AI for Budget Forecasting

Budget forecasting is the discipline of saying, honestly, where a job will finish financially and how confident you are about it. The forecast is not the interesting bit. The movement is. A single number tells you nothing; a number that has drifted the same direction for three months tells you the job has a problem nobody has named yet. The prompts here cover cashflow forecasting, cost-value reconciliation, variance reporting and the commentary that goes around the numbers. The commentary is where the value is, because most forecasting failures are not arithmetic failures. They are people producing accurate numbers with no interpretation attached, and a board reading them as reassurance. WHAT AI IS FOR HERE, AND WHAT IT IS NOT Say this plainly at the top: AI does not calculate. It is a language model, it approximates arithmetic, and it will produce a table that looks immaculate and does not add up. On forecasting, where the margin is a small difference between two large numbers, a modest error in one category becomes a large percentage error in the answer. Every figure comes from your cost system, your valuations and your subcontract ledger. The AI structures, formats and narrates. That is the deal, and breaking it produces confident fiction. Within that boundary it is genuinely useful for four things. Structure and consistency. A CVR or variance report that looks the same every month is what makes the trend readable. AI holds a format better than a person under time pressure. Narrative. Turning a table into the paragraph that explains what moved and why. Most commercial managers are better at the numbers than at the writing, and the writing is what the board actually reads. Scenario framing. What does a six-week delay do to the position. What if the client rejects the pending variations. AI structures the comparison; you supply the figures. Interrogation. Paste in your own draft commentary and ask what a sceptical FD would ask. It finds the soft assertions. WHEN TO USE THESE PROMPTS Monthly, on the same date, as part of the commercial cycle. Consistency matters more than sophistication. When you have inherited a job and need to establish where it actually stands. Before a difficult conversation, so you know your position precisely rather than approximately. When something has moved and you cannot yet size it. WHEN NOT TO Do not use them to produce a number you do not have. If the subcontract forecast is genuinely uncertain, the honest output is a range with the uncertainty stated, not a single figure the AI made look precise. Do not use them on a job too small to warrant it. A £120,000 fit-out needs a cost check, not a twelve-category reconciliation. Do not use them to smooth a bad month. This is the real failure mode in construction forecasting and it is worth naming. The temptation when the position deteriorates is to hold an optimistic forecast, move a risk into opportunities, or delay recognising a loss until next month. Everyone in commercial management knows the shape of it. The problem is that a loss reported late is a far worse conversation than a loss reported early, and the question afterwards is always what you knew and when. A forecast that exists to reassure has stopped being a forecast. FREQUENTLY ASKED QUESTIONS What is the difference between a cashflow forecast and a CVR? They answer different questions and confusing them is expensive. The cashflow tells you when money moves and therefore how much funding you need at the worst point. The CVR tells you whether the job is making money. A project can be profitable and still run out of cash, and it can have comfortable cash while quietly losing margin. You need both, and a board briefed on only one is being told half the story. How often should a forecast be updated? Monthly, aligned to the valuation cycle. Some contractors go quarterly on short jobs, which is defensible but means a problem can run unnoticed for a quarter. What matters more than frequency is consistency of date and method, because the number you actually care about is the movement between periods, and movement is meaningless if the interval or the basis keeps shifting. Can AI calculate our forecast margin? No, and do not let it try. Margin is a small difference between two large numbers, which is the worst possible case for a tool that approximates arithmetic. Calculate in your cost system or a spreadsheet, verify the totals, then hand the verified figures to the AI to structure and narrate. Any workflow where the AI produces the numbers is one where you will eventually report something wrong with total confidence. Should unapproved variations be in the forecast? Show them, clearly labelled, and keep them out of the headline margin. Anticipated variations indicate where value might go. Reporting them in the top line means reporting profit you have not earned and may not get, and it means that when one is rejected you take the whole hit in a single month having never properly booked it. The discipline is to quantify the pipeline separately and let the reader decide how much weight to give it. What is a warning sign in a forecast? Two patterns. A large movement nobody can explain, and no movement at all. The second is more dangerous and more common: a forecast that reports the same margin for four months, then corrects sharply near completion. That pattern almost always means nobody was really forecasting, they were repeating last month's number, and the correction arrives when it is too late to do anything about it. A forecast that never moves is not stable, it is unattended.