It starts with a decision, not a chart
Nobody needs another version of the forecast. What is needed is an answer to the decision actually on the table: buy in, buy more, extend the credit line, sign off the capex programme, or ask for a change before the money moves.
Brenwick starts there. The decision fixes everything downstream — which date matters, which parts of the business have to be modelled in detail, and which assumptions are worth arguing about. A plan examined without a decision in mind produces a document that is interesting and unusable.
Rebuilding how the company makes cash
A forecast is an output. The mechanics that produced it are what can be examined, so those get rebuilt: how demand arrives, what converts into an order, what comes back next year, what it is priced at, when revenue is recognised and when it is actually paid, what each hire costs from the month they start, which commitments are dated and which are intentions, and which financing events the plan assumes will land.
The reconstruction is company-specific. A subscription business, an engineering shop working to order, a manufacturer and a distributor do not run out of money in the same place, and forcing them through one template hides the part worth looking at.
What comes out of this step is not a better forecast. It is a connected model, one where the pieces move each other: a slow quarter of order intake shows up two quarters later as thinner revenue, as a credit line drawn earlier than planned, and as a new hire arriving to a smaller base than the plan assumed.
Ranges instead of one convenient number
Every plan contains numbers nobody can know yet — how many of next year's tenders are won, the month a delayed project is finally invoiced, the day the refinancing is signed. In a spreadsheet each of them becomes a single value, and that value quietly carries the decision.
Here they stay uncertain and say so. An important input becomes a range, a distribution, or a dated possibility with a probability attached, and the dependencies between them are stated rather than assumed away.
This is not sophistication for its own sake. It is the difference between a plan that has one answer and a plan whose answer can be interrogated.
A thousand coherent futures
The connected model is then run many times over — a thousand runs is the standard, more when the question warrants it. Each run samples one set of values from those ranges and plays the business forward day by day. Each one is internally coherent: a single future in which those things happened to be true.
What comes back is not a prediction. It is a population of outcomes, and a population can be counted.
One condition, one date
Do the shareholders have to put money in again before the end of 2027? Does the leverage covenant hold at the December test? Is the earn-out threshold reached in the 2027 accounts? Those are the questions a board actually argues about, and each one becomes a single line the model answers yes or no to at a stated date — cash balance below zero at month-end on 31 December 2027, net debt above 3.5× EBITDA at the covenant test, EBITDA below €4m for the financial year.
Writing it down before the model runs is what keeps the answer honest — otherwise the answer is whichever number the output happens to flatter. The condition is either met in a run or it is not, so the result is a frequency: in this many runs out of a thousand, the answer was yes.
That is a more useful object than a haircut on a forecast. A haircut moves one number by a feeling. A frequency says how much of the plan's value rests on things going better than the middle case.
Which assumptions separate the outcomes
Then the interesting part. Split the runs into the ones where the condition was met and the ones where it was not, and compare the two groups input by input. Most inputs took similar values on both sides of the line. A few took visibly different ones.
Those few are where diligence belongs. The useful conclusion is rarely that downside exists — it is that the investment case leans disproportionately on a small set of conditions, and that they can be named, ranked, and given thresholds a board can hold someone to.
This comparison is observational. It sorts the futures the model produced; it does not intervene in them. A driver can travel with bad outcomes because it marks something else that causes them, and telling the marker apart from the mechanism takes a different test.
Changing one input on purpose
That test is the next step, and it is deliberate. Take the same futures, change one declared input, hold the rest of the background where it was, and run them again. The result is a modelled intervention effect: under this model and these assumptions, moving that input moved the outcome by this much.
Where the management team comes in
Assumptions are grounded in working sessions with the people running the business. That is method, not ceremony. The sessions are where a commitment gets separated from an aspiration, where a dependency nobody wrote down surfaces, and where an estimate gets replaced by evidence.
A plan can be coherent, well argued and entirely sincere, and still be one path through a system that has many.
A person signs it
The model is an instrument. A human decides which decision is being answered, which mechanics matter, which evidence holds, which interventions are worth testing, and what the result actually supports.
Simulation does not sign a conclusion. A named analyst does, and the opinion is bounded on purpose: this question, this date, this model.
What arrives
Two documents. A forecast report that orients the reader to the range the business travels in, and a forensic report that takes one condition, splits the runs at the date, ranks the assumptions that separate them, and turns those into actions with thresholds attached.
The forecast is not a projection. Its P&L and cash-flow statement are one run out of the thousand — a single coherent future, picked because it sits in the middle — with the range the others took drawn around it. So a probable event arriving does not call for a new plan: the possibility was already inside the range, and the range narrows instead of being rewritten. Once the months have happened, actuals can be held against it.
DemoCompany — a fictional company examined with exactly this method — has a published copy of both. Read the specimen, or request the sample report and the pair arrives by email.
Where the method stops
Worth being precise about, because it is what makes everything above usable. The analysis does not predict one future, and putting a range on an assumption does not make it true. The probability it reports is a frequency inside a model, not an empirical base rate for companies like this one. It establishes dependency inside that model, not causality in the world.
What it does is make uncertainty, dependency and consequence legible enough to argue with — in front of an investment committee, a board, or the people running the company. That is the product.