A recurring scene plays out in marketing budget reviews. A channel-level performance table is placed on the table, some rows glowing with triple-digit return figures while others sit almost empty in the conversion column. Occupying the upper rows, typically, are campaigns bidding on the company’s own brand name, retargeting sets served to abandoned carts, and lifecycle email flows; occupying the lower rows are impression-based campaigns, trade publication partnerships, field events and content investment. The outcome of the review usually follows the ordering of the table — incremental budget upward, reductions downward. The decision itself appears disciplined, even evidence-led. Yet every number in every row of that table is the product of a single accounting convention, and that convention is nowhere discussed in the meeting.
The same pattern surfaces in the negotiations channel teams conduct among themselves. The team responsible for brand argues that its activity lifts search volume; the performance team produces its conversion record, and the discussion ends there, because one side holds a number rendered at row level while the other holds a narrative. In corporate decision-making, the general tendency of a number to defeat a narrative is a healthy reflex; what prevails here, however, is not the accuracy of the number but its format. What passes unnoticed is that both parties are describing the same phenomenon, and only one of them possesses a measurement already shaped for the reporting template.
The name of this measurement regime is last-click attribution — the assignment of the entire conversion value to the touchpoint immediately preceding purchase and of zero to every touch before it — and its mechanics rest on an unusually simple assumption: that the purchase decision was formed at the moment it became visible. In enterprise procurement, and equally in high-consideration consumer decisions, the distance between the formation of a decision and its registration is measured in weeks. A buyer encounters a product in a trade publication, speaks with a peer six weeks later, then searches the brand name and purchases. The last-touch rule rewards only the final link in that chain and treats the remainder of it, for measurement purposes, as though it never occurred.
Understanding why the rule proves so durable requires reading it not as an error but as a cost-reducing shortcut. Last-touch attribution is reproducible, auditable, and remains the only model in which two different analysts working from the same data will arrive at the same result; multi-touch models, by contrast, are sensitive to the choice of weighting coefficients, the length of the observation window and the definition of a channel, and may therefore return a different table on each re-run. A finance function preferring a reproducible approximation to an irreproducible truth is behaving rationally. The difficulty lies not in the shortcut itself but in its persistence after the conditions that justified it have changed — as channel count multiplies, as the purchase cycle lengthens, as the weight of spend migrates toward the upper funnel.
The institutional cost appears first in budget composition, quietly and by degrees. The channels that structurally occupy the final position — branded search, direct traffic, retargeting, abandoned-cart flows — are harvest channels; they do not generate demand, they collect existing demand at low friction. Each additional unit of budget directed toward them produces a measurable near-term lift in conversions, because there is demand in the pool to be gathered. The activity that replenishes the pool, however, registers at zero contribution in the same table, and is accordingly trimmed across successive budget cycles. Over three or four quarters the table improves steadily, the centre of gravity of the budget shifts toward harvest, and the organisation discovers that it has stopped financing its own demand generation capacity only once the pool begins to thin.
The second cost appears not in the income statement itself but in the time series of acquisition cost. Cost per new customer declines initially as harvest weighting rises — collecting demand already formed is cheaper than forming it — and then, past a threshold, rises rapidly and steeply, since the same channels are now bidding at higher frequency into a contracting pool. Because this two-phase curve leaves a distance of six to eighteen months between the quarter in which the budget decision was taken and the quarter in which its consequence becomes legible, the increase is typically attributed to competitive intensification, platform cost inflation or macro softness in demand. The causal chain is rarely traced back to the decision that produced it.
The third cost, and the most persistent, is structural. So long as the attribution model remains a reporting preference, it is a contestable assumption; the moment channel targets, bonus structures and the performance thresholds written into agency contracts are indexed to that same model, it ceases to be an instrument of measurement and becomes an institution that generates behaviour. From that point onward, the rational strategy for a team is not to produce incremental contribution but to position itself at the terminal link of the attribution chain — widening retargeting windows, bidding aggressively on branded search, clustering activity around touchpoints proximate to conversion. Internal competition ceases to be competition for the customer and becomes competition for the record, and this transformation occurs entirely within the logic of the incentive structure, without bad faith on anyone’s part.
This tendency is not resolved through individual awareness; a marketing director’s knowledge that the last-touch model is flawed changes nothing about behaviour so long as the table itself remains bound to the budget. The neutralising mechanism is structural, and it separates into four components. The first is the written statement of the attribution model as an assumption, with the model name, the observation window and the excluded touch types disclosed in every version of the report. The second is the maintenance of incrementality as a record distinct from attribution — geo-based or segment-based planned holdout windows, channel-off tests, regional variations in media weight, observations that reveal what occurs in a channel’s absence. The third is tracking harvest-channel spend as a separate line item and elevating its ratio to demand-generation spend into a board-level indicator. The fourth is migrating the incentive structure away from channel-level attributed conversions toward period-level blended acquisition cost and customer lifetime value.
Each of these components carries a different meaning for a different role. For marketing leadership the question is not which model to adopt but which decision will be bound to which evidence, settled in advance. For the finance function the question is how to introduce an incrementality record into the system without surrendering reproducibility. At board level the question is whether to look past a single efficiency ratio and monitor the ratio between demand generation and demand harvesting as an indicator of structural balance. On the enterprise procurement side, the economic weight of these distinctions grows as the purchase cycle lengthens, since the distance between first contact and close typically exceeds the span any attribution window is capable of observing.
The intervention BEIREK operates in capital-intensive, long-cycle programmes begins not with replacing the model but with rendering the decision regime visible. The first mechanism established is an assumption register: which measurement rule directly triggers which budget decision, which line items would change direction if the rule changed, and which line items are indifferent to the model, all recorded in a single table. The attribution choice thereby ceases to be undiscussed infrastructure and becomes a reviewable decision. The second mechanism binds incrementality to institutional cadence — holdout and re-activation windows placed on the calendar, results held in a record separate from the attribution report, and the points at which the two records diverge taken up explicitly in the periodic review.
The third mechanism sits on the incentive line: performance thresholds for channel teams and external agencies are migrated from attributed conversion counts to a structure in which period-level blended acquisition cost and the demand-generation share are assessed together, with the reporting definitions in agency contracts rewritten accordingly. The shared purpose of these three mechanisms is not to declare which model is correct, but to remove the model from its position as the sole basis of the budget decision; the problem encountered in practice is rarely that the model is wrong, but rather that an approximation has been promoted, without anyone noticing, into a capital allocation rule.
When a measurement rule remains unchanged for long enough, it no longer describes what it measures; it shapes what is measured. Last-touch attribution is the most widespread instance of that transformation, because its imperfection is not hidden — everyone knows the rule simplifies — but what the rule simplifies turns out to be the direction of the budget, and this becomes visible only once the demand pool has thinned and acquisition cost has risen sharply. The single question an organisation might put to its own measurement architecture is this: were this rule to change tomorrow, which line of the budget would move in which direction, and which decision, not taken today, would that movement turn out to be the delayed form of?