weekend emergency calls. By her twelfth interview, the story was the same every time, and the practice owner, not the front desk, was always the one who would sign.

Beware the easy kind of convergence. If every interview agrees with you, check that you are talking to people who would actually buy, and that you are asking about what happened last week instead of what they would do someday. The Mom Test is the standard cure for polite agreement.

4. Validation means someone commits something scarce

Customer validation asks whether people will pay, and whether you can sell to them again and again in the same way. Discovery evidence is what people tell you. Validation evidence is what they give you.

Blank's bar in the manifesto post is enough early sales to show a sales process you can repeat and scale, and the number depends on what you sell. He gives three to five early customers for enterprise software and 300 to 500 for a web product. In his 2026 update to the Lean LaunchPad class, he counts a design partner as evidence of validation: a co-developer who gives you feedback, data, access to their workflow and real-world testing in exchange for early access or better terms.

In Dana's case, validation looks like four clinics paying for a three-month pilot, and a fifth that signs after a demo from her slide deck alone, without her cofounder doing the setup. The fifth one matters most, because it suggests the sale works without a heroic first visit. Market validation covers how to ask for a paid pilot instead of counting compliments.

5. When validation fails, go back to discovery and change one guess

The model has a loop: if you cannot find enough paying customers in validation, you return to discovery to learn what you missed. Blank's manifesto post expects most startups to go around that loop at least twice. He calls a change to one or more business model guesses a pivot in the 2010 post.

Change one guess at a time and say which one. Suppose Dana's pilots stall because clinic owners like the tool but will not pay monthly for it. The failed guess is the revenue model, so she goes back to discovery with that one question and leaves the customer and the problem alone. If she changes the customer, the price and the product all at once, the next round of interviews cannot tell her which change helped.

Decide your pivot-or-proceed rule before the evidence comes in. Blank's 2026 class update adds formal pivot-or-proceed criteria to discovery. A rule written in advance is much harder to argue with at 11pm before a board meeting.

6. Do not hire sales and marketing until validation passes

Blank's model keeps spending low until paying customers have validated the business model. In the manifesto post, he says the company should not build its sales, marketing and business development teams until it has proof such as a tested sales roadmap and valid purchase orders.

This is the rule founders break most, because hiring feels like progress. A sales team hired during discovery has nothing repeatable to sell, so it improvises, and the improvising hides whether the product sells on its own. Keep the founders selling until validation passes.

7. Spend on demand only after the sale works without the founders

Customer creation is where the money goes into demand: ads, events, content, partners, a sales team. Blank's 2015 lifecycle post lists signs that you are ready to leave the search stage:

• You have a sales channel that matches how customers want to buy, and you know its costs.
• Sales, or user growth through referrals and network effects, works without heroic effort from the founders.
• You understand how customers are acquired and activated, well enough to estimate acquisition cost and lifetime value for the next 18 months.

If any of those is missing, you are still in validation, whatever the pitch deck says.

How you create demand depends on the kind of market you are in. The manifesto post says market type changes how fast customers adopt and how you position the product. Dana is entering an existing market (clinics already buy scheduling software), so creation for her means showing she is better than the incumbent on one thing clinics care about. A team inventing a new category would spend this step teaching buyers the problem exists.

8. Company building starts when the team outgrows "do what it takes"

The last step turns a company organized for learning into one organized for execution, in the words of the manifesto post. In the 2015 lifecycle post, Blank puts the move from search to build at roughly 40 people and says that is when the company needs formal processes such as a sales compensation plan, expense reports and an HR manual.

You will know you are here when the informal "do what it takes" habits that worked for a small team start causing daily confusion. Blank's advice in the same post is to add CEOs who have been through this phase to your advisory board, and to consider replacing board members whose experience is only in search.

9. Who customer development does not fit

The model is for new products in uncertain markets, where the customer, the problem or the business model is still a guess. It fits a startup and a new product line inside a larger company.

It fits poorly when the customer and the business model are already known. A team adding a feature to a product with thousands of paying users should use product analytics and feature request validation instead of running four steps. A consultancy or agency that sells hours to clients it already has is not searching for a business model at all.

One newer trap: AI-generated customer personas. Blank's 2026 class has students interview AI-generated personas, then compares those answers with live interviews, and the live interviews grade the synthetic ones. Treat a synthetic interview as practice for the real one.

Your next step

Write down which of the four steps you are in, and the evidence that would move you to the next one, before you schedule another call. If you are in discovery or validation and cannot reach enough of the right people, find customers to interview lists places to look. Instant Expert can also find the specific buyers or practitioners you describe and invite each one to a short call or a written or voice answer, and you pay only if someone books or answers. For outside help with the later steps, see the directory pages for startup advisors and go-to-market consultants.