In an investment committee session, a single presentation page that evaluates three structurally dissimilar businesses against one common threshold will, more often than not, pass without objection. The same discount rate, the same payback expectation, the same margin band — notwithstanding that the first business is a contracted operation with predictable cash conversion, the second a fabrication line whose economics move entirely with order tempo, and the third a development portfolio generating no revenue at all. Consolidating the three onto one page is presented as an act of editorial discipline, and institutional courtesy accepts it as such; what is lost on that page, however, is not formatting but information. A year later, when the committee asks which of the three met expectations, the answer arrives as a single blended figure, and the line sitting below it is defended on portfolio logic rather than on its own.
A comparable pattern surfaces at the pricing desk. Where a sales organisation approaches a customer base that diverges materially in account size, delivery complexity and payment behaviour with one list price and one discount authority, the pricing decision itself has stopped carrying information. What the representative then performs in each negotiation is a private compensation for the distinction the framework cannot make — a compensation that frequently works, but which, having been recorded nowhere, never becomes repeatable, and which departs with the representative.
The mechanism beneath this behaviour is **underfitting** — a decision framework that remains too coarse to capture the meaningful patterns present in the reality it operates on. Model here does not denote a statistical equation; it denotes the entire apparatus through which an institution generates decisions: the classification scheme, the approval thresholds, the reporting granularity, the incentive formula, the level at which a given line rolls up on the management dashboard. When that apparatus is built too coarsely, the resulting error is not random. It drifts in a determinate direction, at a consistent magnitude, with regularity. Random error offsets itself in the aggregate; systematic error compounds — and it compounds precisely because the framework producing it stays fixed while the reality it describes does not.
The simplification warrants being understood as a shortcut rather than as a lapse. Producing distinction carries cost: every additional segment implies additional data collection, additional reconciliation, additional debate and an additional approval cycle. Under identifiable conditions the coarse framework is the correct choice — where the economic logic of the business lines genuinely converges, where decision frequency is high and individual decision size low, or where the data infrastructure is not yet mature enough to carry fine granularity reliably. The difficulty lies not in the shortcut itself but in the shortcut outliving its conditions: when the firm moves from one customer profile to three, extends from a single geography into two jurisdictions, or converts from a single product line to a mixed portfolio, the simplification that was correct yesterday becomes today's systematic blindness.
The more insidious property of underfitting is that its output looks consistent. A framework built with excessive complexity responds to noise, its estimates oscillate, and that oscillation exposes it; a framework built too simply errs stably. Stability, in an institutional setting, reads as reliability. Where a budget model drifts in the same direction three consecutive years, the drift is attributed to external conditions rather than to the granularity of the model itself, since the model has by then ceased to be an object available for interrogation and has settled into the substrate of the planning process.
The institutional cost surfaces first as cross-subsidy. Within a mixed portfolio managed to a single margin target, the line falling below target is not closed but fed from the line clearing it — and because that transfer is nowhere recorded as a decision, no one is ever identified as having made it. The predictable consequence is a low-return line sustained structurally and a high-return line whose growth capital is systematically withheld. Over successive years the centre of gravity of the portfolio migrates toward the wrong side, and it migrates not in a strategy document but in the silent arithmetic of capital allocation.
A second surface lies in working capital and inventory. Where inventory items with materially different turnover velocities are managed against a single coverage ratio, the fast-moving item runs in chronic shortage while the slow-moving item accumulates chronic excess; because the aggregate inventory level remains within budget, nothing registers on the dashboard. The same logic governs collections: with days sales outstanding sitting on target in aggregate, the condition in which terms have quietly lengthened across one portion of the customer base while early payment has been purchased through price concession across another remains concealed beneath the mean. An average, in institutional reporting, is a summarising instrument and a concealing instrument in equal measure.
The third and most expensive surface appears at the diligence table. When a buy-side analyst requests revenue disaggregated by segment, customer cohort and product line, and the company can produce only a consolidated statement, the constraint is usually not absent data but data captured at a granularity too coarse to be decomposed after the fact. Two consequences follow predictably from that position: first, a valuation multiple calibrated downward because the recurrence of revenue that cannot be attributed cannot be demonstrated; second, the undemonstrated portion pushed into an earn-out structure or a post-closing measurement, leaving a share of consideration at seller risk. What determines a company's valuation is frequently not performance itself but the ability to show where performance originates independently of the founder — and coarse granularity makes precisely that showing unavailable.
The mechanism that neutralises this tendency is architectural rather than attentional; a better analyst does not resolve it. Four components separate out. The first is a disaggregation requirement — for any proposal above a defined size threshold, at least one breakdown must accompany the single consolidated figure. The second is residual analysis — reading the sign of the variance between forecast and actual in order to distinguish a framework that is too simple from one that is too elaborate, since persistent unidirectional drift indicates coarseness rather than complexity. The third is a segmentation threshold — defining in advance the conditions under which a new segment is opened, absent which the decision is renegotiated on each occasion and the easier position prevails in the negotiation. The fourth is a review cadence — reopening the classification scheme at fixed calendar intervals rather than only when the business structure visibly changes.
The intervention BEIREK operates across complex, capital-intensive projects embeds these four components inside project governance. On capital allocation and pricing decisions the decision record is opened at the moment of proposal rather than at the moment of approval, and the sponsor of the proposal is asked to state in writing the granularity at which the underlying figure was produced and which distinction was deliberately collapsed. Simplification thereby becomes visible as an assumption and remains contestable afterwards; an unrecorded simplification, by contrast, converts into a position requiring defence the moment variance emerges.
The second layer is a residual record within project and portfolio reporting that tracks the sign of variance rather than its magnitude alone. The gap between actual and modelled outcome is monitored directionally across periods, and drift repeating in the same direction triggers a reopening of the classification scheme rather than an explanation grounded in external conditions. The same discipline applies in diligence preparation: installing the breakdown the buy-side will request into the company's own reporting architecture before a process begins is both cheaper and more credible than the hurried decomposition assembled during the data room phase, since granularity constructed retrospectively is typically unauditable, and counterparties price that condition accordingly.
An institution's decision framework remains too simple far more often as the extended life of a once-correct choice than as an act of neglect. The operative question is not whether the framework is sufficiently elaborate today, but whether the institution possesses a mechanism capable of registering the moment the framework stops being adequate; a framework that is too simple never announces its inadequacy through noise, drifting instead, quietly and consistently, in the same direction.