A lot of companies still run finance like it's 2012.
Budgets get locked once a quarter. Forecasts get refreshed once a month. Engineering ships an AI feature on Monday, and finance figures out what it did to cloud spend somewhere around the middle of the following month.
That arrangement holds up fine when the business moves slowly. It falls apart the moment machine learning systems start driving customer behavior, infrastructure usage, and revenue in real time.
The mistake I keep running into inside AI-heavy organizations is treating finance as a reporting function when it needs to be an operational one. The old tooling was built for slower environments. AI-native companies don't work that way. Costs move faster, product behavior shifts daily, and infrastructure usage can spike overnight without anyone touching a budget line.
A paper called "Next-Generation Financial Analytics Frameworks for AI-Enabled Enterprises" makes the same argument in more formal language: finance teams need to sit much closer to the operational layer of the business. That sounds abstract until you hit the actual problems.
The lag is the whole problem
Say you roll out a new recommendation model and engagement climbs 8%. Good news, until you notice GPU inference costs went up 22% to serve it. (Those two numbers are illustrative, but the shape is real.) Revenue goes up, margin quietly leaks out the bottom, and nobody catches it until the quarter closes.
I've watched versions of this play out with idle compute clusters nobody turned off, duplicate training pipelines running in parallel, regional overprovisioning, and AI experiments that never got tagged to a cost center. In almost every case, finance found the problem after engineering had already scaled it. That delay is the expensive part.
Predictive beats a monthly recap
The interesting shift is away from historical reporting toward analytics that are predictive, and eventually prescriptive. A financial system that only tells you what happened last month is a rear-view mirror. What AI-native companies actually need is a system that participates in decisions while they're being made.
That changes the architecture. Instead of a static reporting pipeline that runs after the fact, you need continuous telemetry connecting infrastructure usage, product analytics, operational metrics, and the financial controls sitting on top of them. At that point your finance stack starts to look less like a spreadsheet and more like a distributed software platform.
Unit economics get weird
Traditional SaaS metrics assume customer interactions cost roughly the same to serve. AI systems break that assumption completely. One inference request might cost a fraction of a cent. The next one triggers a long reasoning chain, a vector lookup, two external API calls, and a chunk of GPU time.
If you can't see cost at the level of individual inference behavior, you can end up scaling the workloads that lose you money while congratulating yourself on the growth. I've seen teams do exactly that for a full quarter before anyone asked why the gross margin was drifting.
The integration is messier than the diagram
None of this is clean in practice. Data pipelines drift. Finance tags break. Teams define the same KPI three different ways. Infrastructure owners and finance partners rarely share a vocabulary, and cloud allocation turns political the moment business units start getting charged accurately for what they use.
This is where Technology Business Management (TBM) earns its keep. A TBM model gives engineering, finance, and operations a shared way to talk about cost. Tools like Apptio exist to surface where spend actually originates instead of leaving it buried in one big "infrastructure" bucket.
Simple in theory. In practice I've seen storage costs misallocated for months, shared Kubernetes clusters with no ownership mapping at all, ML experimentation spend that skipped governance entirely, and finance teams that genuinely couldn't tell production inference apart from someone's dev workload. Once AI scales, those gaps stop being annoyances and start being line items.
Forecasting that keeps up
Annual planning cycles don't survive contact with weekly infrastructure pricing changes, models whose usage spikes without warning, and products that iterate continuously. So some teams are now running rolling capital allocation tied directly to operational telemetry: cloud budgets that update on their own, spending thresholds that adjust, anomaly detection that kicks off a governance workflow, forecasts that retrain as new data lands.
Which introduces a new problem. Trust.
The minute a finance system becomes even partly autonomous, executives start asking pointed questions. Why did it move the budget? What drove that recommendation? Why did it throttle spend, and which assumptions changed to make it do that? A forecast nobody trusts gets ignored no matter how accurate it turns out to be. That's why explainability is an operational requirement here, not an academic one.
Governance stops being paperwork
Once a system can influence spending on its own, governance is no longer a document you review at the end of the quarter. It becomes part of the infrastructure. The paper spends real time on auditability and explainability for exactly this reason: these systems now sit close to actual decision authority, and you have to be able to reconstruct why they did what they did.
The organizations getting this right tend to share three things. They have solid operational telemetry. They share financial ownership across engineering and finance instead of throwing it over a wall. And they tie their decision systems to outcomes they can actually measure. The ones struggling usually nail one of those and ignore the other two.
The bigger point is that AI changes what the finance job actually is. In AI-native businesses, the teams that work aren't producing a tidy recap of last quarter. They're building operational control systems for capital, infrastructure, and risk. That's a different job than traditional FP&A, and most companies are still early in figuring out how to do it.