Every growth metric has a failure mode. The problem is not that metrics are bad. The problem is that companies let metrics prove more than they are capable of proving.
LTV hides assumptions. It can smuggle in optimistic retention, expansion, pricing, and margin. A beautiful LTV model can be a useful planning tool or a dangerous story about customers the company has not retained yet.
CAC hides mix and saturation. A blended CAC can look healthy while marginal CAC is deteriorating. It can also hide expensive labor, sales engineering, partner incentives, or implementation work that was not included in the acquisition cost.
ROAS hides margin quality. Revenue is not contribution. A campaign can produce attractive revenue while creating low-margin, discount-trained, refund-prone, or operationally heavy customers.
Payback hides incrementality when it is based on attributed customers rather than caused customers. A payback calculation can look strong if the underlying channel is taking credit for demand that already existed.
Blended averages hide marginal decay. The older efficient dollars make the newer weaker dollars look better than they are. This failure mode gets more dangerous as budgets grow.
MQLs hide sales quality. They can reward volume while pushing low-intent accounts into sales. The cost appears later as low conversion, longer cycles, rep frustration, and pipeline distrust.
Installs and signups hide intent. A product can grow its top of funnel while attracting users who never activate, never retain, or never monetize. Activity becomes a substitute for customer quality.
Attribution hides causality. It can tell the company where credit was assigned, not whether spend changed behavior. Without incrementality, attribution can become an argument for budget preservation.
The metric failure-mode checklist should sit beside every growth dashboard. For each metric, write what it proves, what it does not prove, how it can be gamed, and what companion metric keeps it honest. A metric without a known failure mode is not trusted. It is underexamined.
The checklist should be blunt. For each metric, ask what behavior it encourages. MQL targets encourage lead volume. ROAS targets encourage credited revenue. CAC targets encourage cheaper acquisition. Payback targets encourage faster cash recovery. None of those behaviors is inherently bad, but each can become harmful when detached from customer quality and business model context.
The second question is what cost the metric excludes. CAC may exclude sales engineering. ROAS may exclude discounts. Retention may exclude downsell. Pipeline may exclude close probability. Payback may exclude support load. A metric's exclusions are often where the real business shows up.
The third question is how the metric changes under scale. A number that is useful at low spend may become misleading at high spend. A small channel can look clean because the company is operating in the easiest part of the curve. The failure mode appears when leaders assume the same economics will hold after budget increases.
The fourth question is what companion metric keeps it honest. CAC needs customer quality. ROAS needs contribution margin. LTV needs cohort reality. Payback needs incrementality. Pipeline needs conversion and cycle length. Activation needs retention. No metric should travel alone into an allocation decision.
The final question is who benefits if the metric is believed. This sounds cynical, but it is practical. Every metric has an owner, a narrative, and a budget implication. Good operators do not reject metrics because they are political. They make the politics visible enough that the metric can still be useful.
A strong metric review ends with decision rights. If a metric is only diagnostic, say so. If it can trigger budget movement, say what evidence threshold is required. If it is too weak to allocate capital, keep it in the dashboard but out of the funding conversation.
This checklist should be used before the metric becomes political. Once a target is tied to budget, compensation, or status, people naturally defend it. The cleaner moment is when the metric is introduced. Ask then what it can prove, what it cannot prove, and which companion metric will stop it from becoming a proxy for something it does not measure.
Good teams also retire metrics. A metric that helped during one stage can become noise at another. Early signup volume may matter when demand is unknown. Later it may distract from activation and payback. A dashboard that only gains metrics and never loses them becomes a museum of old anxieties.
The review should also ask what the metric makes invisible. A CAC target can hide lower-quality customers. A pipeline target can hide low conversion. A payback target can hide strategic option value. An NRR target can hide the cost of saving or expanding accounts. Each metric narrows attention, which is exactly why it needs a companion.
The practical answer is not a bigger dashboard. It is a small metric card for each number that enters the allocation review: definition, owner, decision use, known failure mode, companion metric, and kill condition. If a metric cannot survive that card, it should stay in analysis and out of capital allocation. That tiny card often saves the company from a very expensive misunderstanding.
The most useful version of the checklist is written in normal language. 'This number can prove X. It cannot prove Y. It can be gamed by Z. Before we move money, pair it with A.' That small translation keeps metrics from sounding more authoritative than they are.
This is part 9 of 10 in The Capital Allocation Theory of Growth.