Building without demand.

On 1 September 2017, Juicero stopped selling its connected press and produce packs. The machine had launched at $699, later fell to $399, and depended on single-serve packs priced at roughly $5 to $7. The company had raised about $120 million. Its shutdown note said the existing prices could not carry the mission at the scale it wanted.^[3]

The official story became a joke about overengineering. Bloomberg reporters had shown that the packs could be squeezed by hand. The joke was accurate but incomplete. A hand did not beat a machine on engineering. It exposed that the machine had been assigned to a weak job. The company had built a controlled supply system, cold chain, scanner and powerful press before proving that enough households considered fresh juice inconvenient enough to adopt another appliance and subscription.

This is what “no market need” looks like inside a competent company. It rarely feels like nobody wants anything. Early users are curious. Reviewers praise the craft. Investors can see a large adjacent category. Friends understand the pitch. A handful of customers may love the product. The failure is narrower and more lethal: too few people care enough, often enough, at a price the company can survive.

The distinction matters because the usual remedy—improve the product—can deepen the loss. Better engineering increases commitment to a problem whose rank has not changed. More features give current enthusiasts more reasons to praise the product without creating the next cohort. Marketing buys a larger sample of the same weak behavior.

The root decision occurs before these symptoms. The company funds supply before observing sacrifice. It asks customers to predict what they will do, accepts their answer and turns the prediction into a roadmap. Once staff, capital and reputation attach to the roadmap, disconfirming evidence becomes expensive to accept.

Boundary inspection / Want, fit and scale are different findings

The demand question comes first because it identifies whether a ranked problem exists before the company argues about the quality of its answer. Report 004 begins one step later: a real demand pool exists, but the chosen product, segment or delivery model does not create a repeatable relationship. The distinction changes the intervention. Weak underlying demand requires reopening the problem. Weak fit requires changing the pairing.

The test is counterfactual. Remove the current product from the conversation and ask what the customer did the last three times the problem appeared. A demand-positive record contains an incident, a consequence, an owner and a costly response. A demand-negative record contains adjectives: interesting, useful, easier, delightful. Those words may justify further search. They do not justify a scaled organization.

Capital can blur the boundary. Funding allows the company to subsidize price, provide founder service and buy repeated exposure. That can manufacture activity around an unranked problem. The demand review therefore removes investor belief and company effort from the evidence set before it asks what the customer contributes.

01 / The courteous interview

DIAGNOSIS: The team asks whether the idea is useful, not how the customer handled the last occurrence of the problem. The interview measures manners and imagination.

LOAD TEST: For the last ten interviews, mark every claim that describes past behavior and every claim that predicts future behavior. If the case rests on the second column, demand remains untested.

INTERVENTION: Re-run interviews around the last incident: trigger, consequence, workaround, owner, money spent, delay accepted and reason the customer has not already changed.

02 / The waitlist without a cost

DIAGNOSIS: An email address is treated as a purchase. The customer can join without changing a calendar, budget or workflow.

LOAD TEST: Ask waitlisted users to book a dated onboarding slot, import data, invite a colleague, sign a conditional order or place a refundable deposit.

INTERVENTION: Add one reversible but meaningful cost. Measure completion, not clicks.

03 / The investor as customer proxy

DIAGNOSIS: A fund’s belief in a category is converted into evidence that customers want this product.

LOAD TEST: Remove financing, press and advisor logos from the evidence memo. What customer behavior remains?

INTERVENTION: Keep the investment case and demand case as separate documents with different owners.

04 / The founder’s private pain

DIAGNOSIS: The founder truly has the problem, but assumes frequency, budget and urgency transfer to a wider market.

LOAD TEST: Find ten people who share the behavior but not the founder’s job, network or identity. Compare their current spend and frequency.

INTERVENTION: Treat founder insight as a search advantage, not a sample.

05 / The novelty cohort

DIAGNOSIS: Early adopters tolerate friction because being early is part of the reward. Their behavior is projected onto ordinary buyers.

LOAD TEST: Separate users acquired through product communities, crowdfunding and press from users acquired through the eventual channel.

INTERVENTION: Require one cohort from a boring channel to repeat the behavior.

06 / The subsidized yes

DIAGNOSIS: Discounts, concierge labor or free hardware make trial rational even when the underlying problem is weak.

LOAD TEST: Show retention and contribution margin after removing launch incentives and founder service.

INTERVENTION: Raise the price or remove the labor for a small cohort before expanding volume.

07 / The broad noun

DIAGNOSIS: “Parents,” “creators,” “small businesses” or “people who care about health” hides incompatible problems and budgets.

LOAD TEST: Require one role, one trigger and one existing workaround in the segment definition.

INTERVENTION: Narrow until users can refer one another without explaining why they are alike.

08 / The roadmap defense

DIAGNOSIS: Weak use is attributed to missing features indefinitely. Every failed release produces another release.

LOAD TEST: Before shipping, record the behavior the feature must change and the date on which the claim expires.

INTERVENTION: If three material releases fail to move the named behavior in the same segment, reopen the problem—not the backlog.

Demand evidence has an order. The company should not be able to promote itself from one level to the next with prose.

The trigger is simple: do not add fixed cost when the company has only D-01 or D-02 evidence. A prototype is allowed. A launch is allowed. A team built to service projected demand is not.

Trend matters more than a universal threshold. A small number of repeated, costly acts from one coherent segment can justify another experiment. A large number of cheap acts from unrelated people cannot. The ledger exists to prevent volume from laundering weakness.

Reading the ledger / Escalate evidence in order

Each row of the ledger must describe the same segment, trigger and price. Ten waitlist sign-ups from one audience cannot be combined with three paid pilots from another and a retention result from a third. Aggregation makes a weak market appear complete by assigning a different cohort to every missing proof.

The sequence is deliberately asymmetric. A company may move from interviews to a small trial with limited evidence because the cost remains reversible. It may not move from trial to broad hiring merely because one account succeeded. The next commitment must be earned by a more expensive customer behavior. When the company pays all of the cost of producing the evidence, the evidence is about company effort—not market pull.

CF-01 / JUICERO — A complete system around a low-ranking job

• About $120 million raised.
• Press launched at $699 and later sold for $399.
• Produce packs cost roughly $5 to $7 per serving.
• The company stopped sales in September 2017.^[3]

Juicero’s hardware, pack logistics and food-safety controls solved real operating problems. The company’s shutdown does not prove that people do not buy premium juice. They do. It shows that the complete proposition asked for too many commitments at once: counter space, appliance cost, geographic availability and an ongoing pack subscription. Hand-squeezing mattered because it removed the capital-intensive part of the proposition without removing the outcome.

TRANSFERABLE LESSON: Test the minimum customer sacrifice required to obtain the outcome before engineering the maximum system capable of delivering it.

CF-02 / COLOR — Capital and curiosity before a stable job

• Color launched its proximity-based photo app in March 2011.
• The company had raised $41 million before launch.
• The first product drew confused reviews and was reworked.
• By 2012 the consumer app was being withdrawn as the company searched for another direction.^[5]

Color’s public record does not reveal a clean retention table, so the diagnosis must remain bounded. What is visible is a large commitment made before a stable user job emerged. The product asked people to understand a new social behavior—sharing photos with nearby strangers—before an existing network or repeated use case made that behavior legible. Capital bought time to retool. It did not make the original job more important.

TRANSFERABLE LESSON: A large round can finance a search. It cannot count as evidence that the search has ended.

CF-03 / JIBO — Affection without a durable household role

• Jibo began as a crowdfunded social robot for the home.
• The company raised more than $70 million according to IEEE Spectrum.
• Delays pushed delivery into a market where lower-cost voice assistants already performed practical tasks.
• The company’s assets were sold and cloud-dependent functions were later shut down.^[6]

Jibo was capable of producing affection. That is not the same as earning a recurring household job. Its stationary body and limited utility made the social interaction the product, while Amazon and Google trained buyers to evaluate home devices through practical commands at lower prices. Timing and execution contributed. The deeper exposure was that companionship had not been shown to support the cost and technical burden of a dedicated robot.

TRANSFERABLE LESSON: Delight can improve a job. It cannot substitute for one.

These cases do not prove that premium appliances, proximity networks or social robots are impossible categories. They show what happens when a company finances category infrastructure before the customer sacrifice is established.

Survivor check / Products with the same surface do not prove the same job

The failed company should be compared with a survivor serving an apparently similar outcome. The purpose is not to claim that one team executed better. It is to identify the ranked job that changes the causal model. A connected appliance that becomes part of a daily ritual, a social product anchored in an existing relationship, or a household device that reliably completes a recurring task may earn demand that a more technically ambitious peer does not.

This comparison protects the report from declaring an entire category unwanted after one company closes. Juicero does not prove that consumers reject subscriptions or kitchen hardware. Jibo does not prove that households reject robots. Their records support a narrower finding: the delivered outcome did not displace enough existing effort at the required price and frequency. A survivor must show the missing sacrifice and repeat behavior, not simply remain alive through more capital.

The board should reject any post-mortem sentence that cannot survive this comparison. If a successful company faced the same supposed root cause, the diagnosis must move down one level—from category to segment, from technology to job, or from customer praise to costly behavior.

Before a full build, the decision owner writes a two-page warrant:

1. Customer: one role in one context.
2. Trigger: the observable event that creates the problem.
3. Current response: what happened the last three times.
4. Cost of status quo: money, time, delay, risk or lost outcome already borne.
5. Why now: what changed in the customer’s environment, not the founder’s technology.
6. Smallest test: the least product needed to observe sacrifice.
7. Disconfirming result: the behavior that would make the team stop or change the problem.
8. Expiry: the date and runway at which the warrant must be renewed.

The warrant is approved only by evidence from customers, not by the quality of the opportunity narrative. It creates no permission to scale. It creates permission to run the next test.

Authorization protocol / Spend follows the warrant

The demand warrant is completed before a material roadmap, channel or hiring commitment. Product owns the incident record. Sales owns the current workaround and budget owner. Finance prices the full delivery burden. One executive records the disconfirming result that will cancel the spend.

Approval is narrow. Evidence from one role, trigger and price authorizes only the next reversible test for that combination. It does not authorize a broad market claim. Every material change to segment, use case or price creates a new warrant because it changes the causal proposition.

At the review, the author presents refused customers before converted customers. The purpose is to prevent a small number of enthusiastic buyers from hiding the distribution of urgency. If the refusal reason is absence of rank rather than product friction, the decision returns to problem search. Another feature is not an allowed response.

The survivable route has seven steps:

1. The CEO freezes net hiring and non-essential roadmap work for 30 days.
2. Product and growth publish retention, payment and acquisition by segment and source.
3. The team interviews churned, inactive and refused prospects around their last real incident.
4. One founder lists the three strongest existing workarounds, including “do nothing.”
5. The company selects one segment with observed cost and builds the smallest paid intervention.
6. A written review compares the promised behavior with the observed behavior.
7. The board receives one of three decisions: continue the problem, change the problem or return remaining capital through an orderly close.

New evidence can reopen a rejected direction. Enthusiasm cannot.

Normal intervention still works while the company can fund two or three complete learning cycles without assuming a new round. After that point, containment is more responsible than another broad launch.

1. Stop reporting sign-ups without activation, repeat use and source.
2. Split every cohort by segment, price and acquisition channel.
3. Remove discounts and founder labor from one controlled cohort.
4. Identify the last date on which a customer paid to solve the problem without your product.
5. Cancel roadmap items that do not test problem rank.
6. Reduce fixed cost until the company can survive the evidence cycle.
7. Set a board-approved kill criterion before the next release.
8. Prepare customer data export, refunds and vendor continuity if the criterion is met.

Stop rule / Protect the remaining option

The search stops when three bounded interventions fail to produce the named sacrifice in the same segment, or when the runway no longer permits a complete test and an orderly close. A founder may propose a different problem, but the proposal starts with a new incident record and a smaller cost structure. The old product does not carry evidentiary credit into the new search.

This rule is not designed to force premature shutdown. It prevents the team from buying an unlimited series of ambiguous results. A clear negative preserves capital, reputation and time for another thesis. An unclear positive consumes all three while allowing the answer to remain emotionally protected.

The board records one of four outcomes: demand earned, one test repeated because execution corrupted it, problem changed, or company closed. “Continue learning” is not an outcome unless it names the exact behavior, owner, budget and expiry date of the next test.

Building without demand is not a failure of imagination. It is a failure of evidence order.

The company begins with a possible problem and commits to a complete answer. Once the answer employs people and consumes capital, every weak signal becomes easier to promote. Press becomes awareness. Awareness becomes a waitlist. A waitlist becomes a market. The missing step is sacrifice.

The liberating implication is that “nobody wants this” is not an insult to the product or the team. It is a current result about a defined customer, problem and price. Found early, it costs a prototype. Defended for two years, it costs the company.