Finance teams with a December year-end start next year’s budget around now, and it’s the first one they’ll build with the amended FRS 102 already in force. The Financial Reporting Council’s periodic review amendments apply to accounting periods beginning on or after 1 January 2026. Most leases move onto the balance sheet, and a forecast still treating rent as an operating cost will produce numbers that don’t agree with the accounts it exists to predict.
That mismatch isn’t a forecasting-method problem, and neither are most of the others. The list of methods a UK mid-market finance team actually uses is short, and so is the rule for choosing between them. The harder question is what your ledger can feed them, and somebody answered that when they designed your chart of accounts.
What Is Financial Forecasting, and How Is It Different From a Budget?
Financial forecasting is the practice of estimating what your revenue, costs, cash, and balance sheet will actually do over a defined future period, using the history and operational data you already hold. A budget is a different document: it commits you to what should happen, and the board measures you against it afterwards. A financial plan sits above both, setting the strategy the budget funds and the forecast tests. Treat a forecast as a commitment and nobody updates it. Treat a budget as an estimate and nobody enforces it.
Which Financial Forecasting Methods Do Finance Teams Actually Use?
Four methods carry almost all mid-market work: straight-line, percent of sales, moving average, and linear regression. Driver-based and rolling forecasting layer over the top of those four as disciplines. Top-down and bottom-up describe which end you build from. Sage’s March 2026 guide to financial forecasting methods and best practices covers the same ground in more depth. What follows sticks to what a UK mid-market ledger can actually produce.
| Method | What it does | What it needs | Best for | Where it breaks |
|---|---|---|---|---|
| Straight-line | Applies a constant growth rate to a base period | One historical period plus a growth assumption | A fast first-pass number on stable revenue | Assumes the rate holds. No seasonality, no mix change |
| Percent of sales | Expresses costs and balance-sheet items as a fixed share of revenue | Two to three years of P&L and balance sheet | Building a full set of financials from one revenue line | Treats fixed costs as variable, so it overstates how far costs will flex |
| Moving average | Averages the last few periods, weighted or not | Enough clean periods to average | Short-horizon cash and demand, and smoothing noisy data | Lags turning points. It will never predict a change in direction |
| Linear regression | Estimates the relationship between an output and one or more inputs | A dependent series plus candidate drivers over matching periods | Testing whether a driver genuinely explains the number | Correlation isn’t causation, and it needs data quality most ledgers don’t produce by default |
| Driver-based | Builds the forecast from operational quantities multiplied by rates | Operational drivers recorded against the same transactions as the financials | Anything where volume, headcount, or units explain the P&L | Needs the drivers to live in the ledger, not in someone’s workbook |
| Rolling forecast | Re-forecasts a fixed horizon every period instead of to year end | Fast, reliable actuals every period | Businesses whose annual plan stops being useful by month four | Needs a close quick enough that the re-forecast is still current |
| Top-down and bottom-up | Two directions of building the same number | Either a target and a market view, or granular operational plans | Sanity-checking one against the other | Used alone, either drifts. Used together, they have to be reconciled |
| Delphi | Structured, anonymous rounds of expert judgement, converged by a facilitator | A relevant expert panel, a facilitator, and time | New products or markets with no usable history | Slow, and it substitutes opinion for data you may already hold |
| Market research | Primary or secondary research into demand, pricing, and competitors | A budget and a defined question | Entering something you have no history in | Tells you about a market, not about your cost base |
How do you choose a forecasting method?
Match the method to the question you’re answering and to the data you already hold. A short-horizon cash question on noisy numbers wants a moving average. Percent of sales builds a full set of financials from one revenue line, with one caveat worth holding onto: it treats fixed costs as though they flex with sales, so it overstates how far your cost base moves when revenue does. Testing whether a driver genuinely explains a number is regression work. Explaining the P&L operationally is driver-based work.
Where qualitative methods fit
Delphi and market research exist for organisations without usable history. Delphi runs anonymous rounds of expert judgement until a facilitator gets convergence; market research buys you a view of demand, pricing, and competitors where you hold none of your own. Both earn their place for a new product line, a new territory, or a first year of trading. For a fifteen-year-old distributor sitting on fifteen years of ledger data, the problem is reaching that history in the first place.
Where scenario planning sits
Scenario planning sits over whichever method you’ve chosen. You build the same model two or three times on different assumptions about the handful of variables that actually move your result. The Association for Financial Professionals’ 2026 benchmarking survey of 332 finance practitioners in 54 countries, with fieldwork in August and September 2025, found 90% of organisations keep a risk and opportunity list and 89% do contingency planning, while only 38% run scenario planning in structured form. The intent is close to universal. The structure isn’t.
Driver-based forecasting is a chart of accounts decision somebody made years ago.
Why Are Mid-Market Forecasts Usually Wrong?
Mid-market forecasts miss for three reasons, and the method is rarely one of them. ACCA and IMA ran their Q2 Global Economic Conditions Survey between 3 and 17 June 2026, gathering 647 responses from finance professionals. Jonathan Ashworth, chief economist at ACCA, summed up the mood in July: “Despite some improvement in confidence, accountants globally remain very cautious, likely in part reflecting the uncertain and unpredictable operating environment which has become the ‘new normal’ in recent years.” A model your team rebuilds by hand every time conditions move is the wrong instrument for a normal like that.
The actuals aren’t finished when you set the forecast
You can’t re-forecast from numbers you haven’t closed. A team running a three-week close and a monthly re-forecast works most of the month from figures already a period out of date, which is why automating the month-end close usually has to come first.
The ledger doesn’t carry the drivers
Driver-based forecasting needs the operational quantity and the financial value sitting on the same record. Revenue per consultant, cost per delivery, margin per site, contribution per project: each one needs the ledger to know which consultant, which delivery, which site, and which project. Where the chart of accounts records an account code and a cost centre and nothing else, the driver doesn’t exist as data. Someone on your team rebuilds it by hand each cycle, mapping codes to sites from memory and a lookup table. That reconstruction is one of the costs spreadsheets carry without anyone booking them.
The same Association for Financial Professionals survey found only 43% of organisations run rolling forecasts, with most still working to a current-year forecast that stops at December. Adoption sits there because of what the ledgers underneath can feed.
The model doesn’t reconcile to what you’ll report
Your management forecast and the numbers you file have to be the same shape, and in 2026 that shape changed. Where the model still classifies lease rentals as an operating cost, every variance report from here compares two different definitions of the same line. Sage Intacct holds the classification once, on the transaction record, and reports from it.
What Changed for UK Forecasts in 2026?
The amended FRS 102 changed what a UK forecast has to produce this year. Going concern hasn’t changed at all, and it still runs on a clock most models stop short of. The Financial Reporting Council’s periodic review amendments to FRS 102 apply to accounting periods beginning on or after 1 January 2026, with early adoption permitted.
Leases move onto the balance sheet
Lessees no longer split operating leases from finance leases. Most leases now sit on the balance sheet as a right-of-use asset with a matching lease liability, broadly following IFRS 16 with UK simplifications, though leases of twelve months or less and leases of low-value assets keep an exemption. Revenue moves to a single five-step model aligned with IFRS 15, and ACCA’s summary of the key amendments sets out the detail. The forecasting consequence is arithmetic: rent that sat above EBITDA becomes depreciation and interest below it, so EBITDA typically rises while assets and liabilities both increase. Azets makes the point that gearing and net-debt measures look worse, and that covenants built on EBITDA, operating profit, or net assets may need reassessing. That’s a reclassification, and nothing about the trading underneath it has moved.
When the going concern clock starts
Directors assess going concern for at least twelve months from the date they authorise the financial statements for issue, which starts later and reaches further than the year end most models stop at. The Financial Reporting Council updated its guidance on the going concern basis of accounting on 25 February 2025, pulling company law, accounting standards, auditing standards, and the UK Corporate Governance Code into one document and pressing hard on documenting the significant judgements and assumptions behind the forecast. Your bank will ask for the same thing in different words.
What Should a Finance System Give You to Forecast Properly?
Drivers on the transaction record, actuals current enough to re-forecast from, and a planning layer reading the same ledger those actuals live in. A forecasting tab does none of that work. Sage Intacct holds the first and the third in the structure of the system itself.
Drivers recorded on the transaction record
Sage’s dimension documentation lists eight standard dimensions carried on transaction lines: location, department, vendor, customer, employee, project, item, and class, with more available by module and any number you define yourself. Coding a transaction to a project and a location at the point of entry means revenue per site and cost per project become report parameters. That turns the driver-based view into a report you run rather than a model you rebuild.
A planning layer that reads the same ledger
Sage describes Sage Intacct Planning as supporting driver-based models, what-if scenarios, and rolling forecasts, connected directly to the Sage Intacct general ledger, chart of accounts, and dimensions. When the forecast says project margin and the actuals say project margin, both are reading the same dimension off the same records, so nobody has to reconcile the two before anyone trusts the variance report.
Where a finance system stops
No system chooses your growth assumption for you, and none of them knows your pipeline. A badly chosen driver stays badly chosen in any ledger. What the structure removes is the reconstruction work between the ledger and the model, which is where most of the cycle time goes. We implement Sage Intacct for finance teams across more than 30 countries, and we design the dimension structure around how your team actually builds its forecast. Support carries on past go-live because the driver set changes as the business does.
How to Tell If Your Ledger Can Feed a Forecast
Open your last forecast and trace three numbers back to source. Anything that arrived from a workbook is a rebuild your team repeats every cycle. Then take the drivers in the model, whether that’s heads, sites, units, or projects, and ask whether each one exists as a field anywhere in your system. Every no on that list is a manual step in next month’s forecast. That distance between the model and the ledger is the real ceiling on your forecasting, whichever method you pick.
If this sounds familiar, booking a discovery session is a good place to start. Bring your last forecast and we’ll work out which of its numbers the ledger could produce on its own.
References
- FRC: FRS 102 The Financial Reporting Standard applicable in the UK and Republic of Ireland, periodic review amendments
- Sage: Financial forecasting guide, methods and best practices, published 20 March 2026
- Association for Financial Professionals: 2026 AFP FP&A Benchmarking Survey Report, Integrated Planning, published 20 January 2026, fieldwork August–September 2025, n=332 across 54 countries
- ACCA and IMA: Global Economic Conditions Survey Q2 2026, fieldwork 3–17 June 2026, n=647
- ACCA: Cost pressures surge in Q2 according to accountants and CFOs globally, July 2026
- ACCA: Key amendments to FRS 102 following periodic review, May 2025
- Azets: Lease accounting changes, what they mean for EBITDA and borrowing
- FRC: Guidance on the Going Concern Basis of Accounting and Related Reporting (including Solvency and Liquidity Risks), published 25 February 2025
- Sage Intacct Help: Types of dimensions
- Sage: Planning, budgeting and forecasting software for Sage Intacct
