In an investment committee session, the growth presentation of a company operating an intermediation model almost invariably opens with aggregate transaction volume, and where that volume compounds quarter over quarter, the remainder of the discussion tends to organize itself around the pace of growth and the efficiency of marketing spend. What rarely appears in the appendix, and what appears in a compromised form when it does, is the repeat-transaction rate measured at the level of the individual buyer-seller pair rather than at the level of the platform as a whole. Presented on the same chart as aggregate repeat behavior — the probability that a given buyer transacts again with any seller on the platform — the distinction between the two measures becomes invisible. The gap may read as a technical nuance, yet the first measure demonstrates only that the venue clears transactions, while the second demonstrates whether the relationships it originates remain within it, and it is the second that governs the long-run economics of the business.
The same pattern leaves independent traces on three separate surfaces. In the support queue, attempts to exchange direct contact details within the in-product messaging channel cluster immediately after the first completed transaction rather than distributing evenly across the relationship. In the payment infrastructure, the spread between the volume actually settled through the rail and the volume reported by the supply side widens systematically within particular seller segments rather than randomly across the base. And in supply-side interviews, the highest-volume sellers describe, without particular reluctance, a division of labor in which the platform serves acquisition of new counterparties while established counterparties are served through some other channel. Taken individually, each of these observations reads as noise attributable to product friction or reporting hygiene; taken together, they are the same mechanism casting a shadow on three different walls. The distinguishing signature of the pattern is that the loss concentrates not in weak relationships but in the strongest ones.
The mechanism is platform leakage: the continuation of a value-creating interaction the platform originated, sustained beyond a certain threshold outside the platform’s revenue and measurement systems. Its operation is straightforward and requires bad faith from no party. A platform in fact sells two distinguishable goods — the discovery that brings together two counterparties who would not otherwise have found one another, and the assurance that permits matched counterparties to complete a transaction with confidence — yet it prices both through a single transaction-based commission. The value of discovery reaches its maximum at the outset of the relationship and declines with every repeat transaction, since the parties have, by definition, already found each other; the commission, meanwhile, continues to be assessed at the same rate on every transaction. Past the point where those two curves cross, remaining on the platform becomes a quantifiable line item of cost for both sides simultaneously.
The second layer of leakage concerns data rather than revenue, and is institutionally the more insidious of the two. Quality rankings, search results, fraud scores, credit and limit decisions, and increasingly the platform’s own price guidance all operate on the observable transaction set; leakage removes observations from that set not randomly but selectively, because the interactions that escape are precisely those in which the counterparties are most satisfied with one another, transact at the highest volume, and generate the lowest dispute incidence. The observed set therefore drifts systematically toward weaker, more transactional, and more contested interactions, and the platform’s quality models progressively lose the ability to recognize its own best participants. The consequence is that product decisions taken on the strength of the data become miscalibrated over time, not because the analysis is flawed but because the generating mechanism beneath the data has been corrupted.
Leakage does not produce cost under all conditions, and treating it as uniformly pathological invites the risk of overcorrection. During the period in which the supply side has yet to deepen, a transaction completed off-platform circulates through the seller community as evidence that real revenue can in fact be sourced through the venue, which materially lowers the cost of subsequent supply acquisition; at that stage, every restriction designed to suppress leakage penalizes the very behavior that seeds liquidity before liquidity exists. The point of distinction is the direction in which the marginal value of what the fee purchases moves as the relationship matures. Where that value declines, leakage is structural and accelerates in step with growth; where a layer that becomes more critical as the relationship deepens is present — escrow, dispute resolution, receivables financing, insurance, or compliance recordkeeping — leakage remains marginal and tends to bound itself.
The correspondence at the valuation desk is direct. Intermediation models are priced substantially on the durability of the take rate rather than on volume growth alone, and leakage erodes that rate along two paths at once, first by removing the highest-volume relationships from the measured base and thereby depressing the blended rate, and second by pushing the company to respond with discounting to a pressure it has attributed to competition. A response delivered through price reduction typically fails to slow the escape, because the difficulty lies not in the level of the commission but in the deterioration of what the commission purchases as the relationship matures; the discount succeeds only in impairing unit economics and lengthening the marketing payback period. Some quarters later the company finds itself operating at a lower rate and at an unchanged rate of escape.
On the diligence desk, the picture surfaces through several standard exercises. Pair-level cohort analysis separates repeat behavior after the first transaction from aggregate retention; reconciliation between volume settled through the payment rail and volume reported in management accounts exposes definitional divergence; high-level metadata from the messaging module indicates at which transaction in the sequence contact-detail exchange concentrates. The consequence for transaction structure is that definitions are rewritten before the revenue multiple is negotiated: GMV is narrowed to interactions completed within the platform, the earn-out trigger is tied to pair-level repeat volume retained on-platform rather than to aggregate throughput, the revenue definition is brought within the scope of representations and warranties, and the escrow ratio is calibrated separately for this item. The aggregate effect of those amendments is frequently more determinative of outcome than any concession on headline price.
The second institutional cost of leakage emerges where unit economics are computed. When customer acquisition cost is recovered not against the contribution margin of the first transaction but against an assumed series of repeats, leakage truncates the tail of that series, and realized payback extends materially beyond the modeled figure; the cash-cycle pressure that follows is then commonly interpreted as a slowdown in growth. The same distortion propagates into the incentive architecture, since sales and supply-development teams compensated on the occurrence of the first transaction will optimize the organization for matching, leaving the on-platform life of the relationship unowned by any function. In accounts weighted toward founder-originated relationships the picture sharpens further, because those relationships are by construction the least instrumented, and founder dependency and leakage converge into a single diligence finding.
Structural intervention rests on redefining what the fee purchases rather than on prohibiting user behavior, and it decomposes into four components. The first is price architecture: migrating from a pure transaction commission toward a hybrid structure combining a reduced commission with an access or tooling subscription removes the escalating cost pressure that accompanies relationship maturity. The second is the service layer: payment protection, dispute resolution, receivables financing, insurance, compliance and recordkeeping functions grow more valuable as the relationship enlarges, rendering continued presence economically defensible. The third is measurement discipline: unless the divergence between expected and observed pair-level repeat rates is named as an indicator and reported on a fixed cadence, leakage cannot be managed at all. The fourth is the distribution of incentives and authority: deriving compensation from the cumulative contribution a relationship generates on-platform, rather than from first-transaction occurrence, shifts organizational attention from matching to retention.
BEIREK approaches this as a problem of measurement and contract architecture rather than one of user behavior. The first mechanism established is a value capture map, in which every interaction type the platform produces is written out along three separate columns — where the value originates, on which surface it is measured, and through which line item it is billed — and the rows in which those three columns diverge are the candidates for leakage. The second mechanism is a pair-level ledger, in which the on-platform life of each buyer-seller relationship is tracked alongside payment-rail reconciliation and reported volume, permitting the erosion in take rate to be attributed between competitive pressure and escape rather than assumed. The third is a review cadence: these indicators enter management reporting monthly, and pricing decisions are considered in the same session as the product roadmap rather than in isolation from it.
On the transaction and financing side, the intervention concentrates on definitional discipline. Establishing at the term sheet stage that the revenue definition, the earn-out trigger, and the reporting covenants are constructed on volume retained within the platform closes, in advance, the item that most frequently becomes contested after closing and most frequently produces post-completion adjustment disputes. On the diligence side, structuring the cohort tables, the payment-rail data, and the supply-side interviews so that each corroborates the others prevents the distortion that a reading anchored to a single data source predictably generates, particularly where management reporting and settlement data rest on different definitions of a completed transaction. The output of that work is not a schedule of findings but a proposed structure determining which clauses will carry the negotiation between buyer and seller.
What determines the value of a platform is not the number of matches it originates but the proportion of those matches that find it economically preferable to remain on the measurement surface; and that preference is explained not by user loyalty or switching friction but by the direction in which the value of what the fee purchases moves as the relationship matures.