FP&A baseline forecasting: 2027 has no clean base year

Why last year's actuals stopped working as an anchor

Incremental budgeting survives because it is cheap and it is auditable. You take the prior-year actual, apply a growth or efficiency factor, and defend the delta. It works when the composition of spend is stable, because the base year carries most of the information about next year.

That assumption is now weakest where the money is moving. If a company stood up an AI program mid-year, the actual reflects a partial year at a partial run rate. If cloud consumption scaled with usage rather than seats, the actual reflects last year's volume, not next year's. If growth-function headcount landed in the back half, the actual understates the annualized cost of the people already hired. In each case the base is not slightly wrong. It is wrong in a direction that compounds.

The error does not stay in the budget file. A distorted baseline propagates into hiring plans, covenant headroom calculations and, eventually, external guidance. That is what makes this a control issue rather than a modeling preference.

The lines that need a driver, not a growth rate

The practical question is not whether to abandon incremental budgeting wholesale. Most of the cost base is still stable enough to roll forward, and rebuilding every line from first principles is a good way to miss the calendar. The question is which lines earn a driver-based rebuild.

A workable test has three parts. First, did the line change composition during the base year, not just level? Second, is the cost mechanically tied to a volume the business can forecast independently, such as tokens, compute hours, transactions or headcount by month? Third, would a 30 percent miss on this line change a downstream commitment? Any line that clears all three should be modeled from its driver, with the prior-year actual used only as a reasonableness check.

Everything else can stay on a roll-forward with a documented factor. The discipline is in the sorting, not in the sophistication of the model.

Forecasting accuracy is now the stated mandate

The timing is awkward because expectations moved in the same cycle. CFO Dive reported on Aug. 5, 2026 on an FTI Consulting whitepaper arguing that the CFO mandate has shifted past reporting. FTI put forecasting accuracy and scenario planning at the front of its list of value-creation opportunities, alongside better visibility into cash, margins and operational drivers.

So FP&A is being asked for sharper forecasts at precisely the moment its base year stopped being a reliable starting point. That is not a contradiction, but it does change where the work sits. Accuracy in this environment comes from getting the driver logic right on a handful of volatile lines, not from tightening variance tolerances on the stable ones.

Automation is not the same as a rebuilt baseline

FTI paired its list with a caution worth repeating in every planning kickoff: do not automate for its own sake. It is a useful counterweight to the tooling narrative, because the tooling largely arrived first. CFO.com reported in November 2024 that FP&A software implementation more than doubled that year, which means many teams now have planning platforms that are faster at producing the wrong baseline.

Vendor-sourced figures point the same way and should be read as directional only. IBM cites 69 percent of CFOs saying AI is central to finance transformation. Sentiment is not method. A model that pulls last year's actuals automatically and applies a growth rate is still incremental budgeting, just with a shorter cycle time.

The capability gap sits next to the tooling gap. The CFO reported in March 2026 on CFOs working to hardwire desk-ready FP&A excellence, which is a fair description of the constraint: driver-based planning requires analysts who can interrogate an operational driver with the business owner, not only assemble a variance pack.

A narrow ask for this planning cycle

The coverable version of this work is small. Ask the FP&A lead for a one-page classification of the cost base into two buckets: lines that roll forward, and lines that get rebuilt from a named driver with a named owner. Require the second bucket to state the driver, the source of the volume assumption and the sensitivity that would move a downstream commitment.

Then check the annualization. Any line added or scaled mid-year in the base period should be presented at full-year run rate before any growth factor is applied. That single step removes a large share of the error that incremental budgeting quietly imports.

If the list of driver-based lines runs longer than roughly a dozen, the scope is probably too wide to execute well. If it is empty, the plan is almost certainly anchored on a base year that no longer describes the company.