In a multi-line group, the way discussion time distributes across a board presentation reveals more about the decision architecture than the financial statements themselves. Consolidated gross margin sits as a single line on the opening page while the margin table for six distinct business lines occupies the appendix, and most of the available hours are spent on why that one opening figure rose or fell against the prior period. Yet no activity corresponds to that number; neither the assembly line, nor the service organisation, nor the project development unit operates at that margin. What is being debated is a composite describing no unit that actually exists, and it is precisely this property that makes agreement on it so easy to reach.
The same pattern recurs with regularity at smaller scales: a single customer acquisition cost figure, a single average collection period, a single contingency percentage applied across all work packages, one labour productivity coefficient covering every geography, one performance score covering every supplier. The diagnostic signal lies in the form of the answer given when disaggregation is requested; where the difference between subgroups is met with the phrase "broadly similar" rather than a measured distribution, the difference has not been calculated because it was never asked about. Reporting the mean of a quantity does not establish that the distribution around that mean is known, and decisions are usually sensitive not to the mean but to the tails.
The name for this pattern is aggregation bias — the tendency to represent groups that behave in structurally different ways through a single model, a single ratio, or a single set of assumptions. The tendency itself is not an error. Aggregation manufactures comparability, makes a reporting cadence possible, renders incentive systems administrable, and lowers the cost of deciding. Were the only way to run a group to read each unit separately, governance would not scale. Aggregation is, at bottom, a compression operation, and like any compression it is lossless only when the compressed items share a similar production logic. The problem does not reside in the shortcut but in the continued use of the shortcut after the subgroup logics have separated.
The loss takes two typical forms. The first is loss of direction: a relationship observed at the consolidated level — a positive association between order size and margin, for instance — may weaken, vanish, or invert once the data is read by subgroup, and inferring unit behaviour from aggregated data is the direct corporate analogue of the inferential error known as the ecological fallacy. The second is weight drift: every aggregated ratio being a weighted composite of subgroup ratios, the composite moves whenever the weights move, even where no subgroup's performance has changed at all. Reading a mix effect as a performance effect is the costliest manifestation of the tendency, since it prompts either the remediation of a problem that does not exist or the celebration of one that does.
The persistence of the tendency is organisational as much as cognitive. The chart of accounts, the cost centre architecture configured in the ERP, and the periodic reporting template together determine in advance which decompositions are effortless and which require bespoke work; the effortless one becomes the default. Layered onto this is the fact that disaggregation renders accountability visible — a weak line can shelter behind a consolidated figure, whereas a segment table names it. Where the owner of an aggregated indicator and the owner of the aggregation decision are the same person, the probability of that decision being reopened falls structurally. This is a design condition rather than a matter of intent; the reporting architecture determines who is obliged to see what.
The most visible surface on which the institutional cost lands is the valuation table. Where a mixed-structure group is discussed through a single multiple, one of the first things the buy side does in diligence is separate revenue and margin by line; once separated, the line generating repeatable cash and the line carrying optionality are priced differently, and the blended multiple typically comes under downward pressure. A seller who has never maintained a segment-level ledger is then obliged to negotiate without its own data, and where the counterparty's decomposition is the only decomposition on the table, the discussion proceeds on the counterparty's assumptions. Earn-out structures tied to the performance of a line the seller cannot document in isolation generally originate in the same gap.
On the financing side, the cost accumulates in the space between where cash physically sits and where the covenant is measured. A consolidated DSCR may present at a comfortable level while a material share of cash flow remains trapped behind a distribution test at the project level; the structural subordination between an obligation at the holding tier and free cash at the asset tier is simply not described by the consolidated ratio. Similarly, a model carrying one availability or one degradation assumption across an entire portfolio forces divergent climates, technologies, and operating contract regimes into identical behaviour, and even where such a model produces a defensible total, deviation accumulates in the individual periods in which debt service actually falls due.
The same mechanic operates more quietly along the operational line. A single contingency percentage applied across all work packages underfunds the high-variance package and overfunds the low-variance one; total contingency appears adequate while the package where delay actually originates is left short on its own terms, and the schedule impact propagates to the whole. One labour productivity coefficient spanning geographies systematically misplaces the programme in regions where permitting regimes and site conditions diverge. Where supplier performance is monitored through a single average score, deterioration in a critical single-sourced item stays invisible inside the healthy average produced by a broad supplier base — until the delivery window is missed.
The tendency is governed not by individual attention but by an architecture that removes aggregation from the category of default and converts it into a recorded decision. Four components form the spine of that architecture: (a) deriving the segmentation criterion from the decision to be taken rather than from reporting convenience — customer class for a pricing decision, cash conversion cycle for a capital allocation decision, the legal tier where cash physically rests for a debt structuring decision; (b) defining a variance threshold and withholding decision authority from the consolidated figure whenever the dispersion across subgroups exceeds it; (c) carrying a weight table alongside every aggregated ratio, so that mix movement can be read separately from performance movement; (d) a direction test, periodically examining whether the relationship observed at the consolidated level holds in the same direction across the principal subgroups.
On projects BEIREK manages, the build direction of the model is fixed for this reason: the portfolio- or group-level figure is produced as the composite of asset- and package-level items constructed from the bottom up, never through an allocation running the other way. The assumption register holds each aggregation decision as a separate entry — which units were combined under a single assumption and on what grounds, how large the difference between them was at the moment of combination, and what threshold that difference must cross for the separation to be reopened. Contingency is calibrated package by package, permitting and interconnection assumptions are held separately by jurisdiction, and offtake and supplier concentration are tracked not through an average but through the share carried by the single largest counterparty.
On the governance side, the intervention is placed in the ordering of the information pack: in the decision document the segment table precedes the consolidated table, and the consolidated figure is presented only alongside its weight distribution. The recurring question at gate reviews is not what the number is but which combinations produced it, and where that question is asked on a fixed cadence, the request for disaggregation ceases to be a personal objection and becomes a routine step in the process — which is precisely what makes it easy to ask. Where disaggregation is genuinely costly, the decision may well be not to disaggregate, provided that this is recorded as a choice rather than continued as a habit.
How much of what an institution knows about itself describes an actual operating unit, and how much of it is merely a composite, is in most cases first measured on the counterparty's diligence list; every disaggregation question left unanswered until that list arrives is a question whose answer will be supplied by someone else, on someone else's assumptions.