Your 2027 Budget Has a Problem: Last Year’s Numbers May Be the Wrong Starting Point

Last year’s actuals: $4.2 million.

Add 6%.

New budget: $4.45 million.

Done. Put it in the presentation. Someone find the green check mark.

Except the company launched an AI product in July, moved customer support to a usage-based platform in September, and hired half its sales team in the fourth quarter.

So the $4.2 million is accurate. Beautifully accurate. It just describes a business that no longer exists.

This is the problem I keep coming back to as companies build their 2027 budgets. Historical actuals are useful evidence. They are not a legally binding prediction of the future, no matter how nicely we format the cells.

A September 2026 analysis from CFOmatics raises exactly this issue: AI, cloud consumption and growth-function hiring can make the previous year a poor baseline for the next. And Forrester’s 2027 budget planning research reports that more than 80% of surveyed business and technology leaders expect their budgets to increase in the coming year.

That’s an interesting combination. More money to allocate, more business models changing underneath us, and the same old temptation to take last year and add a percentage.

I’m not here to abolish incremental budgeting. I’m here to ask whether we’re applying it to the right things.

The Trouble With a Number That’s Technically Correct

I understand why Finance starts with actuals. We can reconcile them. We can explain them. The auditors aren’t going to ask whether we made them up during a particularly enthusiastic strategy meeting.

And in a stable business, a historical baseline can be excellent. Rent is often predictable. Insurance may have known renewal assumptions. Certain administrative costs follow a fairly steady pattern. Rebuilding every expense line from zero would be an expensive way to prove we have too much time.

The trouble begins when the composition of the cost changes, not just the amount. A new team hired in October isn’t reflected as twelve months of payroll. A cloud service added in June doesn’t have a full year of usage. A product with consumption-based AI costs doesn’t behave like a traditional fixed software subscription.

When we apply a growth rate to those actuals, we’re not necessarily forecasting. Sometimes we’re taking a partial-year expense, treating it like a full year, and giving it a modest raise.

Then everyone acts surprised when the actual bill arrives.

I’d separate two questions at the start of the process: Is last year’s number accurate? And is last year’s number representative? Finance is usually very good at answering the first. The second takes a conversation with the business.

The 2027 Budget Is Trying to Fund Two Different Futures

Forrester’s July 2026 budget guidance suggests business and technology leaders are moving toward renewed investment. More than four in five respondents expect larger budgets, and as many as one-quarter anticipate increases of 10% or more. That does not mean every company or every cost center is growing. It does mean the 2027 planning conversation is about more than maintaining the existing operation.

Management may want to fund AI initiatives, expand revenue capacity, modernize systems and improve efficiency at the same time. All reasonable ambitions. They also compete for the same cash, talent and management attention.

Here is where the historical baseline can become a little too comforting. A department requests an 8% increase because its current expenses are $10 million. Another requests 5%. A third promises savings of 3%. Finance consolidates everything and presents a neat operating expense trend.

But what if the first department’s costs will scale with customer usage, the second’s new hires are only partly reflected in the base year, and the third’s promised savings require a capital investment?

The percentages aren’t necessarily wrong. They’re answering different questions.

I’d want the 2027 budget to distinguish the cost of continuing today’s business from the cost of changing it. Those are two separate conversations, and I don’t think we do ourselves any favors by hiding them inside one growth assumption.

First, Find the Costs That Don’t Behave Like Last Year

Not every line needs a complete rebuild. That’s important, because the alternative to lazy budgeting should not be a nine-month archaeological expedition through the general ledger.

I’d start by flagging costs that meet one or more of four tests. Did the underlying activity start or change materially during 2026? Does the expense move with a measurable operating driver? Did the pricing model change? Would getting the forecast wrong materially change hiring, cash or margin decisions?

AI inference costs may depend on usage rather than seats. Cloud infrastructure can change with workloads, storage, data transfer and service architecture. Sales compensation may change with bookings, headcount and quota attainment. Customer support costs may shift with tickets, automation rates and service-level commitments.

Then there are the less fashionable items. Freight. Utilities. Insurance. Vendor renewals. Overtime. They deserve attention when their drivers change, even if they don’t come with an AI announcement and a very expensive keynote.

For everything else, I’d keep a sensible roll-forward with documented assumptions. A stable office lease doesn’t need a neural network. It needs the lease agreement.

The skill is deciding where additional modeling effort actually changes the answer.

A Midyear Hire Is Not Half a Person in Next Year’s Budget

Let’s use a hypothetical example. A company hires 20 employees on July 1 at an average annual fully loaded cost of $150,000 each. The 2026 expense is approximately $1.5 million, assuming six months of employment and no other timing differences.

If Finance takes that $1.5 million and adds 5%, the 2027 budget becomes $1.575 million.

But those 20 people already represent approximately $3 million in annualized cost before raises, hiring changes or attrition. A 5% increase on that annualized amount would be $3.15 million.

That’s a $1.575 million difference between the naïve roll-forward and the illustrative full-year plan. We haven’t hired a single additional person. We’ve merely acknowledged that July is not January.

I know that sounds obvious. It is obvious when you say it out loud. Budgeting has a remarkable ability to turn obvious things into reconciliation meetings.

And the expense is only part of the story. What did those hires make possible? More capacity? Faster product development? Additional revenue? Better customer retention? If the cost is annualized, the related productivity and business assumptions should be examined too.

Otherwise we’re modeling the payroll very carefully and leaving the business case in a PowerPoint deck.

Cloud Costs Have a Volume Problem, Not Just an Inflation Problem

Consider a hypothetical company that spent $900,000 on cloud services in 2026. A 10% uplift gives us $990,000 for 2027. Reasonable? Maybe.

Now imagine the business expects transaction volume to increase 35%, but an infrastructure optimization program should reduce cloud cost per transaction by 12%. If the entire $900,000 were truly variable with transactions, a simple driver model would produce $900,000 × 1.35 × 0.88 = $1.069 million.

That is about $79,000 above the 10% roll-forward. And it may still be wrong, because cloud spending usually includes fixed commitments, tiered pricing, minimums and one-time changes.

Which is why I wouldn’t stop at the calculation. I’d ask Engineering what drives the bill, Procurement what the contract says, and the product team how customer behavior might change.

Perhaps only 60% of the spend is variable. Perhaps the new architecture reduces data transfer. Perhaps the company has committed-use discounts. Each answer changes the forecast.

This is where Finance earns its seat at the table. We don’t need to become cloud engineers. We need to know which operating decisions create financial consequences and which assumptions deserve a challenge.

And if someone tells me the cloud bill is ‘basically fixed,’ I’d ask whether they’ve met the invoice.

AI Spending Needs Its Own Economic Logic

AI budgets are especially vulnerable to the wrong baseline because many organizations are still experimenting with how the technology will be used and paid for. Some expenses are fixed subscriptions. Others are metered by usage, tokens, calls, compute or vendor-defined units. Some sit in operating expenses; some may be directly associated with delivering a product or service.

I wouldn’t treat those as one homogeneous line called ‘AI investment.’ That’s a wonderfully vague category if the goal is to avoid a useful variance explanation.

For an internal AI assistant, I might model users, adoption, usage per active user, price per unit and support costs. For AI embedded in a customer product, I’d want expected customer activity, inference cost per interaction, pricing and gross-margin implications.

Then I’d ask what benefit we’re buying. Does the tool reduce an external vendor bill? Increase output? Improve customer retention? Allow a team to absorb growth without hiring as quickly? Those are different benefits, and only some produce immediate cash savings.

A time-saving estimate is not automatically a P&L saving. If employees save three hours a week and remain employed at the same salary, we’ve created capacity. That may be valuable. But Finance still needs to explain how that capacity turns into a financial outcome.

I’m optimistic about useful AI. I’m less enthusiastic about an ROI calculation where the benefits are annualized at full speed and the costs are based on a three-month pilot.

A Better Budget Begins With Three Views of the Same Cost

For material, fast-changing lines, I like separating three views. The first is the reported 2026 actual. The second is the normalized exit run rate: what would the business cost if today’s operating structure continued for a full year? The third is the 2027 planned outcome after growth, efficiency, pricing and deliberate investment decisions.

Those views answer different questions. Actuals tell us what happened. Exit run rate tells us what we’ve already committed ourselves to. The plan tells us what we intend to change.

Suppose a function spent $8 million in 2026, but midyear hires and new vendor contracts create a $10 million annualized run rate. Management then proposes a $1 million efficiency program and $2 million of new initiatives. A rough 2027 planning bridge might be $10 million minus $1 million plus $2 million, or $11 million, before timing and execution adjustments.

Compare that with taking $8 million and adding 10% to get $8.8 million. Both calculations are easy. Only one begins to describe the operating choices management has made.

I’d want the bridge visible. Not buried in notes. Visible enough that a leader can say, ‘I agree with the existing run rate, but I don’t approve that initiative yet,’ and we can show what changes.

That’s a budget conversation worth having.

The 2027 Baseline Test I’d Actually Use

I’d put the major cost lines through a short review before locking assumptions. Not a 40-page questionnaire. Something the business can answer without hiring a second person to complete it.

For each material line, I’d ask: What activity creates the cost? What changed during 2026? What is the current monthly or quarterly run rate? What portion is committed? What portion moves with volume? Who owns the volume assumption? What decision would we change if the estimate were 20% too high or low?

That last question is useful because it keeps Finance from over-modeling immaterial items. If a 20% miss doesn’t change a meaningful decision, the line may not deserve a heroic amount of work. If a 10% miss threatens cash headroom, it probably does.

I’d also require an explicit bridge from 2026 actuals to the 2027 plan. The bridge doesn’t need to be elegant. It needs to explain why the number changed.

And I’d have the operating owner sign off on the driver, not merely on the final dollar figure. Agreement on a number without agreement on what creates it tends to expire at the first forecast update.

Don’t Confuse a Budget Target With a Forecast

Here’s another trap. Leadership wants a 12% margin. The business forecast supports 9%. Somewhere between the second review and the board deck, the forecast mysteriously becomes 12%.

I’d love to know what changed. Did pricing improve? Did the product mix shift? Was a cost program approved? Did someone discover a new operating lever?

Or did the number become more ambitious because it was presented to someone more senior?

As I explored in the trade-off between top-down and bottom-up budgeting, targets are useful. They tell the organization what management wants to achieve. Forecasts are useful for a different reason: they describe what the evidence suggests is likely under stated assumptions.

If we turn the target into the forecast without documenting the gap, we lose the very information needed to close it. I’d rather show the 9% base case, the initiatives required to reach 12%, their owners, costs, timing and probability.

Then management can decide whether the plan is achievable, what additional resources are needed, and what happens if execution slips.

There’s nothing unambitious about telling the truth about the starting point.

Scenario Planning Is Where the New Baseline Gets Tested

Once we’ve normalized the base and modeled the key drivers, I’d want at least three coherent scenarios: an operating case, a plausible upside and a plausible downside. Not simply revenue plus or minus 5% while every expense politely stays where we left it.

In a downside case, perhaps sales volume slows, cloud commitments remain fixed and new hires are delayed. In an upside case, transaction volume grows faster, but support and infrastructure costs rise with it. Those are different operating stories, not just different percentages.

The Association for Financial Professionals’ 2026 FP&A benchmarking research found that only 38% of organizations use structured scenario planning. The organizations that do report budgeting cycles about 11% faster on average. That is survey evidence of an association, not proof that scenarios automatically make every budget faster.

Still, I understand the appeal. When management can see how the drivers connect, it spends less time asking Finance to rebuild the same plan from scratch every time an assumption changes.

I’d also connect scenarios to actions. At what demand level do we postpone hiring? When do we renegotiate capacity? What cash threshold changes our investment schedule? A downside scenario with no decision attached is an elaborate way to be worried.

What I’d Tell the CFO Before the Budget Goes to the Board

I’d want one page that separates three things: the cost of the existing business, the cost of already-approved changes and the cost of proposed new decisions.

Then I’d show the material assumptions behind each. Headcount by month. Usage volumes. Contract pricing. Capacity. Productivity. Expected benefit timing. And the cash implications.

I’d highlight the lines where history is a weak predictor. Not every uncertainty needs a red warning icon. But the board should know when a forecast depends on an adoption curve that has barely begun or a savings initiative that still needs operational approval.

I’d also distinguish a management target from a finance-supported forecast. If the two differ, that’s a discussion, not a formatting problem.

And I’d insist on a short list of triggers that will prompt a reforecast. The budget isn’t a promise that the world will hold still until December. It’s a plan made with the best information available today and a way to respond when the information changes.

That is a much stronger story than ‘We increased everything 6%, except the departments that negotiated harder.’

You Don’t Need to Rebuild Every Line. You Need to Know Which Lines Lie.

I don’t want Finance to throw away history. Historical actuals are among the most useful things we have. They show patterns, seasonality, cost discipline and what the business has actually been willing or able to spend.

But a good baseline has to describe the operation we’re planning. When the operating model changes, the forecast has to change with it.

For 2027, I’d keep incremental budgeting where the business is stable. I’d annualize the commitments we already made. I’d rebuild the genuinely variable or structurally changed lines from drivers. And I’d make management choose which new investments belong in the plan instead of allowing them to disappear inside a percentage.

That’s not anti-budgeting. It’s budgeting with fewer pleasant surprises in the presentation and fewer unpleasant surprises in the actuals.

Last year’s numbers deserve respect.

They just don’t deserve the final vote.