Quarterly marketing reviews tend to reproduce a recurring scene: a campaign that came off air six weeks earlier is assessed against the sales recorded during the weeks it ran, while in the same meeting the search volume and direct traffic it left behind are credited to the performance channel whose budget was recently increased. Read channel by channel, the table is clean, every line has an owner and the columns reconcile. What the same table records nowhere is the effect that persisted in the system after the campaign stopped running, and precisely because it is recorded nowhere, that effect accrues to whatever happens to be measurable. This is not an accounting error; it is the natural consequence of a reporting unit whose periodicity does not match the temporal structure of the phenomenon being measured.
The inverse of the same pattern is equally observable. Where a brand line whose budget was cut in the third quarter continues generating qualified demand through the fourth, that continuity is customarily read as evidence that the cut was harmless, and the reduction is deepened in the following cycle. The measurement window itself is rarely the company’s own decision either: post-click conversion windows are inherited from a media platform’s default setting and become a reporting standard bearing no relation whatever to the category’s actual purchase cycle — evaluation processes running in weeks on one line and in quarters on another. The firm has thereby tied the rhythm of its own capital allocation to somebody else’s technical assumption.
The mechanism beneath this pattern is what is termed adstock misestimation — the mis-parameterisation of advertising’s carryover effect across time. Carryover is not one quantity but three, each moving independently of the others: the lag before the effect reaches its highest point, the rate at which it decays thereafter, which is to say its half-life, and the saturation curve beyond which additional frequency yields diminishing returns. In practice all three are compressed into a single intuitive assumption, with money spent matched against outcomes measured inside the same reporting period, the lag treated as zero, the decay as instantaneous, the saturation as linear. No model has been built, and yet a model is nonetheless in force.
Part of this shortcut is entirely functional. In promotion-intensive categories with short purchase cycles, where the decision is largely made on price, assuming that the effect exhausts itself within the period is a defensible approach and materially reduces the cost of measurement; under those conditions the marginal return to modelling lag does not justify the modelling burden. The difficulty lies not in the shortcut itself but in its persistence after the product mix, the channel structure or the customer profile has changed. Once an institutional buyer, a dealer network or a line with a long evaluation cycle enters the portfolio, the temporal structure of the category shifts, whereas the measurement framework, once settled, typically does not — there being no natural owner whose interest it serves to demand its revision.
A second difficulty resides in estimation itself. A decay parameter can only be identified from a dataset in which spend varied meaningfully; where the history shows spend held within a narrow band across periods, decay rate cannot be statistically separated from seasonality, price movement and distribution expansion. A model fitted on such data does not measure the parameter — it returns the analyst’s starting assumption dressed as goodness of fit. The result is a self-confirming framework: because half-life is assumed to be short, short-cycle channels perform well; because they perform well, they receive budget; because they receive budget, variation in spend narrows further, and the parameter becomes marginally less estimable with each successive period.
The institutional cost of this mechanism accumulates as a gradual and difficult-to-reverse migration in budget allocation. Channels that harvest demand — branded search, retargeting, cart recovery — convert an intent generated somewhere upstream of them and therefore always appear superior within the measurement frame, while demand-creating lines, leaving the greater part of their effect outside the measurement window, are perpetually undervalued. Each budget cycle shifts a few points, and in no single cycle does the shift look wrong; after several years the mix has become a structure that draws down the stock of generated demand without replenishing it. That no individual step in the sequence was mistaken does nothing to prevent the outcome from being structural.
Depletion becomes visible not on the marketing line but two or three items away from it. Branded search share retreats, the quality composition of traffic changes while the conversion rate holds, the discount required to close a sale deepens and promotional frequency rises. What the income statement then displays is an improving ratio of marketing expense to revenue alongside a deteriorating gross margin, two figures rarely placed side by side in the same meeting because they live in different reports owned by different people. That divergence — the efficiency indicator improving as the pricing-power indicator erodes — is the earliest observable signal that carryover has been consumed, and it typically appears several periods after the decision that produced it.
The same structure translates directly into valuation language in a change-of-control process. The question asked at the diligence table is not whether growth was recorded but whether it can be shown to be repeatable independently of the founder and of incremental spend; where the demand-generation mechanism is undocumented, a buyer prices that gap either through a discount on the multiple or by shifting part of the consideration into an earn-out structure. The critical asymmetry sits here: earn-out measurement periods are typically constructed over twelve to twenty-four months, whereas in considered-purchase categories the decay tail of carryover may run beyond that horizon. The mismatch makes harvesting the existing stock of demand economically rational on the seller’s side, and leaves the buyer, even where the contractual targets are met in full, inheriting a demand base whose tail has been spent.
What neutralises this tendency is not analyst vigilance but the architecture built around the decision, and it has four separable components. The first is that the decay parameter be generated through planned variation rather than estimated from historical data: geo-split holdout tests, staggered launch calendars and pre-declared dark periods are the only inputs that render the parameter identifiable. The second is that decay and saturation be kept distinct in the record, since a decline in a channel’s return demands entirely opposite budget decisions depending on whether it originates in the effect fading or in a frequency threshold having been crossed. The third is that the review rhythm be tied to the category’s half-life rather than to the fiscal calendar. The fourth is the separation of the owner of the in-period number from the owner of the carryover assumption, so that defending the assumption and defending the result do not converge in the same individual.
BEIREK operates these four components through an assumption register. For each channel the lag, half-life and saturation threshold are recorded in writing at the moment the budget is proposed rather than at the moment it is approved, so that the end-of-period assessment compares outcomes against a previously declared assumption rather than against one recalibrated in light of what actually happened. An annual test calendar runs against that register: variation in spend is not left to circumstance but produced deliberately, at the minimum amplitude required to keep the parameter identifiable. A bridge is also maintained between model output and the income statement, on the principle that an improvement in the marketing efficiency indicator is not sufficient for a decision unless it is read on the same page as gross margin and average discount depth.
In a transaction context, the same register is converted into a tail calendar. The demand base being acquired or divested is decomposed into the portion arising from stock already generated and the portion sustained by ongoing spend; the earn-out measurement window and the category’s decay tail are laid over one another, and the gap between them is addressed within the definition of the target itself. In practice this means altering the composition of the target rather than extending the measurement period — measuring, alongside revenue at period end, the branded search volume handed over, the repeat-purchase cohort and the share of sales transacted without discount. The undertaking thereby ceases to be a harvestable quantity and becomes a stock that must be replenished.
Advertising effect is a continuous process; the fiscal period is a legal construct, and consumer memory is under no obligation to conform to it. What determines whether a company has priced its marketing investment correctly is not which model it uses, but whether the temporal assumption embedded in that model was declared in writing and whether the variation required to test that assumption was produced deliberately.