The weekly operations meeting opens with the late order list, and each line carries its own explanation: a supplier shipment moved, a batch was held in quality control, a document was missing at customs, a customer requested a late revision to the drawing. The composition of the list changes from week to week while its length remains remarkably stable, and that stability never reaches the agenda, because as long as every line is individually explainable, the pattern formed by the sum of the lines stays invisible. The question the meeting does not ask is this: who produced the date on this order, against what load profile, and at what moment.
In most manufacturing and project businesses, the answer originates in the same place. At the point of order intake the customer asks for a date, the person on the commercial side reaches for the standard lead time embedded in the catalogue or in habit, shortens it somewhat according to the temperature of the negotiation, and the date is born; planning learns of it only after the order lands in the system. At that moment the date is not the output of a capacity calculation but an instrument for closing the sale, which means that whether it will later hold is nobody’s responsibility at the point of its creation. Three independent sales representatives can commit the same line for the same day within a single week, and the collision becomes visible only when the production schedule is drawn.
At this point delivery delay — the failure of an order to reach the customer on the date committed — stops being an event reducible to a single root cause and becomes a composite formed by three superimposed layers: how the promise was produced, how buffer time is consumed along the line, and how quickly information about slippage travels upward. These three layers sit with different functions, run on different rhythms, and never appear side by side in any standard report, which is why the delay is read each time as a failure of the last layer, namely execution. Execution, however, is attempting to honor a commitment that was handed to it, having played no part whatsoever in producing that commitment.
The mechanism feeding the first layer is planning fallacy — estimating duration against an idealized run free of disruption rather than against the realized durations of comparable past work — combined with anchoring, in which the standard lead time behaves as a fixed reference and pulls every new estimate toward itself. The anchor is a reasonable approximation while the plant runs at normal load; the difficulty is that it stays in place when the load profile shifts, when the order mix grows heavier, or when a maintenance window opens on a critical machine. Add overconfidence to that, and the estimate deviates systematically in one direction: delays do not cancel one another out, they accumulate, because early completion is rare and, when it occurs, the gain is not passed downstream.
The second layer concerns how buffer time is distributed. Each stage holds its own safety margin internally, and that margin behaves asymmetrically in two directions: a unit that finishes early does not release its output early, since early delivery invites a shorter allocation next cycle, whereas a unit that runs late transfers the entire loss downstream. Time gained thus disappears inside the system while time lost accumulates in full, and the aggregate buffer capacity of the line, though theoretically generous, protects no individual order in practice. At the level of the individual unit this behavior is entirely rational; what is irrational is a design that permits buffer to be distributed stage by stage in the first place.
The third layer depends on the direction of information flow. That a job has slipped is usually known at the lowest level of the line weeks in advance; it reaches management only when the customer asks or when the shipping window is missed. Bad news travels upward slowly, because each intermediate level first attempts its own recovery, and throughout that attempt the delay does not formally exist. The operational consequence is that the intervention window — the interval in which low-cost options such as material substitution, an alternate supplier, or a sequence change remain open — closes quietly, so that by the time intervention finally begins, only the expensive options remain.
The first and most visible line of institutional cost is the liquidated damages provision, though the substantive cost is rarely found there. The LD cap in most supply contracts is set at a limited percentage of contract value, and that cap does not represent the true burden of lateness; the true burden sits in expedited freight, in the batch shifted to air cargo, in weekend overtime, in the packaging and customs charges doubled by a split shipment, and in the capacity stolen from the schedule of the following job. These items are collected in the accounts under logistics expense, personnel expense, or manufacturing overhead; unless cost is tracked at the order level, no report reveals what a given delivery actually cost, and gross margin erosion is attributed to price pressure.
A second cost accumulates in working capital. Undelivered finished goods sit in inventory, no invoice has been issued, and consequently nothing appears in the receivables ageing table; the cash conversion cycle lengthens while the source of the lengthening is read as collection performance. A third cost forms on the counterparty side and returns: a customer who does not trust the promised date raises its own safety stock, orders earlier and larger than actual need, and this behavior distorts the demand signal, making production planning less accurate still. Delay therefore establishes a loop that feeds itself, with the company allocating progressively more capacity against a progressively less reliable forecast.
The valuation counterpart of that loop surfaces at the diligence table through a single question: what is the OTIF rate, and against which date is it measured. A meaningful share of companies either do not measure it at all or measure it against the most recently revised date; the latter renders the metric void, since a target repositioned each time it is missed will by definition produce high performance. The buyer typically discovers this through a sample reconciliation of order records against dispatch documents, and where the finding combines with customer concentration, the response comes through structure rather than price: a performance-linked earn-out, expanded representations and warranties, a higher escrow ratio. Delivery performance carried by the founder’s relationship management does not receive full credit in valuation unless it can be shown to be repeatable independently of the founder.
The mechanism that neutralizes this tendency is not individual discipline but a reconstruction of the commitment architecture, and it has four separable components. The first is changing the source of the promise: the date is generated from a capacity and material availability model, and the commercial function may depart from it only through a named approval authority and an exception whose justification is recorded. The second is freezing the original commitment date and logging every revision together with its date, its rationale, and the function requesting it; revision itself is not prohibited, revision without a trace is. The third is measuring delivery performance against that frozen date and classifying delays by root cause — commitment error, supply, capacity collision, customer-driven revision. The fourth is withdrawing buffer from the individual stages, consolidating it at line level, and tracking its consumption weekly.
BEIREK’s intervention in a structure of this kind begins not with replacing the planning software but with capturing the moment the commitment is produced: at order entry, the model or exception the date came from, the person who approved it, and the line load prevailing at that moment are collected into a single commitment record, and that record is made non-rewritable thereafter. A weekly rhythm is built on top of it, one in which late orders are not discussed case by case; what is discussed is the root cause distribution of delays and the buffer consumption curve, because what is manageable is the distribution rather than the individual instance. On the contractual side, the gap between the LD cap and realized expediting cost is quantified, and the delivery commitment, the acceptance criteria, and the schedule-suspending effect of customer-driven revisions are recalibrated against that gap.
The delivery reliability of a company is read less from how fast its line runs than from how it produces the dates it gives and from whom it tells, and when, once a date is missed; where neither mechanism is on the record, improvement effort lands on execution every time, and finds nothing there to work with.