News

Pankaj Prasoon is Sr Director Finance Tech & Strategy at LinkedIn, leading finance technology and building AI fluency across finance teams.

gettyConsumption pricing hands finance an operating expense with the volatility of a commodity, a demand curve tied to its own success and no natural owner.​

Every cost that finance manages has a behavior we understand. Headcount increases in steps. Rent is largely fixed. Software was predictable enough to be boring: Negotiate annually, count the users, forecast the expense.

AI is changing that equation.

In an earlier article, I argued that capacity released is not value realized. The cost side has the opposite problem. Consumption is happening before finance fully understands what it is paying for.

In Deloitte’s second-quarter 2026 CFO Signals survey, cost uncertainty or lack of transparency was the top internal concern about AI, cited by 46% of CFOs. The Wall Street Journal reports finance chiefs are now trying to track AI usage before token-based pricing produces sticker shock.

That is an unusual position for finance. We are rarely the function surprised by a cost.

Here are three things that change the equation.​The cost is unpredictable. Two employees can use the same AI tool and generate vastly different expenses. Model selection, request complexity, repeated attempts and workflow design can all change the bill.

Success can increase spending. When more employees adopt a useful AI workflow, total consumption rises. A successful rollout can create a budget variance even when the underlying economics are improving.

The people making spending decisions are often not accountable for the budget. An engineer changes a model, adds a processing step or adjusts how often an agent runs. The decision may improve performance, but its financial impact appears somewhere else.

We spent decades putting approval processes around spending decisions. AI has quietly moved some of those decisions into technical configurations finance may never see.

Build the unit before the budget.​The obvious response is a spending limit. But a cap tells people when to stop. It does not tell them whether the money was well spent.

Start with the unit of work. What does it cost to process an invoice, prepare a reconciliation, resolve an exception or complete a close activity?

Then examine what drives that cost. Did volume grow? Did the price change? Did the workflow switch to a more expensive model?

This is familiar territory: price, volume and mix. The same discipline applies, with one addition: We must understand how much AI consumption it takes to produce a successful result.

A rising bill may reflect more work being completed. It may also reveal waste. Without that distinction, finance cannot respond intelligently to either.​

Cheaper does not always mean better.​Once finance measures cost per transaction, the instinct is to drive it down. That is the next trap.

Consider two workflows. One costs 50 cents and produces a reconciliation that needs 20 minutes of human correction. Another costs $2 and produces an accurate result with minimal intervention.

At $60 an hour in loaded labor cost, those 20 minutes add $20. The 50-cent workflow is the expensive one.

Cost per transaction is half the story. The other half is the quality of the outcome and the human effort still required.

The objective is not the lowest cost of intelligence. It is the best economic outcome from using it.​

Decide who pays before the arguments begin.​Central funding encourages experimentation, but it can make consumption feel free to the teams generating it. Charging every department from day one creates the opposite problem. Teams stop experimenting because they inherit costs before they understand the benefits.

Fund early experimentation centrally, but make consumption visible to the teams using it. When a workflow moves into production, establish a business owner, an agreed cost baseline and a clear expectation of what it should deliver.

As benefits become measurable, spending accountability should move closer to the leaders receiving them. Cost ownership and benefit ownership should not live in separate organizations indefinitely.​

Do not build next year’s forecast from a pilot.A pilot with 40 employees tells us little about consumption across 4,000. Heavy users, edge cases and exceptions only show up at scale.

Multiplying pilot spend by 100 is not a forecast. If the business expects usage to double in the third quarter, finance should know what that means for spending in the same quarter. If the operating plan assumes productivity gains, it must also account for what those gains cost to produce.

The adoption plan, cost forecast and benefit assumptions need to tell the same story.​

The line needs an owner.​Engineering understands the technology. Procurement understands the contracts. Business leaders understand the work being transformed.

Finance must connect those decisions to their economic consequences.

Finance does not need to choose models. It needs clear expectations around spending, measurable results and accountability before consumption scales.

In my organization’s finance AI program, weekly token and license spend reporting came first. Our proposed scorecard measures cost to serve against value delivered, alongside error and rework rates. We deliberately do not track licenses assigned or pilots launched. Both rise reliably while value stays flat.

The monthly review should ask more than whether AI exceeded budget. Did the additional spending produce additional value? Did the cost of a successful outcome improve? Are we paying for work that still requires substantial human intervention?

Ask these questions while decisions can still be changed, not months later when the invoice arrives.

The next budget will reveal what we have learned​.For years, finance has helped organizations do more with less. AI introduces a more complicated possibility: Doing substantially more may require spending more.

That should not make us uncomfortable. It should make us more precise.

Time saved is not necessarily value realized. The reverse is equally true: A rising AI bill is not necessarily money wasted.

The real test is whether finance can explain what the business received for each meaningful increase in AI consumption.

The goal is not to make AI cheaper. It is to make every additional dollar worth spending.

The information provided here is not investment, tax or financial advice. You should consult with a licensed professional for advice concerning your specific situation.​

Forbes Finance Council is an invitation-only organization for executives in successful accounting, financial planning and wealth management firms. Do I qualify?