When a technical question surfaces in the weekly progress meeting of a capital-intensive project, the identity of the person who will answer it is usually predicted not by consulting the organisation chart but by recalling who answered the same question last month. The chart shows responsibility distributed across disciplines, yet everyone in the room knows, without anyone saying so, whether the real addressee is the contract manager or the senior engineer who has been tracking the same interface problem for two years. That knowledge derives not from formal authority but from the accumulation of past answers, and this accumulation is the most valuable record the organisation keeps nowhere.
The same pattern becomes visible in calendars. When the quarterly meeting hours of a portfolio are aggregated by person rather than by discipline, a disproportionate share of attendance time is typically found concentrated in a handful of names, and those names commonly sit not at the top of the hierarchy but in the middle layer, where institutional memory runs longest. The same three or four names recur in approval chains, in copy fields, in the justification memorandum accompanying a contested change order, and in the routing address for RFIs arriving from site. They occupy those positions precisely because they produce the highest contribution, and the difficulty originates there.
This concentration is termed **collaboration overload** — the accumulation of collaborative demand at an organisation's most capable nodes, independent of the capacity those nodes actually hold. Its mechanism is straightforward and self-reinforcing: whoever answers a question quickly and accurately raises the probability that the next question will arrive at the same address, each successful response reinforcing the routing of future demand. The decisive asymmetry is this — requesting costs the requester almost nothing, a message or a meeting invitation, whereas the cost of responding remains wholly with the responder. Like any externality left unpriced, such demand is over-consumed until a corrective mechanism is installed.
This tendency is not an error; under certain conditions it is the cheapest coordination instrument an organisation possesses. In the pre-FID development phase of a project, where uncertainty is high and the return on writing procedure is low, the informal network operates both faster and more accurately than any formal process, since the nodes of that network carry within themselves the knowledge of which information holds in which context. The difficulty lies not in the shortcut but in the shortcut persisting after the conditions have changed. As a portfolio moves from two projects to twelve, demand volume rises roughly in proportion while a single node's capacity to absorb it does not, and waiting time lengthens disproportionately as the saturation threshold is approached.
The demand reaching such a node is also not of a single kind, and that distinction is rarely drawn. Part of it constitutes a genuine information request, where technical or historical data held only by that person is required. Part of it constitutes a decision request, where the answer is already known and what is being sought is the exercise of a decision right. The third portion, frequently the largest by volume, constitutes a legitimacy request: the person who will decide has already decided, and the name is added to the chain so that the decision becomes uncontestable within the firm. This third layer consumes capacity while producing no information whatsoever, which makes it the easiest to reduce and the hardest to measure.
Contrary to expectation, the first institutional cost of concentration does not appear in output quality. A saturated node generally continues to complete the work that reaches it at the same standard; what changes is the interval preceding the start of that work. The number of days an interface approval, a supplier deviation request, or a design revision spends waiting on the critical path never enters the progress report, since the progress report measures completed items rather than the waiting queue behind them. Schedule slippage is therefore recognised late in most projects, and once recognised it is attributed to a discipline rather than to a person — although the source is not the discipline's capacity but the queue standing at a single address within it.
The second cost surfaces at the transaction table. When a company or a portfolio is prepared for sale, the buyer's diligence team establishes within a few sessions where the knowledge actually resides, attending less to which questions the data room answers than to which questions can be answered only by speaking with a particular individual. Where single-node dependency is identified, the response is seldom written as a discount to the headline price; it is typically embedded in structure — a key-man clause in the credit agreement, a binding post-closing transition service undertaking, an extended escrow period, or a broadened indemnity scope. Cost embedded in structure proves more durable than cost reflected in price, since it continues to serve as the reference point in every subsequent negotiation.
The third cost falls on the personnel side and is recognised latest of all. The saturated node's own substantive production — design review, contract analysis, model calibration — is pushed to the margins of the day to make room for the inbound demand stream, and those marginal hours are precisely the hours in which error density peaks. Every quarter this configuration persists, the probability of departure among the firm's highest-contributing staff rises, and the cost of such a departure appears not in the payroll line but in the reconstruction of the routing network itself. Even where the successor is technically fully adequate, learning who asks which question in which context can take as long as a full budget cycle.
This tendency is not managed through individual awareness or time-management discipline; even where a person learns to decline, the demand stream does not disappear but merely redistributes itself or goes unanswered. The neutralising intervention sits in institutional architecture and has four separable components: the separation of decision rights, meaning that for every recurring decision type the roles that decide, that are consulted, and that are merely informed are named and periodically audited; the visibility of the request queue, meaning that questions travel through a recorded routing line rather than a personal channel; the naming of deputies, meaning that a second signature is pre-assigned for each critical node and actually exercised at regular intervals; and a codification trigger, meaning that a question recurring above a defined threshold is converted into a document rather than answered again.
In projects BEIREK manages, this architecture is operated through three registers established at the outset. The first is the decision register: alongside the substance of the decision, the role that will take it, the input set it will rest on, and the date by which it is due are written in advance, so that approval traffic circulating as a legitimacy request falls away at source, the decision right being already defined. The second is the interface and RFI routing matrix: for every question arriving from site, from a supplier, or from a lender, the primary address, the alternate address, and the response window form part of the contract management documentation rather than of a personal relationship. The third is the queue register — an ageing list of outstanding requests holds a fixed place on the weekly agenda, and the quantity measured is the waiting time of requests, not the utilisation rate of individuals.
The combined effect of these three registers is less the equalisation of load than the disclosure of where the load originates. When a node's queue lengthens over several consecutive weeks, the cause is typically not that the individual has slowed but that a domain in which decision rights remain undefined is being resolved through that individual, and the queue register exposes that undefined domain together with its address. Items carrying single-node dependency on the project's critical path are separately flagged, and for those items deputisation is operated as a procedure tested at regular intervals rather than as a contingency expected to occur. The fragility of institutional memory becomes measurable only where the substitute has in fact been used.
Whether the confidence an organisation places in its most capable people is the same thing as the volume of demand accumulating on those people is a question most organisations never pose; yet the difference between the two is itself the difference that determines whether a portfolio can be repeated independently of its founder or of a few key names.