The Four Steps to the Epiphany

Webvan built the company its forecast described

In May 1999, Webvan let about 1,100 people test an online grocery service. One month later it opened its first market; two months after that it filed for an initial public offering. Its first automated distribution center had cost about $40 million, and before meaningful demand evidence arrived, the company signed a deal with Bechtel for as much as $1 billion to build more. Six months after launch, Webvan was receiving roughly 2,500 orders a day against the 8,000 its plan required. The machinery worked, but far below the volume for which it had been designed.¹

Those figures and the causal interpretation come from Steve Blank's opening Webvan case in The Four Steps to the Epiphany. Blank's point is deliberately unsettling: Webvan did not simply fail to execute. Its managers, board, investors, engineers, marketers, and operations teams executed a product-development plan whose customer assumptions had never faced a comparable stop-or-revise gate. Warehouses, hiring, promotion, and national expansion converted a forecast into commitments before customer behavior could contradict it cheaply. ²

The collapse has explanations beyond Blank's framework, including dot-com-era capital markets, grocery margins, acquisition, and operating choices. But its human cost was not metaphorical. When Webvan shut down in July 2001, contemporary reporting counted about 2,000 workers losing their jobs and 750,000 active customers losing the service. Blank's animating question is therefore larger than “how can a startup launch?” It is: how can people discover whether there is a business before they organize many other people's work and expectations as though the answer were already known?³

A refused sale became evidence at E.piphany

Blank developed his answer from a career in Silicon Valley startups and from a course he began teaching after the dot-com collapse. He describes the book as his class text and his company stories as the experiences from which he derived the method. They are participant recollections, not a comparative study of startup survival.⁴

One recollection shows the change in practice better than the four-step diagram. While starting E.piphany, Blank visited Silicon Graphics marketing vice president Joe DiNucci. Their conversation seemed to confirm E.piphany's proposed marketing-analysis product, so Blank asked whether SGI would become an early customer. DiNucci said no: SGI had already built and deployed its own system, called “Mine Your Own Business.”⁵

The sales call had failed. On the drive home, however, Blank reinterpreted what he had heard. SGI had found the problem painful enough to write working software when no vendor supplied it. Blank returned to see the program, recruited its creator John McCaskey as an E.piphany cofounder, and, in Blank's retrospective account, licensed the code from SGI for one dollar; DiNucci later joined E.piphany as vice president of sales. A “no” had distinguished lack of a customer from lack of a problem and changed the product, coalition, and evidence available to the startup. ⁶

This episode also reveals a boundary. E.piphany's later performance cannot prove that this encounter, or the method Blank later named, caused its success. Retrospective stories select memorable turns from many concurrent decisions. The case is most useful as a concrete account of what became visible when a founder treated a prospective customer's existing behavior—not enthusiasm for a pitch—as evidence.⁷

The four steps keep search from masquerading as execution

Blank calls the missing operating system customer development. Product development can proceed while a second, recursive process tests who has the problem, what they do about it now, who can buy, how a sale happens, and whether the economics can repeat. The book's four steps name changes in the kind of evidence and organization required:

• Customer discovery turns claims about problem, customer, product, channel, and price into hypotheses confronted outside the building. A homegrown remedy like SGI's can be stronger evidence of pain than a favorable interview.
• Customer validation asks whether actual purchases and a sales path recur. A bespoke deal or a few enthusiasts may establish interest without establishing a scalable model.
• Customer creation increases demand according to the market being entered; it is delayed so promotion does not conceal weak retention or economics.
• Company building adds departments, managers, and execution measures only after the organization has something evidenced to repeat.

⁸

The circles and backward arrows matter more than the numbering. In a product schedule, returning to an earlier phase looks like delay. In customer development, failed validation sends the venture back to discovery while its commitments are still small enough to change. The method's sharpest distinction is consequently search versus execution: a startup searches for a repeatable and scalable business model; an established company executes one it already understands.⁹

That distinction connects the book to innovation, entrepreneurship, and renewal. It also differs from The Innovator's Dilemma: Christensen explains why incumbent allocation systems reject emerging markets, whereas Blank proposes a process for a new venture that does not yet know its market. Innovation and Entrepreneurship offers an earlier contrast: Drucker systematizes the search for changes that can become opportunities, while Blank moves discovery into repeated contact with a specific market. That is a useful reading sequence, not evidence that Drucker's book directly shaped Blank's method.¹⁰

The evidence supports a discipline, not a guarantee

Four Steps was first published as a practical class text, and its examples carry the strengths and limits of that origin. It makes experienced tacit knowledge inspectable and gives teams shared gates for spending. It does not estimate the effect of completing each step, supply a representative sample, or show that its four-part sequence outperforms alternatives across industries. The Wiley reissue also preserves the earlier text rather than updating it for new venture forms.¹¹

Later research tests related propositions without validating the whole book. A randomized trial of 116 Italian startups found that teams trained to articulate theories and test predictions made more precise predictions and pivoted more often than a control group that received other entrepreneurship training. That supports disciplined hypothesis testing under the study's conditions; it is not evidence that customer interviews are infallible, that every pivot is wise, or that Blank's exact sequence travels unchanged into public agencies, science, cooperatives, or capital-intensive work.¹²

The lineage is nevertheless direct. Eric Ries adapted Blank's customer development while advising and building IMVU, then joined it to agile software practice and ideas borrowed from lean production in The Lean Startup. Build–Measure–Learn and innovation accounting are later formulations, not terms to read backward into Blank's 2005 text.¹³

A cheap test for the company can still be costly for others

Webvan makes one experimental burden obvious: scaling a belief enrolls workers, suppliers, investors, and customers in its failure. Customer development tries to reduce that exposure by postponing irreversible commitments. But its criterion of success remains a profitable, repeatable, scalable business. It does not by itself establish that the business is safe, truthful, ecologically viable, or beneficial to those who cannot purchase or refuse it. ¹⁴

The word customer can obscure several parties. An enterprise buyer may authorize a purchase without being the employee whose behavior is monitored. A platform user may supply attention and personal data while an advertiser pays. A community or ecosystem may bear consequences without appearing in a sales road map. The book teaches founders to distinguish users, influencers, and economic buyers for the sake of a sale; responsible experimentation must also identify who is exposed to the test and who has authority to consent. ¹⁵

Research on the ethics of online controlled experiments adds questions that the commercial method does not answer: autonomy, fairness, non-maleficence, beneficence, protection of vulnerable participants, and review proportional to risk. Those are later ethical criteria, not claims Blank makes. They clarify why the smallest experiment that can test demand may still be too large, opaque, or irreversible for the beings who bear its effects. ¹⁶

Read Four Steps for the moment when confident execution should yield to inspectable search. Then ask what Webvan's spreadsheet and the E.piphany sales call each made visible, what evidence could stop the next commitment, and whose cost must be counted before “going backward” can honestly be called cheap.

The links to organizational intelligence and benefit for all life are editorial reading paths. The first tests whether search improves collective sensing and stopping; the second widens success beyond a paying market to everyone exposed to the venture. Neither relation is attributed to Blank.¹⁷