Why capex forecasts are inaccurate, and why the miss goes unmeasured

Capex forecast accuracy is the measured difference between what a company forecast it would spend on its capital investments in a period and what it actually spent in that period. It applies to the spending forecast inside a live capital programme, the figure a group controller collects each quarter. It is what tells a CFO how much weight the capex guidance given to the market can carry.

Capex forecasts become inaccurate for several reasons, but optimistic estimation receives most of the attention. Teams estimate low, permits arrive late and suppliers slip. In its 2022 work on capital expenditure management, McKinsey found overruns against original cost and schedule estimates frequently passing 50 percent, across industries, public sector and private alike.

A less visible problem is that controllers rarely measure the error at all. Each new forecast replaces the one before it: the wrong figure disappears, and the evidence of how wrong it was goes with it. Most organisations discuss forecast accuracy. Very few calculate it.

Part of the optimism is psychology: teams estimate what a project should cost if it goes as planned, rather than what comparable projects actually cost. Part of it is incentive, since a lower figure stands a better chance of getting through the capex approval process. Either way it explains one estimate coming in low. It says nothing about why the same group comes in low again next year, and the year after.

In more than thirty years of working inside capex processes at industrial groups, we have yet to meet one that could produce last year’s quarterly forecasts without rebuilding them by hand.

Operational people usually resist the idea that the damage is done at the estimate, and fairly so. Anyone who has run a capital project has watched costs move during execution: the contractor variation, the extra crane weeks, the scope that grew because the old line was in worse condition than the inspection suggested.

Execution feels like where money escapes because execution is where money is watched. Sanction the project and the machinery starts: monthly cost reports, commitment tracking, variation orders, a steering group asking awkward questions. An overrun in execution and tracking gets seen, escalated and eventually explained.

The estimate gets none of that. Someone produces it over a few weeks, under time pressure, with engineering that is nowhere near complete. It gets challenged in the approval meeting, adjusted and approved. From that moment it becomes the baseline, the figure everything else is measured against, and the one figure no one reopens. So the machinery spends three years measuring deviation from a number that was wrong on the day it was set. Which is also why most capex software is built around execution: that is where the visible pain is.

In our study of 20 European capex organisations, 14 of the 20 ran their capex calculations on standardised in-house Excel templates. For several of them the template was the request form.

Every forecast cycle produces a version, and the version it replaces is rarely kept.

A new file replaces the old file, or a new column replaces the old column. Answering how accurate last year’s forecasts were then means reconstruction: find the old file, work out its scope, pull the matching actuals, align the currencies. For one investment that is an afternoon. For a portfolio of two hundred it stays on the list of things the team would do with more capacity, and after a couple of years the controller stops asking. The same study lists this among the most common mistakes in capital allocation: follow-up that is discussion-based rather than analytical, so the lessons are never gathered.

Capex software vendors, Weissr included, agree that the capex process needs a feedback loop, and in this market the feedback loop means the post-completion review. Compare delivered cost, schedule and benefit against what was approved, record what was learnt, carry it forward. It is sound practice, and it has earned its reputation.

The limitation is the interval. A post-completion review closes the loop once, at the end, and in practice on material projects only, since most groups staff the review function for the big ones and leave the rest. It reports after the capital is committed. It can establish that a project finished in 2024 was estimated 14 percent low. That may help when a comparable project next comes up for sanction, but it is too late to correct the forecast itself. Running a full review every quarter is not practical for most groups, and the finding would still arrive after the spend.

If the point of the loop is to correct estimating, it has to close on the rhythm the estimating runs on: every period, on every investment, regardless of size.

Weissr Capex closes the loop at that interval. Post-completion review runs as it does elsewhere in this market. Underneath it, the forecast that stood at each period close is kept rather than overwritten by the next cycle. Deviation is recorded when the period closes, and the remaining forecast is recalculated against it. These are the mechanics behind capex forecasting and financial control.

One change in the mechanics, and the series survives. The forecast that stood in February is still there in September, and the distance between them is a figure rather than an impression. At that point a capex forecast stops being an expression of intent and becomes a measurement with a known error. That version of a forecast, a CFO can put in front of a board.