A recurring scene plays out in quarterly review meetings. Funnel conversion has run materially below target for a third consecutive quarter, customer acquisition cost has multiplied over the same period, the average sales cycle continues to lengthen — and the decision reached at the end of the presentation is not to change the model but to run the same model for one more quarter behind a larger marketing budget. No one in the room asserts that the model is working; what is being asserted, rather, is that the model has not yet been tested adequately. The two propositions look almost identical, yet the first requires evidence and the second does not.
An asymmetry operates inside that same room. Were a business model proposed for the first time today while carrying the performance observed over the preceding three quarters, clearing the investment committee threshold would in all likelihood be impossible; the identical performance, attached to a model already in force, is accepted as grounds for continuation. What determines the direction of the decision is not the content of the data but the side on which the burden of proof happens to sit: a new proposal must establish itself, whereas an incumbent model is abandoned only once it has been affirmatively disproven. This allocation appears in no governance document, and yet in practice it functions as the most binding rule in the building.
The tendency has a name — pivot resistance, the delay in redirection produced by emotional and organizational attachment to an existing business model — and its mechanics assemble in three layers. The first is sunk cost: the time, capital, and reputation already committed to the model migrate into a forward-looking calculation as a backward-looking line item. The second is escalation of commitment, in which each fresh approval to continue, by retroactively justifying the approval before it, raises the price of reversal. The third and quietest layer is identity: the story the founder tells outside, the thesis sentence in the investor deck, and the meaning the company offers its own employees are all constituted by the model, so that questioning the model means questioning three narratives simultaneously.
It is worth recognizing that this resistance is not an error. An organization that answers every adverse early signal by redirecting never climbs a single learning curve to its top; supplier relationships reset with each turn, the sales team never deepens in any segment, and the product roadmap degrades into a sum of unconnected experiments. Separating market signal from noise typically requires several full sales cycles, and remaining committed to the model across that interval is a rational choice. The difficulty lies not in the heuristic itself but in the heuristic persisting after the condition that validated it has dissolved — which is to say, not in the presence of resistance but in the absence of any defined measure governing its duration.
Once that condition changes, the first place the cost appears is not the income statement but the cash conversion cycle. Every quarterly decision to continue ties working capital to assets specific to that model: inventory configured for a particular customer profile, logistics agreements structured around a single distribution channel, tooling and equipment procured for one use case, a sales organization hired against one segment. What these assets share is thin secondary market value, which means a change of direction requires writing down most of their carrying amount. The balance-sheet counterpart of the delay is therefore usually concealed not in an impairment line but in an inventory position acquired a year earlier for which turnover still cannot be computed.
The second cost surfaces in the allocation of management attention. The longer examination of the model is postponed, the further internal debate drifts from strategy toward execution: because the model is treated as settled, the explanation for poor outcomes must be sought in the quality of implementation, and that search hardens into reciprocal attribution among teams. Sales points to gaps in the product; the product function points to pricing; pricing points to the target segment. When this loop runs for several quarters, the observed increase in personnel turnover generally begins among the most senior and most mobile people on the roster — precisely those capable of diagnosing first that the model, not the execution, is what fails.
The third cost materializes at the valuation table. An investor or acquirer does not treat a prior change of business model as a problem in itself; redirection is ordinary in venture valuation. What registers as a problem is the absence of any record of how the change was decided. The question posed during diligence is usually a narrow one: when, and against which measure, was it determined that this model was not working. Where that question cannot be answered with a board minute, a threshold written in advance, or a dated decision memorandum, the inference drawn on the buy side concerns not the weakness of the model but the fact that the decision mechanism consisted of founder intuition — and that inference operates on structure rather than on price, appearing as a heavier earn-out, a longer schedule of conditions precedent, and a wider escrow.
This tendency cannot be managed through individual awareness, for the simple reason that awareness is weakest precisely in the person who owns the model. What can be managed is decision architecture, and that architecture has four components. The first is writing the falsification threshold at the moment of proposal: in the meeting where the model is approved, the metric and the level below which the model will be reopened are entered into the minutes with a number and a date attached. The second is periodically reversing the burden of proof, so that at least once a year the continuation of the existing model is defended exactly as a new proposal would be defended, on the basis not of its history but of the cash it is expected to generate in the period ahead.
The third component attaches the counter-argument role to a position rather than to a person. Whoever is charged with presenting the weaknesses of the model must know that discharging the assignment carries no career cost; otherwise the role exists formally and stands empty functionally. The fourth is sorting decisions by reversibility: tooling investment, long-dated leases, single-source supply agreements, and headcount expansion are one-way doors, and deferring them while the model remains open to reconsideration lowers the price of optionality directly. The shared logic across all four components is straightforward — a threshold defined after the fact will calibrate, predictably, to ratify the performance already on the table.
The mechanism BEIREK establishes in capital-intensive projects and portfolio companies is the operational expression of that logic. For every allocation decision, a decision record is opened at the moment of proposal rather than at the moment of approval; the record carries the three to five assumptions on which the decision rests, the indicator against which each assumption will be measured, and the threshold that will trigger reopening of the model. The record is archived independently of the individual who made the decision, and each subsequent review reads from that document rather than reconstructing one. Debate then proceeds over whether a previously agreed threshold has been crossed, not over how the outcome ought to be interpreted.
Two distinct cadences run on top of this. The quarterly review examines execution — schedule, cost, quality, procurement. A separate annual session examines only the model and admits no execution performance to its agenda whatsoever, since good execution discussed in the same room will reliably conceal a bad model. The practical consequence of the separation is that a change of direction becomes a decision available in a calendared session rather than in a crisis; and the effect on the cash position of a pivot taken outside of crisis conditions is typically an order of magnitude smaller.
Abandoning a business model does not require conceding that the model was wrong; it requires conceding only that the conditions which made it right no longer hold. The distinction looks minor, yet in terms of the institutional weight it carries it is decisive: the first passes judgment on one person's reasoning, while the second acknowledges that an assumption has a shelf life. The task of decision architecture is to determine, in advance, which of those two sentences the company will eventually be obliged to say.