A recurring scene plays out in quarterly marketing budget reviews: the channel performance table is placed on the table, the cost per conversion attributed to mobile inventory appears materially higher than that of desktop search, and by the close of the meeting a portion of spend has been reallocated from the former to the latter. Read on its own terms, the reallocation is entirely defensible; no arithmetic was mishandled and no figure was massaged. Yet within that same table, a user who first encountered the product on a handset, returned two days later by searching the brand name on a desktop machine, and completed the purchase there has been counted as two distinct people — the first touch generating cost, the second booking revenue. When search volume then grows more slowly than forecast in the following quarter, that shortfall is tabled as its own agenda item, discussed on its own merits, and never connected to the allocation decision that preceded it.
A second and less visible pattern surfaces on the reconciliation side, where the conversions claimed collectively by agencies and platform dashboards exceed the order count sitting in the financial system. That variance is habitually treated not as a defect but as noise intrinsic to digital measurement, and the reports continue to circulate unchanged inside each channel owner's deck. For the channel owner this is not even a question of interest; every manager is obliged to report the figure produced by the dashboard assigned to them, and no dashboard has ever been engineered to discount its own contribution. The organisation therefore operates two parallel accounts of the same commercial reality, each internally coherent, neither ever placed alongside the other on a single page.
The mechanism underneath this pattern is the cross-device attribution gap — the inability to resolve a customer's touchpoints across separate devices into one identity, and the consequent assignment of credit to the wrong channel. Its origin is infrastructural rather than behavioural: measurement rests on an identifier scoped to a device, and that identifier cannot travel beyond it. Resolution is available through one of two routes, each with a distinct failure surface. The deterministic path, built on authenticated sessions across both devices, is reliable but achieves broad coverage only in business models where logging in is structurally unavoidable. The probabilistic path, inferring linkage from address, timing and behavioural pattern, extends coverage but rests on a signal set that has narrowed steadily as browser-level restrictions on third-party cookies and mobile operating-system tracking permission frameworks have become the default rather than the exception.
It matters to recognise that this shortcut is not an error; under a specific set of conditions it is entirely functional. For a product whose purchase cycle closes within a single session, whose price point is low and whose consideration window is measured in minutes, device-scoped attribution is both cheap and sufficiently close to reality, and the marginal return on constructing a more elaborate measurement architecture would not cover its cost. The problem lies not in the shortcut itself but in its persistence after the conditions have moved: as basket size rises, as the product begins to demand comparative evaluation, and as the division of labour between mobile discovery and desktop purchase deepens, the same measurement logic converts gradually into a systematic bias. That bias is not randomly distributed — it runs consistently in one direction, favouring whichever channels sit nearest the final step of the conversion.
Organisational structure tends to reinforce this bias rather than correct it. Where channel budgets are assigned to separate owners, each owner is placed in the position of defending the conversions their channel claims, and the contribution of upper-funnel activity that manufactures demand appears at the top of no one's agenda. To the extent that variable compensation is tied to the attribution dashboard, interrogating the measurement architecture becomes individually costly while leaving it unexamined remains individually rewarding. This configuration pushes the decision-maker in a predictable direction — expand what can be measured, contract what cannot — and over any short horizon both movements present themselves as rational stewardship.
The first layer of institutional cost sits in allocation. Every incremental pound directed toward harvesting channels produces a strong near-term conversion figure, because such channels capture demand that already exists; as demand-creating activity contracts, the pool available for capture shrinks with a lag of several quarters, and that contraction reveals itself not in the channel report but in branded search volume, in the share of direct traffic, and in the slow erosion of the new-customer ratio. Where the lag runs long enough, the causal link decays out of institutional memory altogether, and the resulting growth shortfall is written off to competitive intensity, pricing pressure or macro conditions. Channel-level customer acquisition cost, meanwhile, continues to be reported below its true value, the duplication remaining embedded in the denominator.
The second layer surfaces directly in valuation. In an acquisition or investment process, the buy-side commercial diligence workstream attempts to construct the relationship between marketing spend and revenue not from platform dashboards but from the financial record and from cohort-level payback analysis; where those two sources fail to agree, the unexplained variance is typically resolved against the seller rather than in the seller's favour. The consequence rarely appears as an adjustment to the headline multiple. It appears instead as a shift in deal architecture: earn-out triggers indexed to acquisition metrics, an expanded representation and warranty perimeter covering customer data and measurement claims, and a higher escrow proportion released against post-closing verification. Each of these is a single proposition expressed three different ways — that the growth engine has not been shown to be reproducible independently of the agency, the dashboard and the incumbent team.
The third layer is data governance, and it is ordinarily the last to be noticed. Cross-device identity resolution is legitimate only insofar as the consent actually collected extends to the resolution being performed; contracts with third-party identity graph providers frequently define the processor role narrowly, leaving accountability with the controller at the moment of inspection. The question put at a diligence table tends to take a specific form: for which population, on what lawful basis, and from which date can the linkage be evidenced. Where that question can be answered with a versioned consent record and a documented data flow rather than a screenshot, the claim stands; where it cannot, the measurement assertion and the compliance exposure arrive on the table simultaneously, and each makes the other harder to settle.
The mechanism that neutralises this tendency is architectural rather than attentional, and it separates into four components. The first is an identity spine: a record layer, owned by the company and sitting outside every channel dashboard, in which authentication, membership, order and service events are bound to a single durable customer identifier. The second is a measurement hierarchy, in which platform attribution remains at the base as an instrument of operational optimisation, incrementality evidence built through geographic or time-based holdout testing sits above it, and allocation decisions are taken only on the upper tier. The third is financial reconciliation, whereby the aggregate of dashboard-reported conversions is compared against booked orders and revenue on a fixed cadence, with the variance left open until it has been explained. The fourth is the decision record, in which the attribution assumption underlying any reallocation is committed to writing at the moment of proposal rather than at the moment of approval.
The fourth component costs materially less than the first three and generally carries the highest return, for the simple reason that an assumption written down becomes an assumption that can later be tested. The sentence recorded at the point of proposal — that this reallocation rests on last-click cost per conversion as reported by the channel — is the only stable ground on which a review two quarters later can assess the outcome; absent that record, the review collapses into a mind that already knows the result reconstructing the reasoning that preceded it. The same discipline requires that every change to the measurement architecture be dated, since a series break created by a revised identifier policy is exceptionally easy to mistake for a genuine movement in performance, and the mistake tends to be made in whichever direction is more comfortable.
BEIREK's intervention in this area is not the reconstruction of the marketing dashboard but the establishment, at governance level, of the link between the measurement claim and the financial record. In practice this operates through three concrete instruments: a fixed-cadence reconciliation session in which channel-level claims are set against the income statement and the variance is closed only with a stated explanation; a decision ledger in which the attribution assumption behind each reallocation is recorded at proposal and reopened at the subsequent review; and a measurement calendar that fixes in advance which channel enters a holdout test in which period. In a transaction-preparation context the same material is reorganised on data-room logic, so that the lawful basis for identity resolution, the version history of the measurement methodology and the underlying source of the cohort payback analysis stand documented before the buy-side questions are asked rather than after.
The cross-device attribution gap is less a technical deficiency than a timing mismatch — a measurement infrastructure designed for one commercial era and still being operated in another — and mismatches of this kind rarely announce themselves as failures. They present as reporting that is consistently, unremarkably optimistic. The genuine test of any claim a company makes about its marketing efficiency is not what the dashboard displays, but whether the claim survives with the dashboard switched off.