Eric Ries

Failure exposed the difference between delivery and discovery

Eric Ries was born in 1978. He graduated from Yale in 2000 with a B.S. in computer science, worked as a senior software engineer at There.com, and co-founded IMVU with Will Harvey in June 2004, serving as its chief technology officer.¹²

There had spent five years in development and raised $40 million before its October 2003 launch. The Stanford teaching case reports only $20,000 of revenue in the first month, flat sales for the next six months, and a turn toward military simulation. A later independent profile reports the investment as a rounded $50 million; the more specific case figure is used here rather than treating the estimates as interchangeable.³

Ries made that experience into a management problem: a team can deliver software and still fail to discover whether its business premise is true. The failure did not prove the method he later proposed. It supplied a participant's diagnosis, reconstructed after the outcome and without a comparison venture that differed only in its learning process.³

IMVU joined several lineages

Steve Blank agreed to invest in IMVU on the condition that Harvey and Ries audit his customer-development course at UC Berkeley. Blank's later first-person account and Ries's published introduction agree on the relationship: customer discovery needed a discipline comparable to the one engineering already had. An independent profile also records the condition and Ries's subsequent use of Blank's work.⁴

Ries already practiced agile software development. He combined its short development cycles with customer development, then interpreted waste and small batches through lean manufacturing. Ries explicitly traced that last influence to the Toyota Production System. The lineage is selective translation, not direct equivalence: Toyota organized the production of known physical goods, whereas IMVU was searching for a product, customer, and business model.⁵

The early operating practice was concrete. A six-person IMVU team released a paid beta in November 2004 while it was still buggy enough to crash some users' computers. It spent about five dollars per day for one hundred advertising clicks, followed daily cohorts through registration, download, chat, repeat use, and purchase, and split-tested changes before wider release.⁶

That evidence describes what IMVU did. The teaching case depends heavily on company records and participant interviews, while Ries's 2008 account is first-person testimony. Neither source isolates which practice caused IMVU's later trajectory.⁶

Progress became a tested change in belief

The Lean Startup, published on September 13, 2011, defines a startup as a human institution creating something new under extreme uncertainty. Ries names evidence that advances that search validated learning and organizes it into Build–Measure–Learn: turn an assumption into a product intervention, observe behavior, and decide whether to persevere or change strategy.⁷

A minimum viable product is minimal relative to the learning required. It may be less than a finished product, but it is not valuable merely because it is cheap or fast. If the artifact cannot expose the relevant assumption—or if defects and deception change the behavior being measured—its result does not support the inference the team wants to make.⁷

Innovation accounting asks teams to establish a baseline, compare cohorts or controlled changes, and distinguish actionable evidence from cumulative totals that can rise without showing improvement. A pivot changes an important part of strategy while retaining a larger vision; perseverance keeps the current hypothesis. The method structures attention and review, but does not make either judgment automatic.⁷⁶

Measurement defines the edge of what can be learned. IMVU's funnel could show where observed users stopped or paid. It could not by itself explain why a change worked, represent people who never entered the funnel, discover an unimagined form of value, or determine whether more use benefited the user. Those are limits of the measurement design, not evidence that cohort analysis has no value.⁶

Independent tests support parts under conditions

Later research supplies evidence beyond the IMVU story. Harms and Schwery surveyed 100 young Berlin software ventures and found a moderately strong, positive association between a five-part lean-startup capability measure and reported project performance. The result is useful but observational: the sample was limited to responding software ventures, and common self-report measures do not establish that the method caused performance.⁸

Leatherbee and Katila followed 152 National Science Foundation I-Corps teams in a standardized implementation. Hypothesis formulation, probing, and convergence linked as the method predicted, while MBA-trained team members were less likely to use the process even though they valued it afterward. The study supports a learning mechanism and identifies team composition as a boundary; it is not a randomized test of the entire framework.⁸

A randomized trial placed 116 Italian startups in ten-session programs. Both groups learned market feedback and feasibility assessment, but only the treated group learned to articulate theories, derive predictions, and test them rigorously. Treated firms performed better and pivoted more often without dropping out more often. A 2024 replication and extension combined that trial with three more, covering 759 firms. It found that the approach could be taught, increased idea termination, and produced a nonlinear pattern in which treated firms made a few radical pivots rather than none or repeated pivots.⁹

These experiments are the strongest causal evidence in the set, but their treatment is a theory-based scientific approach, not a bundled test of Ries's book. Together the studies support explicit hypotheses, disciplined probing, and willingness to stop or redirect an idea in particular programs. They do not establish that every minimum viable product, metric, or fast release improves venture performance, nor that the method transfers unchanged across sectors. ⁹

Critique begins before the experiment

Felin, Gambardella, Stern, and Zenger argue that Lean Startup says too little about where a consequential hypothesis comes from. Readily observable customer feedback may direct search toward what is easiest to validate and encourage incremental change rather than a novel theory of value. Bocken and Snihur reply that the method was not designed for ideation; once a venture has a vision and an initial model, iterative experiments can reduce uncertainty, engage stakeholders, and allow novelty to emerge.¹⁰

This is a conceptual dispute, not an outcome comparison between ventures using the two positions. It identifies a real sequence problem: evidence can reject or refine a stated hypothesis, but it cannot guarantee that the hypothesis was important, imaginative, or directed toward a worthwhile purpose. A rapid loop around a weak question remains a weak inquiry.¹⁰

Customers were participants, not only observations

The IMVU case makes the distribution of experimental cost visible. Roughly half of early subscribers were eighteen or younger. In one seven-day funnel, 2.7 percent of active adults converted to payment and no teenagers did; some young users mailed cash. After IMVU enabled charges through mobile-phone bills in May 2005, monthly revenue rose from less than $700 to $1,400 and then $2,200 in June. The founders attributed the increase to easier payment, but the sequence alone does not prove that cause.¹¹

For the company, the release limited cash committed before learning. For users, the same experiment involved crashes, attention, behavioral observation, and money. The case says the founders accepted temporary user frustration, but it does not report informed-consent procedures, child-safety review, or independent oversight. Absence from the case is an evidence gap, not proof that no protective practice existed.¹¹

Later ethics research on online controlled experiments treats autonomy, non-maleficence, beneficence, fairness, accountability, and context-sensitive review as separate from business impact. It also argues that experiments above minimal risk may warrant review structures and that transferring a result requires checking whether relevant conditions still hold. This is a later normative framework, not evidence that Ries or IMVU used it.¹²

The distinction matters most where outcomes are delayed, exposure cannot be reversed, or the people bearing risk cannot meaningfully refuse. Medicine, credit, employment, education, public benefits, infrastructure, and ecological intervention require domain-specific safety, rights, and public-accountability constraints. A test can produce commercially actionable evidence and still be impermissible or harmful.¹²

Transfer to institutions remains a separate claim

The Startup Way, published in 2017, extends entrepreneurial management to established companies, nonprofits, nongovernmental organizations, and governments. Its publisher describes the argument as grounded in five years of Ries's work with such organizations. That is an official record of his claim and experience base, not an independent cross-sector performance evaluation. ¹³

Independent corporate evidence is narrower. A multiple-case study examined two software innovation projects in two large companies through seven employee interviews. It identified top-management support and cross-functional teams as enablers and found different inhibitors depending on how a project began, its objective, and its product. The study helps explain implementation but cannot validate the method for all large organizations, let alone public institutions. ¹³

An internal venture's authority, career incentives, budget cycle, and access to customers can differ sharply from an independent startup's. A public institution also owes duties to residents who are not reducible to customers or conversion rates. No net conclusion about benefits or harms to users, workers, communities, or ecosystems follows from the venture-performance studies above; their outcome measures were not designed to provide that account. ⁸⁹¹²

Connections and types

The Lean Startup is Ries's documented work and the subject most directly associated with him. The Four Steps to the Epiphany is a documented intellectual influence. Toyota and the Toyota Production System are a partial manufacturing lineage; the connection does not claim that Ries worked at Toyota or reproduced its production system in software.⁴⁵

Organizational intelligence and benefit for all life are editorial frames for asking what the method helps an institution learn and whose welfare bounds that learning. The strongest topical connections are learning, quality, and reliability and innovation, entrepreneurship, and renewal. Decision and bounded rationality, measurement and control, strategy and adaptation, and executive attention and sensing are analytical comparisons. The emphasis scores are editorial coordinates, not historical measurements, causal estimates, or claims that Ries used those labels.

Ries's durable question is what a team learned that should change its decision. The necessary companion questions are who bore the test's cost, who could refuse or contest it, and what evidence would show that learning faster served a purpose worth pursuing.