Unable to reach product-market fit.

In July 2015, Homejoy said it would shut down. Co-founder Adora Cheung identified worker-classification lawsuits as the deciding factor in the company’s inability to raise more capital. That explanation was real. It was not complete.

Contemporaneous reporting found that Homejoy had also expanded while customer retention remained weak, individual cleanings lost money and the service depended heavily on deal-driven acquisition. By the time the company tested higher-intent channels and service improvements, the runway was short.^[2]

The market did not lack a problem. People already paid to have homes cleaned. The product did not fail to function. A cleaner arrived and cleaned. The fit broke between the specific customers Homejoy acquired, the consistency they required, the workers the marketplace retained and the price at which the service could be delivered.

This is why “we have customers” is a weak PMF claim. Customers can arrive for a discount, remain because a founder rescues the account or sign a contract that will not renew. One cohort can love the product while the next six use it once. An aggregate revenue chart turns those different facts into one reassuring line.

The root cause is not failure to discover a magic feature. It is the decision to scale before locating the complete unit of fit: a coherent segment, a recurring job, a product behavior that creates value and an economic route for delivering it. When those elements remain blended, every department solves a different company.

Boundary inspection / Demand can exist without fit

The existence of a real problem is necessary and insufficient. Homeowners want reliable cleaning. People want packages shipped without losing an afternoon. Viewers want entertainment in idle moments. None of those observations proves that a specific product, segment and delivery model can sustain repeated value.

The boundary test begins without the product. If customers already incur meaningful cost, the demand pool exists. The next test holds the problem constant while changing the pairing. Does one segment retain more strongly? Does a different context produce repeat use? Does standard delivery preserve the outcome after founder intervention disappears? A positive answer points to a fit search. Repeated negatives across credible pairings may move the diagnosis back to Report 003.

Fit must also be bounded by time. A customer who uses a tax product once a year may be retained on an annual cadence; a daily workflow that disappears for a month is not. The board writes the natural recurrence interval before it reads the cohort. Otherwise every frequency can be interpreted as patience.

01 / The segment average

DIAGNOSIS: Retained and unretained customers are reported together, hiding the use case that works.

SIGNAL: Retention, gross margin and acquisition cost change sharply when cut by role, trigger, source or use case.

PROTOCOL: Define the narrowest retained cohort and rebuild reporting around it.

02 / Activation without value

DIAGNOSIS: Setup completion is mistaken for the customer receiving the promised outcome.

SIGNAL: Users finish onboarding but do not repeat the core behavior on its natural cadence.

PROTOCOL: Name the first value event and measure time to it, not screens completed.

03 / Founder-manufactured fit

DIAGNOSIS: Senior attention, custom work and relationship capital make early accounts succeed.

SIGNAL: Outcomes worsen when delivery passes to ordinary sales, success or operations staff.

PROTOCOL: Re-run the motion with capped founder time and standard product boundaries.

04 / Discount fit

DIAGNOSIS: A low price creates usage that disappears at sustainable economics.

SIGNAL: Repeat behavior or conversion falls when promotions end.

PROTOCOL: Test the viable price on a small, high-intent segment before buying more traffic.

05 / Wrong usage context

DIAGNOSIS: The job is real, but the product appears at the wrong place, device or moment.

SIGNAL: Customers recreate the outcome through exports, workarounds or another interface.

PROTOCOL: Follow the job into the existing workflow; do not force a new ritual without evidence.

06 / Service variance

DIAGNOSIS: The product promise depends on a variable human, supplier or local operation.

SIGNAL: Retention and support cost vary more by provider or city than by product cohort.

PROTOCOL: Instrument the service layer as part of the product.

07 / Expansion before transfer

DIAGNOSIS: Success in one segment or geography is assumed to transfer to another.

SIGNAL: New markets require different pricing, onboarding, supply or product behavior.

PROTOCOL: Treat each expansion as a new fit hypothesis with its own evidence.

08 / Growth as proof

DIAGNOSIS: Acquisition volume is used to overrule flat retention or negative contribution margin.

SIGNAL: The company stops growing when paid spend or founder outreach stops.

PROTOCOL: Freeze scale spend until the same cohort produces improving repeat behavior.

Review six signals monthly by segment and acquisition source:

No universal score creates PMF. The action trigger is divergence: if acquisition grows for two review cycles while repeat use, depth or unsubsidized margin worsens, fixed growth spend pauses. The company must explain which segment is producing the divergence before it resumes.

Cohort discipline / The average is not a customer

The fit panel is rebuilt at the lowest level the company can operate: role, trigger, acquisition source, price and delivery method. Averages are permitted only after the team has shown that the component cohorts behave similarly. If one cohort retains and another churns, the blended line is not a trend. It is a decision the company has postponed.

Every reported value event must be observable. “Activated,” “engaged” and “successful” are labels until they name the customer outcome. The record should show when the triggering incident occurred, when value arrived, what work the customer contributed and whether the action repeated without a discount or senior rescue.

The same cut applies to margin. A cohort is not fit if it retains only through unpriced human labor. Founder time, exception handling, refunds, supplier variance and local operations belong in the delivery burden even when accounting convention places them elsewhere.

CF-01 / HOMEJOY — The growth curve hid the retention curve

• Home cleaning was an established paid category.
• Homejoy raised roughly $40 million.
• The public closure explanation emphasized worker-classification lawsuits.
• Reporting also documented weak retention, unprofitable cleanings and costly expansion.^[2]

Homejoy’s legal exposure mattered because it threatened the labor model and the next financing. It does not explain why the business was already fragile. Deal customers churned. Cleaner quality varied. Direct relationships could move off-platform. The company scaled cities and staff before one retained, profitable service unit was secure.

TRANSFERABLE LESSON: Marketplace PMF must hold on both sides and in the transaction economics. Consumer bookings alone cannot prove it.

CF-02 / SHYP — A loved service in the wrong frequency segment

• Shyp removed the work of packing and sending parcels.
• It raised more than $60 million.
• Founder Kevin Gibbon later said the company embraced growth at all costs and did not build sustainability from the start.
• The company eventually narrowed to businesses and San Francisco, but ran out of financing time.^[4]

Shyp solved a real problem well. The consumer segment encountered it too infrequently to support the courier and warehouse system at the original price. Small businesses had higher frequency, but the company pivoted toward them after building an organization and brand around consumers.

TRANSFERABLE LESSON: Satisfaction without frequency can create a loved product and a weak company.

CF-03 / QUIBI — The job existed; the format did not fit it

• Quibi raised $1.75 billion before its April 2020 launch.
• It offered paid, short-form premium video designed primarily for phones.
• The company added television support after launch.
• It announced closure six months later; Jeffrey Katzenberg and Meg Whitman said the idea may not have been strong enough or the timing may have been wrong.^[6]

Short video clearly had a market. Streaming clearly had a market. Quibi’s task was to prove that commuters wanted professionally produced episodes, in a paid standalone service, on a phone. The pandemic damaged the commute thesis, but the fast move toward television also exposed a product-context mismatch. COVID was a severe contributing condition, not a complete diagnosis.

TRANSFERABLE LESSON: Evidence for each adjacent behavior does not prove the combined behavior.

The survivor check is decisive. Handy continued in home services; logistics companies serve frequent business shippers; TikTok and YouTube proved short video demand. The categories survived. These particular product-segment relationships did not.

Survivor check / Keep the problem constant

Each case has a misleadingly easy comparison. Homejoy can be compared with the continued existence of household services, Shyp with shipping platforms, and Quibi with mobile video. Category survival establishes that the underlying jobs were not imaginary. It does not establish that every entrant should have survived.

The useful survivor comparison keeps the customer problem constant and inspects what changed in the relationship. A survivor may target a higher-frequency user, enter the workflow at a different moment, rely on a less variable service layer, or charge the buyer who receives the economic value. Those differences are not implementation detail. They are the fit.

This prevents two opposite errors. The first is blaming the market whenever a product relationship fails. The second is blaming execution without naming the relationship that execution needed to create. If a company could have become viable only by changing customer, context, product core and economics at once, the original pairing did not have a repair plan. It had another startup hidden inside it.

The comparison ends with a falsifiable sentence: for this segment, at this trigger and price, standard delivery should cause this behavior to repeat by this date. A survivor can suggest the sentence. Only the company's own cohort can confirm it.

For each candidate segment, maintain one row with:

1. Segment: role, environment and trigger.
2. Job: the progress the customer is trying to make.
3. Current alternative: including manual work and doing nothing.
4. First value event: an observable product outcome.
5. Repeat cadence: when the job naturally returns.
6. Retention evidence: cohort behavior at that cadence.
7. Delivery burden: product, service and founder cost.
8. Viable price: tested, not imagined.
9. Expansion route: why the same fit can reach more customers.

One row may receive scale capital. Mixing rows into a total is prohibited until each has independently passed its own review.

Operating the map / One pairing per test

The fit map is owned jointly by product, finance and the executive responsible for the segment. Product defines the value event. Finance records the full delivery burden and viable price. The segment owner identifies the trigger and acquisition route. No function may change its field during the test without restarting the clock.

The company then exposes a new cohort to standard conditions. Founder sales can recruit the cohort but cannot rescue adoption. Discounts are either removed or recorded as part of the hypothesis. Support follows the service level the scaled model can afford.

At review, the team compares the promised recurrence with observed behavior and names the largest leak. One variable may change for the next cycle. Changing price, segment, feature set and channel together produces a new anecdote, not learning. Expansion remains frozen until two consecutive cohorts from the same pairing behave within the accepted range.

Intervention route / maximum 45 days

1. CEO freezes expansion, broad paid acquisition and net hiring.
2. Data owner rebuilds cohorts by segment, source, price and first value event.
3. Product interviews retained, churned and refused users from the same segment.
4. Finance publishes contribution margin with founder and service labor included.
5. The team selects one retained segment and removes non-essential promises.
6. A new cohort runs at the viable price and standard delivery burden.
7. The decision memo records continue, change segment, change product or stop.

Repair remains possible while the company has enough runway to observe the natural repeat cadence at least twice. If the product renews annually and the company has three months of cash, normal experimentation is over; the company must seek a bridge tied to a precise proof, sell, or contain the shutdown.

1. Publish segment-level cohorts to the board.
2. Remove customers whose outcome depends on permanent custom work from PMF claims.
3. Stop geography or vertical expansion.
4. Price the retained segment at sustainable delivery economics.
5. Cut roadmap work not tied to first value, repeat or margin.
6. Assign one owner to the fit map.
7. Set the evidence and date that reopen growth.
8. Prepare an alternate capital plan before restarting acquisition.

Stop rule / Do not scale an unresolved relationship

The search contracts or stops when no segment shows improving repeat behavior under viable delivery, when each promising cohort depends on a different custom intervention, or when the remaining runway cannot complete another natural usage cycle. The company may continue serving retained users, but it cannot describe maintenance revenue as proof that broad fit is approaching.

A working narrow segment creates three honest options: concentrate the company around it, redesign the capital model for its size, or use its observed behavior to test one adjacent segment. A non-working segment creates two: change one side of the pairing or stop. “Grow and learn” is disallowed because growth changes the cost structure before the relationship is understood.

The board decision records the segment, recurrence interval, viable price, delivery burden and next expiry date. If any field is blank, the company has not made a fit decision. It has extended an aggregate.

Unable to reach PMF is not the same as building something nobody wants. It is more deceptive. Someone wants it. The problem is real. A launch works. Revenue appears. The company fails because the observed value belongs to a narrower segment, different context or heavier service model than the plan admits.

Growth then becomes an anesthetic. It lets the company postpone the argument about who stays, why they stay and whether serving them creates a business. The eventual cash crisis is the invoice for that postponed argument.