What is FP&A? Financial planning and analysis in the AI spend era

FP&A also now owns the newest board-level question: whether AI spend is producing measurable returns.

The earnings call that explains the whole job

In May 2026, Duolingo CFO Gillian Munson stood on a Q1 earnings call and did something finance leaders spend entire careers avoiding.

She guided gross margin down. On purpose.

Q1 margin had expanded year over year and beaten the company’s own expectations.

Munson’s guidance anyway: approximately 71% in Q2, then trending down to roughly 69% by year end “as AI-powered feature use in our products expands,” per the call transcript. Margin walking out the door, announced with a straight face, because the company wants it that way.

The culprit wasn’t a pricing mistake or a supply chain fire. It was success. Duolingo published 20,500 course units in a single quarter, a pace AI made possible and more than ten times what it shipped per quarter two years earlier. Daily active users grew 21%. Adjusted EBITDA hit $83 million, about 29% of revenue. And every AI-powered video call and speaking exercise arrived with an inference bill stapled to it.

Munson said the quiet part out loud on that same call: “we have started to see some pretty big increases in AI costs internally.”

Here’s the nuance that separates a good finance team from a nervous one. Duolingo’s per-unit AI spend is falling. Aggregate AI spend is rising anyway, because adoption is outrunning the unit savings. Munson’s team modeled that tradeoff, decided the margin was a price worth paying for growth, and could prove it. Then she defended the math to Wall Street in real time.

That work happens in FP&A. And whether your company has 50 employees or 50,000, that’s the job now.

(Full disclosure, and honestly a bit of a flex: Duolingo is a CloudZero customer, and we’ve published the case study to prove it. We are not claiming credit for their earnings call. We’re just noting that the company in this story, the one trading margin on purpose with its eyes open, runs its spend intelligence on the platform this article’s publisher built. Draw your own conclusions.)

What does FP&A mean?

FP&A meaning, in one line: the finance team that turns raw financial data into decisions. That’s the working FP&A definition; the textbook version adds budgeting, forecasting, and analysis, which is accurate but undersells the job.

So what is FP&A in finance organizations, practically? If accounting is the company’s rearview mirror, FP&A is the windshield. Accounting closes the books. FP&A opens them back up and asks the uncomfortable questions. Why did cloud spend grow 30% while revenue grew 12%? Which beloved product is quietly eating the margin everyone brags about? Fund the expansion, or kill it?

What does FP&A stand for in the org chart? The function reports to the CFO, alongside accounting, treasury, and tax. At a startup, it might be one analyst with a heroic spreadsheet habit and a mild caffeine dependency. At an enterprise, it’s a full team split between corporate planning and business unit support.

The title on the door says planning and analysis. The actual product is confidence: leadership’s confidence that the numbers in the board deck will survive contact with reality.

What does an FP&A team actually do?

Four core jobs, plus whatever fire is burning this quarter.

That last column isn’t hypothetical. In CloudZero’s 2026 AI ROI survey of 260 finance leaders, 46% called managing AI spend the most stressful part of their job. Not the close. Not the audit. Not budget season. The AI bill.

And the pressure flows downhill from the boardroom: 66% of finance leaders in the same survey said their boards now tie AI funding directly to demonstrated returns.

What does the FP&A process look like, step by step?

The FP&A process runs on an annual loop with monthly and quarterly checkpoints. Five steps:

The loop hasn’t changed in decades. The difficulty setting has. When most organizations miss their AI spend forecasts by 11 to 25%, step four stops being a formality and becomes the main event.

What’s the difference between corporate FP&A and business unit FP&A?

Corporate FP&A sits at headquarters and owns the consolidated picture: the company-wide budget, the board deck, the forecast the CEO gets grilled on. Business unit FP&A embeds with a division or product line and works its specific P&L.

Air traffic control versus pilots. Corporate makes sure the fleet lands where the plan said. Business unit teams fly individual planes and radio back when something’s off course.

The tension between them is a feature. Business units push for resources and sunny assumptions. Corporate pushes back with portfolio discipline. Somewhere in that argument, a realistic plan gets born.

AI spend scrambled this structure in a way neither side enjoys. Inference bills don’t respect organization charts. One AI feature might burn OpenAI API credits, Claude tokens, and GPU compute across three cost centers. Neither corporate nor business unit FP&A can see the whole picture from where they sit. That’s how AI budgets die: a thousand unattributed cuts.

How did AI change the FP&A job?

Short answer: AI made FP&A’s tools sharper and its subject matter much messier. Both at once.

The tools got better

AI in FP&A workflows now handles the grunt work that used to eat analyst weekends: data consolidation, first-draft variance commentary, anomaly flagging, scenario generation.

The adoption curve is early but steep. Gartner forecasts AI agent software spending will hit $206.5 billion in 2026 and $376.3 billion in 2027. Only 17% of organizations have deployed AI agents so far. More than 60% expect to within two years.

Translation: the 2028 FP&A analyst spends less time reconciling exports and more time doing the thing the job was always supposed to be, which is thinking.

The subject matter got wilder

Now the messy half. And to understand it, rewind fifteen years, because finance has watched this movie before.

Cloud was the first line item that broke the FP&A playbook. Infrastructure used to be CapEx: predictable, committed, depreciating on a schedule an analyst could model in their sleep. Then compute went usage-based, and the bill started scaling with engineering decisions nobody looped finance in on. An entire discipline of cloud cost management grew up to get that spend attributed, forecasted, and defensible.

The teams that actually won the cloud wave learned something bigger than cost control. The size of the bill was the wrong question. The right question was what the bill bought: cost per customer, cost per feature, gross margin per product. Cloud spend stopped being a cost to minimize and became an investment to measure returns on.

Cloud taught finance to think in ROI. AI is the final exam, and it’s timed.

Look at the velocity difference side by side:

Sources: CloudZero’s AI era recalibration report with Benchmarkit, February 2026; The State of AI Costs.

The macro numbers say the acceleration is just starting. Gartner’s May 2026 forecast puts worldwide AI spending at $2.59 trillion for 2026, up 47% year over year, with more than 45% of it flowing to infrastructure: the same cloud compute finance just learned to measure, now consumed at AI velocity. Spending on AI models and platforms alone will hit $64 billion, up 63.4%, with generative AI model spending growing 117%.

Gartner’s John-David Lovelock calls 2026 the inflection year because, in his words, “Enterprises have yet to really flex their spending potential.” His colleague Arunasree Cheparthi added in July 2026 that enterprise AI budgets now face “increased focus on usage efficiency, cost control and measurable outcomes.”

Every dollar of that flex lands on an FP&A model built for different physics. Traditional spend is fixed (rent), stepped (headcount), or committed (SaaS contracts). Cloud broke that taxonomy first. AI breaks it harder, because the rate card is public (ChatGPT’s pricing, Gemini’s pricing, all of it) while your usage pattern, the thing that actually determines the invoice, is the variable nobody hands you.

The efficiency bill is already arriving

Here’s the stat that should be in every FP&A team’s next deck. CloudZero’s February 2026 research with Benchmarkit found that median Cloud Efficiency Rate fell from 80% to 65% in a single year. In plain finance terms: the typical company now sends 35 cents of every revenue dollar to cloud providers, up from 20 cents.

That happened while formal cloud cost programs nearly doubled, from 39% to 72% of organizations. More governance, worse efficiency. The AI surge is outrunning the controls built for the cloud era, which is precisely the kind of sentence that gets FP&A invited to more meetings.

What does FP&A become next? CloudZero’s take

We have a strongly held opinion here, as the company that put “the AI ROI company” on its homepage after a decade of doing the same math for cloud: FP&A becomes the AI ROI desk, and the teams that get there first become the most powerful function in the building.

The reasoning runs on three facts.

First, the ROI question is now unavoidable. Boards tie AI funding to returns (66%, per the survey above). Yet only 51% of organizations can confidently evaluate AI ROI, per CloudZero’s State of AI Costs research of 500+ software leaders. Someone has to close that gap, and it won’t be the engineering team volunteering.

Second, the gap is telemetry, not talent. ROI is a simple formula with brutally hard inputs. The cost side only exists once every dollar is attributed to the customer, product, or feature it served, and half of finance teams still wait days, or until the invoice lands, to see AI spend at all. One CloudZero customer discovered $600,000 of annual AI spend traced to a single employee running 40 different models. Nobody was stealing. Nobody could see. Arguably worse.

Third, visibility measurably changes the outcome. In the same State of AI Costs research, more than 90% of companies tracking AI spend in a third-party platform reported high confidence in calculating AI ROI. Against the 51% baseline, that’s the difference between guessing and knowing, quantified.

CloudZero founder and CTO Erik Peterson’s bet on where this ends: “it’s going to be the ones with the best AI unit economics” who win, not the biggest spenders.

What that looks like mechanically

This is the part most vendors hand-wave, so here’s the actual machinery, researched from our own platform releases rather than adjectives:

Skyscanner, Grammarly, and Upstart run their cloud and AI economics this way, alongside Duolingo. The pattern repeats: attribution first, then the forecasts start holding, then the board conversation stops being scary.

Will AI replace FP&A?

No. But it will replace FP&A teams that can’t answer AI questions with FP&A teams that can.

The judgment work at the core of FP&A, deciding what a variance means, pressure-testing an operator’s sunny forecast, telling a CEO their favorite initiative doesn’t clear the Rule of 40 math, is exactly what AI is worst at. Models are excellent at producing plausible numbers and terrible at knowing which numbers matter.

What AI does replace is the mechanical layer: consolidation, formatting, first-pass commentary. Most of the hours, least of the value.

The career math is unusually clean. AI spend is the fastest-growing line item in most budgets, the least understood, and the one the board asks about every quarter. Duolingo’s CEO Luis von Ahn put the stakes plainly on that same Q1 call: “AI has fundamentally changed what’s possible for us.” The finance person who can price what’s possible owns the most valuable answer in the building.

The board meets in six weeks. Walk in with a number.

Somewhere on your board’s next agenda is a version of this question: is the AI investment working? You can answer with a shrug dressed up as a range, or with cost per customer, gross margin impact by AI feature, and a forecast you’d stake your credibility on.

The second answer requires seeing every AI and cloud dollar attributed to the outcome it served, the way CloudZero’s financial control plane for finance teams does it.

Three ways in, depending on how your week looks:

FAQs: financial planning and analysis

Author Spotlight

Lyne Carolyne

Lyne Carolyne has several years of experience in AI and cloud economics and brings that understanding into the content she creates. Outside work, she's an avid explorer.