The framework doesn't fail because the problems get hard. It fails because operators misidentify which problem they're actually in — or solve the visible version while missing the structural one quietly building underneath it.
The three failure modes below are not the obvious ones. Obvious failures get priced into valuations before you can act on them. These are the ones that look like health from the outside — right up until they don't.
Failure Mode One: Subsidized demand
There's demand that clears at market price — and demand that exists because a subsidy is holding it up. From the outside, they look identical. The difference only becomes visible when the subsidy ends, usually at the worst possible moment for the supply side of the business.
Peloton is the cleanest case study of the decade. The demand insight was real: people want connected fitness that makes exercise feel like a class, not a chore. That insight wasn't wrong. But two subsidies were propping demand far above its natural clearing level simultaneously.
First: COVID. Gyms closed. Demand for home fitness equipment was structurally redirected, not structurally generated. The question an operator must ask — and Peloton asked too optimistically — is: does this demand exist at market-clearing price when my customers have alternatives? The honest answer, in late 2020, was "we don't know." They proceeded as if they did.
Second: hardware pricing. Peloton priced its hardware below the margin that would make standalone unit economics work, betting software subscription LTV would carry the difference. This is a legitimate strategy — but it's a subsidy. The trade only works if the LTV assumption holds. When churn accelerated as gyms reopened and COVID immunity faded, the LTV math broke from both ends simultaneously: lower retention and lower new subscriber volume.
The capital allocation consequence was severe. They had built supply infrastructure — the Tonal acquisition, the Precor factory in Ohio — to meet demand that existed partly because competitors were physically closed and partly because hardware was priced below cost. When both stimuli faded, they were carrying the fixed costs of a supply machine scaled for demand that never existed at those economics without the subsidies.
The question isn't whether demand exists. It's whether demand exists at a price that makes your supply economics work — without the tailwind.
The diagnostic for any high-growth demand story: what happens to churn and acquisition if I raise prices 15%? If I have no pricing sensitivity data, I don't know whether I have real demand or subsidized demand. Operators who've never stress-tested WTP against a scenario where the tailwind reverses have a blind spot that gets expensive when conditions normalize.
The broader pattern: every VC-backed business that raised at 50x+ ARR in 2020–2021 on the back of pandemic-accelerated adoption was sitting on a version of this problem. Some had real demand underneath the acceleration. Many had subsidized demand dressed up in strong cohort data from an anomalous period. The ones that didn't distinguish between the two made supply investments they're still rationalizing.
Failure Mode Two: Demand debt
Borrowed from software: technical debt is when implementation shortcuts that feel fine today compound into problems tomorrow. Demand debt is when implicit promises made to early customers are sustainable at current scale and start breaking as you grow.
The mechanism is almost universal in growing businesses, and almost universally underestimated. Your first 100 customers get the real product: founder attention, bespoke implementations, things that obviously don't scale. Those customers have exceptional outcomes. They renew at high rates. They expand. They become the case studies in your pitch deck and the testimonials on your website.
Your next 1,000 customers get the scaled version. It's good. It works. But it's not the same product. The gap between what your early customers experienced and what your scaled customers experience is the demand debt — the implicit promise you made with your early success that you can no longer fully honor.
Demand debt shows up in the data in a specific, non-obvious way. Aggregate NRR looks fine — say, 108%. But cohort analysis tells a different story: your 2020 and 2021 customer vintages are at 122% NRR. Your 2023 vintages are at 91%. The aggregate is being held up by early cohorts who got the founder-delivered product. The newer cohorts — the ones that determine whether the business is compounding or decelerating — are already quietly breaking.
This is why aggregate NRR is one of the most gamed metrics in SaaS investing, and why operators who only report it are often hiding something, sometimes even from themselves. The longitudinal cohort view — broken out by acquisition vintage — is the demand debt x-ray. If you can't see it, you can't manage it.
Demand debt is most dangerous in businesses with strong early NRR. Strong early NRR tells you the product worked for the first cohort. It tells you almost nothing about whether the scaled product will work for the tenth.
Professional services firms accumulate demand debt in a specific flavor: the senior partner promise. Every major strategy firm and law firm sells on the implicit assurance of senior engagement. The partner who won the relationship will be involved. As they scale, the partner-to-client ratio shifts. Newer clients get the associate version of the product. The demand debt isn't booked on a balance sheet — it accumulates as brand equity borrowed against a promise that erodes with each new hire class.
Consumer brands accumulate it differently. Lululemon's first decade was built on a specific demand promise: technical quality, community culture, premium positioning. Mass distribution and rapid category extension blurred the product delivery while the marketing held the original promise. The demand debt wasn't visible in revenue — it showed up in weakening pricing power at the margin and in the gradual compression of the brand's ability to justify its premium against increasingly credible alternatives.
The corollary, worth stating directly: demand debt is hardest to see in businesses with strong early NRR, because those early cohorts are doing exactly what you want — renewing, expanding, referring. They are living proof that the product works. What they are not is proof that the scaled product works for the customer who signed up after the founders stopped doing implementation calls personally.
Failure Mode Three: WTP ceiling compression
Every demand problem has a ceiling: the maximum price a customer will pay, determined by the cost of their next best alternative. Most operators think of this ceiling as static — set it, defend it, price below it to win. But the ceiling moves as alternatives improve. And right now, for a significant swath of enterprise software, the alternatives are improving fast.
The mechanism is specific. Foundation models can now replicate the core analytical output of many specialized software tools at dramatically lower cost, with no meaningful integration overhead. A contract review platform charging $200K annually for AI-assisted standard clause analysis faces a WTP ceiling that has structurally compressed — because a general-purpose model with a well-crafted prompt does 80% of the job. Not all of it. 80%.
For the 80% of customers who needed 80% of the capability, the alternative just improved substantially. Their willingness to pay $200K — which was set by the cost of doing it manually or the cost of the next-best tool — just got recalibrated. The ceiling moved. The vendor's pricing power, which felt structural, was actually downstream of alternative quality — and alternative quality just step-changed.
The businesses that are genuinely not exposed to this compression share a property: their value is not primarily informational. Veeva Systems serves pharma and biotech with compliance-grade data management built on 15 years of customer-specific integrations, FDA audit trail requirements, and switching costs denominated in regulatory risk, not software preference. A foundation model generating similar-looking output doesn't solve the compliance problem. The WTP ceiling for Veeva's customers is set by what it costs to switch and maintain regulatory standing — not by whether a general-purpose model can produce similar analysis.
That distinction — ceiling set by software substitutability versus ceiling set by something more durable — is the key diagnostic for any SaaS business in the current environment.
The two viable operator responses:
Become the alternative. Embed AI deeply enough that your product is how your customers access AI, plus your proprietary data layer and workflow integration. The moat is no longer the analytical capability — it's the data accumulation and the workflow that would be expensive to rebuild elsewhere. You're not competing with foundation models; you're powered by them.
Find WTP niches where the ceiling is non-substitutable. Regulated industries, physical-world service delivery, institutional trust relationships, government contracting. Places where "a model can do 80% of this" is not acceptable as a substitute because the 20% it can't do carries legal, operational, or safety consequence.
There is a third response that looks like the first but isn't: adding AI features to an existing product without addressing whether the core value proposition is fundamentally substitutable. That's a roadmap update. It is not a ceiling defense.
| Failure Mode | Surface Signal | Underlying Reality |
|---|---|---|
| Mode 01 Subsidized Demand |
Revenue growing, engagement strong, cohort data healthy. Looks like product-market fit. | Demand exists because of a structural tailwind (crisis, subsidy, competitor absence) — not because willingness to pay holds at market-clearing price. Supply investment is being sized to the subsidized demand curve. |
| Mode 02 Demand Debt |
Aggregate NRR at 105–110%. Expansion revenue growing. Case studies strong. | Early vintage NRR is 120%+; recent vintages are 88–92%. Aggregate is being held up by cohorts who received the founder-delivered product. The machine serving new customers is quietly different from the machine that earned the brand. |
| Mode 03 WTP Ceiling Compression |
Pricing power intact. Renewals stable. No obvious competitive threat in traditional comp set. | The ceiling on willingness to pay is determined by alternative quality — and alternatives are improving faster than the product roadmap. The compression isn't visible yet. It arrives as churn at renewal from customers who discovered a cheaper substitute mid-contract. |
When your customer is also your supply chain
In platform businesses, the demand problem and the supply problem are not neatly separable — because the customer is part of the supply chain. Airbnb's hosts are the inventory. Uber's drivers are the production capacity. Reddit's users are the content. You're managing the incentive structure of a distributed, uncontracted supplier network that can, at any moment, simply stop supplying.
When that relationship breaks, the supply problem becomes visible instantly — and publicly, which compounds the demand problem simultaneously. Reddit's 2023 API pricing change triggered a moderator blackout. Their supply chain went on strike. The platform survived, but the episode exposed the structural vulnerability: third-party moderators were providing the curation that made the platform valuable, with zero contractual obligation to continue.
The durability calculation for platform businesses is structurally different from asset-heavy businesses. Supply durability isn't about owning infrastructure — it's about maintaining incentive alignment across a distributed, voluntary network. That's a different management problem, requiring a different diagnostic. The question isn't "do we have enough capacity?" It's "do our suppliers believe this relationship is worth maintaining?"
Phase transition errors
Most businesses move through phases where one problem is more binding than the others:
Peloton's capital allocation failure was a phase misread: they were still in Phase 01 — the demand thesis was unvalidated at market price — and they allocated capital as if they were deep in Phase 02. The supply infrastructure was right if the demand assumption was right. It wasn't.
WeWork's S-1 introduced a metric the company called "community-adjusted EBITDA" — a non-GAAP figure that excluded stock-based compensation, interest, taxes, depreciation, and, notably, most of the costs associated with building out and operating their locations: design, development, and pre-opening expenses.
The result was a metric that made the unit economics appear positive while the business was losing approximately $219,000 per hour on a GAAP basis. The supply problem — long-term commercial leases with fixed costs against variable demand — was not only unresolved; it was architecturally excluded from the metric the company chose to present as the measure of health.