Planning, budgeting and forecasting: what software can take over and what stays with people

Software can already gather actuals, roll forward baselines, produce rolling forecasts from driver models and flag variances. People still set budget policy, agree targets with budget holders and decide what a variance means. None of it works until actuals tie to the ledger and cost centres map cleanly to owners.

Steps a machine handles well

Preparing the periodic budget is mostly assembly. Actuals come out of the ledger, headcount comes out of the HR system, and the prior period's run rate becomes the starting point for each line. A planning tool can do that assembly unattended and hand each budget holder a prefilled template. Nobody should still be retyping ledger extracts into spreadsheets.

Forecasting is the strongest case. Statistical and machine learning models are good at projecting recurring costs, seasonal revenue and expected receipts from history. Cash flow projections benefit most, because payment schedules for approved invoices and expected collections already sit in the payables and receivables systems. A model can combine them continuously and highlight any unusually large deposit or disbursement ahead, so treasury is not caught short.

Variance analysis splits in two. Calculating the gap between actual, budget and forecast, then ranking which lines moved most, is mechanical. AI can also draft a first explanation by reading invoice descriptions and journal narratives. Revenue earned but not yet billed is a common source of false variances. An accrual routine run before the comparison removes much of that noise.

Reporting on budget execution can run automatically too, including checks that spending stays inside approved limits. The condition is that every reported figure traces back to a general ledger balance.

Steps that still need a person

Budget policy is a set of decisions about authority: who may move money between lines, and when a reforecast is triggered. Software can store and enforce those rules. It cannot decide them, and it should not be the only place they are written down.

Setting targets is a negotiation. A model can propose a number, but the budget holder has to believe it. Belief comes from conversation about what will actually change in the coming period, such as a new contract or a team being merged.

Turning a plan into action belongs to operations. Finance can publish budgets and send alerts. Hiring freezes and supplier renegotiations are management calls.

Explaining variances is where automation most often oversells itself. A drafted commentary is only a starting point. Someone close to the business has to judge whether an overspend is a timing quirk or the first sign of a structural problem. Breaches of spending limits need a named person to review and report them.

What the data must look like first

Actuals must reconcile. If numbers in the planning tool cannot be traced to ledger account balances, every forecast built on them inherits the error. Budget holders notice quickly and stop trusting the output.

The chart of accounts and the cost centre hierarchy need to be stable and owned. Each cost centre should map to one accountable person. Restructures belong in the mappings at the moment they happen, not patched at period end.

History has to be clean enough to learn from. Unusual items and accounting corrections need tagging. Otherwise a forecasting model treats them as patterns and repeats them.

Operational drivers matter as much as financial ones. Headcount, volumes, pipeline and contract dates usually live outside finance. Agree who supplies each driver, how often, and in what format before building anything that depends on them.

Revenue recognition and accrual timing must be consistent between periods. Inconsistent cut off produces variances that reflect booking habits, not performance.

Questions to ask the people who run it

• Which spreadsheet do you open first when the budget cycle starts, and who built it?
• When actuals in the planning tool disagree with the ledger, which do you believe, and how do you fix it?
• Which cost centres have no real owner, or more than one?
• What adjustments do you make to the forecast by hand that never appear in the documented method?
• Who warns finance about a big payment or receipt before it lands, and how?
• How do you decide which variances deserve an explanation and which get ignored?
• Where do budget holders push back, and what usually changes as a result?
• Which approvals happen by email or in a meeting instead of in the system?

The answers often reveal a second, informal process running alongside the official one. That informal process is usually what keeps the numbers credible, and any automation has to absorb it or replace it deliberately.

APQC's Process Classification Framework® (PCF) is an open standard developed by APQC, a nonprofit that promotes benchmarking and best practices worldwide. To download the full PCF or to view definitions and measures, please visit www.apqc.org/pcf.