When a platform company convenes its quarterly budget review and the question of how to divide the user acquisition allowance is opened, the weight of the allocation moves, almost predictably, toward whichever side produces a countable signal, because on that side the chain running from click to registration to first transaction to repeat transaction has been instrumented end to end, and every unit of spend therefore reappears in a table the following week. The question of where scarcity is actually forming is frequently never raised in the same room; the assumption that supply will arrive of its own accord as demand grows sits beneath the discussion as a premise rather than as a proposition under examination. Redirecting a portion of the allowance toward the other side — not because that side cannot be measured, but because measuring it is expensive — rarely reaches the agenda at all.
A second observable pattern in that same review is that the distribution itself is inherited from the preceding quarter: because the incentive line already exists, the debate proceeds on what percentage the total budget will grow rather than on how the allocation between sides should be set. The probability that an existing line is reapproved is markedly higher than the probability that the same line would be approved if proposed for the first time, and this asymmetry holds the budget in place even where the platform's liquidity dynamics have changed entirely between two quarters. In a growing platform, moreover, the migration of scarcity from one side to the other is the ordinary course rather than the exception; as transaction volume rises, the bottleneck typically moves from demand to supply, and then from supply in general to a particular quality, geography, or response time within supply.
The pattern has a name — **subsidy-side misidentification**, the platform subsidizing the side whose behavior is best measured rather than the side that in fact carries the cross-side network effect — and its mechanism operates on two layers. The first layer is measurement asymmetry: data infrastructure is built wherever money passes through, and an allocation decision in which only one of two options generates feedback will, foreseeably, drift toward the option that generates it. The second layer is the conflation of price elasticity with visibility, since the side that speaks most audibly on the platform, that has account managers assigned to it, and that is represented in satisfaction surveys need not be the side that is genuinely responsive to incentives; in the meeting room, however, the first regularly stands in for the second, and no participant is required to notice the substitution for the allocation to follow it.
Recognizing that this shortcut is functional under specific conditions is a precondition of the diagnosis rather than a qualification of it. While both sides of the platform remain scarce, every unit of acquisition raises the probability of a match irrespective of the side on which it is spent, and in such a period leaning on the measurable side is not merely inexpensive but defensible before a board, because the rationale can be supported with numbers. The difficulty lies not in the shortcut itself but in its persistence once conditions change: when the match rate approaches saturation and the obstacle in front of the marginal transaction is no longer demand but supply of a particular character, the same allocation logic ceases to manufacture liquidity and merely purchases existing demand at a lower price.
A third and less frequently discussed layer concerns the difference in attachment structure between the two sides. On many platforms one population works within a single venue while the other conducts the same activity across several venues simultaneously, and an incentive directed toward the multi-homing side buys a temporary price differential rather than a preference, with the consequence that the same side is typically the first to withdraw once the incentive is reduced. Allocation that produces durability, by contrast, supports not the side whose cost of leaving is already high but the side whose continued presence widens the option set available to the counterparty. That distinction becomes legible in a cohort table constructed per side; it does not become legible in a campaign report, where spend and conversion appear together and durability appears nowhere.
The financial expression of this tendency is usually concealed not in the incentive line itself but in blended acquisition cost. Once the acquisition cost of two sides is consolidated into a single average, the aggregate indicator can appear stable even while marginal cost is climbing steeply on one side because of saturation and declining on the other, and that visual steadiness postpones recognition of the misallocation by roughly the length of a full budget cycle. When contribution margin is instead decomposed by side, the resulting picture is generally different in kind rather than in degree: the payback period on the subsidized side lengthens quarter over quarter, while each participant added on the unsubsidized side may open transaction volume several times larger, a divergence that the blended figure is structurally incapable of surfacing.
The second financial surface is accounting classification. Incentives transferred directly to users, even where they were planned at the design stage as marketing expense, may be reclassified during audit as a reduction of revenue, and that reclassification bears directly on gross revenue and therefore on every valuation discussion anchored to a revenue multiple. A growth narrative told through take rate produces a materially different curve once the portion of the incentive netted against commission is separated out, and resolving that treatment while the incentive line is being designed is, as a practical matter, considerably cheaper than resolving it at the closing table, where the question arrives coupled to a purchase price adjustment and to a compressed timetable.
The third surface sits on the transaction desk itself. When an acquirer or an investor rebuilds the platform's cohorts on a per-side basis, the question posed is not the growth rate but which side eroded, and at what speed, during periods in which the incentive was withdrawn or reduced; where a company cannot answer that question from its own data, the subsidy-dependent portion of volume is priced conservatively, at the upper bound of what is plausible. In practice the consequence takes one of three forms — a direct valuation discount, an earn-out conditioned on post-subsidy retention, or representations and warranties concerning incentive accruals that push the escrow percentage upward. Much as with founder dependence, subsidy dependence renders not performance itself but the repeatability of performance the contested question.
The mechanism that neutralizes this tendency is decision architecture rather than individual awareness, and it separates into four components. The first is per-side unit economics: blended acquisition cost is removed from management reporting altogether, and each side is reported with its own contribution margin, its own payback period, and its own retention curve. The second is a constraint record, in which the side holding the bottleneck in front of the marginal transaction is determined in writing each period against a single indicator — match rate, utilization, or first response time. The third is the coupling of budget authority to that record, so that when the constraint changes sides the allocation changes as the output of a rule agreed in advance rather than as the outcome of a fresh negotiation. The fourth is that every incentive line carries a sunset date written at the moment of its creation, with the behavioral difference measured in a tapered region treated as the measure of the incentive's genuine effect.
BEIREK typically constructs this intervention through three records. The constraint record is made a periodic document in which the side holding the constraint, the indicator on which that determination rests, and the change from the preceding period are held on a single page, so that the allocation discussion does not restart without memory each quarter. Incentive lines are bound to a contractual calendar, with the opening rationale of each line, the indicator against which it will be assessed, its taper steps, and its termination date recorded as the line is opened; kept at the moment of proposal rather than at the moment of approval, that record prevents retrospectively constructed rationales from carrying the budget forward. Accounting classification is resolved while the incentive is being designed rather than during the audit period, when the cost of resolution has already migrated to the valuation.
Alongside these records, the question of where the constraint will migrate under scenarios in which volume grows several times over — holding the current allocation fixed — is worked through in advance, and this stakeholder pre-mortem forces the discussion onto the platform's next bottleneck rather than its present one. The finding that emerges most often in practice is not that the incentive is flowing to the wrong side, but that it is flowing to the right side and has been fixed there, while the platform's growth has already moved the correct side elsewhere. A budget that was accurate at the moment it was set thereby becomes, without any decision having been taken to that effect, an expenditure that purchases volume the platform would have obtained in any event.
What determines the value of a two-sided business is less the identity of the subsidized side than the capacity to demonstrate on which side the platform would remain standing were the subsidy removed; for as long as that demonstration is unavailable, growth is priced as volume rented rather than owned.