In a plant running three shifts on a single line, producing the same product from the same material lot under the same maintenance regime, a measurable spread in yield between those shifts is an ordinary finding rather than an anomaly. The spread does not originate in the equipment, which is identical, nor in the input, which comes from a common batch, and in most cases not in headcount either, which is held constant by the staffing plan. It originates in the presence, on one shift, of a person who knows what to do when the acoustic signature of a machine changes, and in the absence of that person on the others. Acting on a threshold sense accumulated across years rather than on any clause in the operating procedure, that person makes an intervention the document does not describe; the line keeps running, no scrap is generated, and nothing enters a record. Because nothing enters a record, management reads the spread for years as a difference in discipline or motivation.
The same pattern presents itself far more sharply across locations. Two facilities commissioned by the same vendor, carrying the same line configuration and governed by the same maintenance contract, will nonetheless show ramp-up durations that differ materially; seconding the technical team from the stronger plant to the weaker one narrows the gap partially, and returning that team reopens it within a quarter or two. Filed at the investment committee table under the heading of local labor quality, the difference is discussed as a recruitment and training question, whereas the question that actually deserves the table is through which channel, in what form and at what speed knowledge produced at one site travels to the other. In most groups that channel exists as a box on an organization chart, an owner named in a governance document, and a recurring line on a meeting agenda, without any physical counterpart in the way work is actually recorded.
The mechanism underneath this pattern is known in the operations literature as knowledge-transfer failure — the inability of process knowledge to move from the point where it is produced to the point where it is needed — and its core lies in the fact that knowledge is held in two distinct forms rather than one. The first form is explicit: work instructions, parameter tables, maintenance schedules, quality control sheets. This form is written easily, copied easily, and audited easily, which is precisely why it is the only thing most organizations mean when they speak of knowledge management. The second form is the judgment exercised at the moment of deviation: which departure from nominal is tolerable and which requires stopping the line, which supplier lot compels which offset, which symptom precedes a failure and which merely accompanies it. Being a conditional decision tree rather than a procedural clause, this second form seldom finds its way into writing.
That it is not written is neither negligence nor indifference; under a particular set of conditions it is an entirely rational choice. The operator intervening at the moment of deviation is optimizing for keeping the line up, not for documenting the reasoning behind the intervention, and documentation attempted after the heat of the event has passed loses fidelity while registering, to the person asked to produce it, as an additional and uncompensated workload. Beyond that, knowledge retained in a person generates an implicit bargaining position — indispensability being a lever that is never discussed in institutional language yet shapes behavior powerfully. The difficulty therefore lies not in the shortcut itself but in the persistence of the shortcut after the conditions that justified it have changed: in a single-site, single-shift operation with low turnover, knowledge resting in individuals carries no cost, whereas in a multi-site, multi-shift structure with rising attrition, the identical choice becomes a systematic leak.
Shift handover is the most visible point of that leak, because in most facilities handover has been designed as a ritual for transferring responsibility rather than as a mechanism for transferring knowledge. What the handover sheet carries is state information — which machine is running, which lot was left half-complete, which fault has been logged with maintenance — and state information is an asset that expires within fifteen minutes of being written. What does not cross the boundary is judgment: which parameter the outgoing shift adjusted and on the strength of which observation, which symptom it registered and deliberately elected not to act upon, which supplier lot it has learned to treat with a standing correction. Where this distinction is not grasped, extending the handover window produces no improvement in outcome; information of the same category is simply exchanged for a longer period, at a higher labor cost.
The institutional cost remains invisible for a long time precisely because it never gathers into a single account. A few points of elevation in the scrap rate, a chronic floor beneath which rework hours never fall, a recurring signature in unplanned stoppages, a tail on the commissioning of a new line that runs materially beyond what was budgeted — each of these is reported in its own review, justified in its own terms by its own owner, and none is traced back to a common cause. The same cost accumulates on the labor side as well, where the plainest available indicator is often ignored: a measurable deterioration in yield during the week a key operator takes annual leave demonstrates, without further analysis, that the plant's knowledge architecture contains a person-dependent node. In most companies this comparison is never run, because the underlying data sit in separate reports written for separate audiences.
The accumulation converts into a qualitatively different cost the moment the company sits at a transaction table. Operational diligence conducted by a buy-side team reduces, in substance, to a single question: is this performance repeatable once the current crew changes? The evidence for the answer is not sought in the thickness of the procedure binder, which any organization can produce on request, but in the rationale field of the deviation records, which cannot be reconstructed retrospectively. Where that field is empty, or uniformly populated with formulations such as resolved by operator intervention, the review team attributes the observed performance to a handful of individuals rather than to the institution. The consequences in deal structure follow predictably: earn-out tranches conditioned on the retention of named personnel, key-employee agreements elevated into conditions precedent, an expanded scope of operational representations, and a repeatability discount taken directly on the multiple.
In multi-site groups the same gap distorts capital allocation as well. A yield improvement achieved at one plant compounds across the portfolio to the extent that it can propagate; where it cannot, the identical improvement is financed again, from zero, at every subsequent site, and each instance is approved on its own merits by a committee unaware of the repetition. What prevents the group center from perceiving the difference is a reporting architecture that measures outcomes without carrying the method by which those outcomes were produced — the management pack states the yield achieved at plant A while remaining silent on the setting regime plant A adopted to achieve it. Under those conditions the investment committee reads recurring expenditure as new investment, and the group pays for the same body of knowledge several times over the life of the portfolio.
The mechanism that neutralizes this tendency is neither individual awareness nor additional training, but a change in the moment at which the record is created. The determining design decision is that the rationale for a deviation is captured at the point of intervention, in the language of the person intervening, rather than afterwards in report format, with the recording burden lifted from the operator at the line and assigned to the shift supervisor as a defined duty carrying defined time. Three components accompany that decision. The first is a handover sheet that carries a decision field structurally separated from the state field, recording which parameter was changed and on which observed symptom. The second is a monthly sweep in which recurring deviations are promoted into procedure, converting individual judgment into institutional rule. The third is the explicit mapping of person-dependent nodes: a maintained and reviewed list of which operations only one individual can currently perform.
In the multi-site industrial and infrastructure portfolios BEIREK manages, the intervention we operate binds those three components to a single operating rhythm rather than treating them as separate improvement initiatives. Rebuilding the handover architecture, we separate state information from decision rationale into distinct fields, move the deviation record as close as practicable to the moment of intervention, and locate ownership of that record at shift supervisor level so that the operator's attention remains on the line. We then establish a review session that promotes any rationale recurring beyond a defined threshold into standing procedure, and we consolidate the output of that session at group level rather than plant level, so that a correction learned once is not relearned four times. The person-dependent node map is maintained as a separate record and becomes a standing annex to the investment committee pack, placing the performance of a plant and the portability of that performance on the same table at the same time.
A second line of intervention addresses the dependence of inter-site transfer on the physical movement of people. Where the setting regime developed at one plant is carried to another by seconding an expert, the knowledge remains lodged in the individual and travels back with them when the assignment ends, leaving the receiving site with a temporary improvement and no durable capability. Framing the transfer instead as a validation loop conducted at the receiving plant rather than as a document delivery changes the outcome materially: the incoming regime is tested under local conditions, every departure from the source configuration is logged with its rationale, and the corrected regime returns to the originating plant as a revised standard. To the extent that this bidirectional loop converts knowledge from the property of one facility into a common asset of the group, it also terminates the practice of financing the same improvement repeatedly across the portfolio.
Where a company's operational knowledge weakens measurably the moment its carrier goes on leave, that knowledge does not yet belong to the company; it sits in the reflexes of a small number of individuals rather than anywhere on the balance sheet, and it will be priced accordingly by any counterparty who thinks to test it. The shortest route to locating where that line currently falls does not require a diagnostic program or an external assessment. It requires opening last month's deviation records and examining how much of the rationale field was actually filled, and by whom.