A recurring pattern shows itself in growth budget reviews: the request to increase customer acquisition spend is anchored, almost invariably, to the same ratio, and the lifetime value figure sitting in the numerator of that ratio holds steady across quarters. The cohort base may have widened, a portion of the first-generation customers may have passed through their renewal decision, pricing may have been reset once — and yet the number on the slide is either unchanged or has been marked up. That no downward revision was made over the same interval is, on its own, sufficient evidence that the figure functions as an assumption rather than a measurement, since genuine measurements move in both directions.
A second and less visible pattern emerges in the institutional journey the number takes. The figure typically originates in a cohort table maintained on the commercial side, migrates into a single cell of the finance model, and then arrives on the summary page of an investment memorandum; the assumption schedule behind it — how many cohorts were observed, over what window, how the retention curve was extended, which costs were netted out — goes missing somewhere between those three stations. What reaches the board table, accordingly, is a single number severed from its derivation chain, and a number severed from its derivation chain cannot be argued with, because arguing requires an assumption to argue against.
The name for this pattern is LTV overestimation — the systematic upward drift of customer lifetime value — and it arises not from one error but from the accumulation of several small choices that all lean the same way. The first is extension of the retention curve beyond the observed window: when the behavior of a cohort observed for twelve months is carried out to five years, the error compounds along the tail rather than growing linearly, because the extended portion carries the majority of the modeled value. The second is survivorship in the cohort set, since the most mature cohorts are also those acquired under the most favorable conditions, and their behavior does not represent the customer arriving today through broader and cheaper channels. The third is the conflation of gross margin with cohort-level contribution margin: with support load, implementation cost, the selling cost of the renewal conversation, and payment infrastructure expense left undeducted, what has been computed is lifetime revenue rather than lifetime value. The fourth is the failure to discount future cash; the fifth is measuring churn on customer counts rather than on revenue.
None of these choices is irrational at the moment it is made. An early-stage company can justify spending that has not yet reached payback only through a forward-looking thesis, and extending a retention curve while data remains thin is a functional shortcut that lowers the cost of deciding — without it, the growth decision could not be taken at all. The difficulty lies not in the shortcut but in its persistence once conditions change: once the observation window has grown long enough to test the assumption on which the estimate rests, the estimate ceases to be a bridge and hardens into a constant. What accelerates that hardening is the incentive architecture, since where the team that maintains the number is also the team that spends the budget the number unlocks, downward revision becomes an institutionally costly act, and costly acts are deferred.
The first institutional consequence appears in the cash cycle. Acquisition spend is a present, certain, and irrecoverable outflow, whereas lifetime value is a contingent future inflow, and the two never meet within the same reporting period. To the extent that an inflated LTV shortens the apparent payback period, it also understates the working capital requirement, with the result that the company must finance a shortfall not yet visible in the income statement either through unplanned bridge facilities or through an abrupt halt to growth. The second of these carries a secondary cost that is rarely priced in advance: sales headcount is a durable expense commitment while the revenue it was hired against is not, and the decision to reduce that headcount destroys both cash and institutional memory.
The second consequence materializes at the diligence table. In a financing or acquisition process, among the first tasks a commercial diligence team undertakes is rebuilding the cohorts from raw transaction and billing data and setting the resulting curve against the one presented; that comparison tests not the seller's model but the seller's assumption discipline. Where the gap proves material, the center of gravity in the negotiation typically shifts from the valuation multiple to the quality of the revenue base to which the multiple is applied, and that shift produces two structural outcomes — a portion of consideration is tied to an earn-out, and the escrow percentage moves up. Price can be settled in a single conversation; doubt about revenue quality, by contrast, can push the post-closing collection schedule out by months.
The third consequence accumulates in contract tails and credit documents. In portfolios where revenue-based retention — net revenue retention — concentrates in a handful of large accounts, logo-based churn can look stable while the revenue-based measure remains materially more volatile, and structures are observed in which that distinction opens the gap between the two metrics to an order of magnitude. Lenders typically meet this volatility with a covenant threshold derived not from the company's own model but from the creditor's conservative case, and once written into the documents, that threshold constrains growth more forcefully than any internal LTV assumption ever did. An assumption never debated internally thus returns from the outside as a binding constraint.
The mechanism that neutralizes this tendency is not more careful estimation but institutional fixing of the frame within which estimates are produced, and that frame has four separable components. The first is an observation-window rule: the retention curve is capped so that it does not extend beyond the actually observed period by more than a stated multiple, the tail is either truncated or bound to an explicitly finite horizon, and the contribution of the extended portion to total value is reported as a separate line. The second is the margin definition: LTV is computed on cohort-level contribution margin rather than gross margin, support and renewal costs are charged back to the cohort, and future cash is discounted at the company's cost of capital. The third is churn decomposition, with gross revenue loss, contraction, and expansion each measured separately, since contraction masked by expansion tends to surface all at once several quarters later. The fourth is the use of a band in place of a point estimate, where the lower bound enters budget approval and the upper bound enters scenario analysis only.
What makes those four components durable, however, is not a technical correction but a recording rhythm: every LTV estimate is logged together with the date it was produced, the cohort set it rests on, and the person accountable for it, and at defined intervals the estimate made at that date is compared against value actually realized since. This backward test is run not to establish who was right but so that the institution can learn the direction and magnitude of its own drift; an organization that knows which way and by how much its estimates typically miss can calibrate future estimates against its own history. That the record is kept at the moment of proposal rather than the moment of approval is decisive here, since a record kept at approval documents only the outcome, while a record kept at proposal documents the assumption.
BEIREK's intervention in structures of this kind begins not with rebuilding the model but with building the governance around it. Cohort data is reconstructed independently from raw transaction records and set line by line against the curve the company reports; an assumption register is opened in which each assumption carries an owner, a date, and a stated basis, and maintenance of that register is assigned to a line of accountability separate from the team consuming the growth budget. A quarterly review rhythm then runs: prior estimates are compared against realization each period, the direction and magnitude of drift are documented, and the following period's band is calibrated against that documentation. What these three elements produce together is not a more optimistic or a more pessimistic number, but a number that a counterparty's own team can reproduce in a financing, acquisition, or credit process — and therefore need not discount.
Where a company's lifetime value estimates have moved in only one direction over the past two years, the question worth asking is not whether those estimates are correct, but whether the estimation process was ever constructed in a way capable of producing a downward revision at all; and the answer to that question tends to lie less in the model itself than in the identity of whoever owns it.