The cost of deciding on assumptions

Every material authorisation rests on premises nobody put to a vote: the customer pays in sixty days, welders can be hired locally at the model's rate, the grid holds. Those premises do the load-bearing, and because they are not line items they are never priced, tested or dated.

Our position is arithmetical, not moral. Verifying a load-bearing assumption is a priced option: two weeks of a controller's time, a metered log, a countersigned letter from a supplier. It pays out exactly where the authorisation would otherwise have been wrong. Boards never see these options offered, because nobody wrote the premises down as choices.

The instrument follows: an assumptions register travelling with every material authorisation.

Unpriced premises do measurable damage at scale. Flyvbjerg, Holm and Buhl found the decision-stage estimates for 258 transport projects systematically misleading: not incompetent arithmetic, but premises that entered the estimate unexamined.

The institutional response is a rule about premises: UK central government appraisal treats appraiser optimism as a demonstrated defect and requires explicit, empirically based adjustments drawn from comparable completed projects.

Uplifts correct an output; what a board authorises is a set of inputs. RAND's assumption-based planning treats the assumptions as the planning object. Load-bearing is a test a director can apply without a model: if this premise is false, does the decision reverse?

The cheapest surfacing technique is prospective hindsight: state the failure as accomplished, then require reasons. The construct comes from Mitchell, Russo and Pennington (1989), and we grade it down deliberately. The much-quoted claim that it improves identification of reasons by 30 per cent travels through secondary accounts; the reported findings concern outcome certainty, and the first experiment found little effect of temporal perspective. We use it because it is free, not because it is measured.

Our position, graded as interpretation: outcome distributions justify suspicion of premises, an appraisal rule institutionalised that suspicion, and RAND built the register that acts on premises directly. It belongs in the board pack.

Evidence cards

The register earns its price fastest where assumptions travel: a holding board authorising capital for an operating company in Abidjan, Dakar or Lomé imports payment terms, logistics timings, staffing availability and power continuity from the market where the template was drafted, and where each was a fact rather than an assumption. They fail silently: a stated sixty-day term is also a cash-conversion assumption, absorbed by working capital before anyone reads it as a premise failure. We carry no measured corridor statistic on payment behaviour, logistics reliability or grid availability. We looked, verified none to our standard, and would rather say so than borrow a number.

Sources and limitations

Sources and limitations. The outcome distribution rests on Flyvbjerg, Holm and Buhl (2002), a peer-reviewed study of transport infrastructure and not of corporate investment; the appraisal rule rests on HM Treasury's supplementary Green Book guidance on optimism bias, which binds UK public appraisal and no board; the register instrument rests on RAND's assumption-based planning as documented in MR-114-A (1993) and Dewar (2002), a method definition supported by no outcome evaluation we cite. Prospective hindsight is carried at SEG-3 and marked down on purpose: the "30 per cent" improvement figure attached to it in general circulation is a secondary account, not a finding we can trace to the reported results of Mitchell, Russo and Pennington (1989), and we do not use the number. The corridor section deliberately carries no measured statistic on payment behaviour, logistics reliability or power continuity in WAEMU markets — we did not verify one to a specific citable publication at our standard, and the corridor claims are therefore firm interpretation at SEG-2, not measurement. The step from these anchors to the assumptions register is the firm's position, graded SEG-2: consistent with the cited architecture, not a measured result, and we cite no dataset comparing authorisations made with and without a register because we found none meeting our standard. Nothing here is investment advice, and nothing here claims that a tested assumption cannot still be wrong.

1. The CFO: attach an assumptions register to the next material authorisation. Name the load-bearing premises, what each would cost to test, which were tested, and which are carried open and on whose decision.

2. The company secretary: give every open assumption a falsification condition and a named owner in the operating company. An assumption with neither is not being carried; it is being forgotten.

3. The sponsoring executive: mark every premise written in another country as imported, and record the local evidence tested against it. Imported assumptions are the ones nobody examines.