Every finance conference and expo I have sat through this year has had one centre of attraction: someone saying Artificial Intelligence (AI) is going to replace FP&A and the financial analyst within the course of five years.

I disagree.

In my experience, the real dividing line is simple: AI works where the constraint is data and throughput, but FP&A still relies on human judgment and accountability.

The more in-depth I went into integrating my own systems, which involved planning, budgeting, and forecasting, the more I understood that AI still cannot provide the niche market understanding, or read the trend of a business that is not based on algorithms or cyclical patterns, but on the human intelligence of understanding market emotion.

AI is changing the way the FP&A profession works, but not in the way the SaaS companies hyping this rhetoric in the boardroom would have you believe. It lacks the authenticity of human intelligence and the vast market knowledge of a person who deals with the market on a day-to-day basis. Do not get me wrong, AI has lifted a real burden off the FP&A profession on the efficiency side, where the work is repetitive tasks and visualisation rather than cognitive thinking. What follows is my honest opinion on AI and how it can change our FP&A profession's life.

Where AI Earns Its Place

There has been a major winning streak for AI on speed, accuracy and data ingestion. It stitches reports together from multiple data sources, guards the data, converts it into KPIs, applies business logic through calculations, and analyses the apparent market shape. This genuinely helps the financial analyst cut down decision time and walk into a board meeting with a quick read on the numbers. Anyone who has ever taken over a chart of accounts after an acquisition, or tried to piece together cost-centre structures from multiple old systems, knows how messy finance data can get. They know the soul-erosion of dimension mapping.

In one recent build, I worked with almost 21,000 rows of primary mapping table sitting across a fallback table, applied weekly across multiple entities and cost centres. AI assisted with the major ETL part. It helped clean and trim the data, understand how it was set up, and made some bold suggestions on outlier-value mapping logic to ensure I stayed in line with the report's accuracy. The model never made the judgment call. I did. AI assisted me with code generation and mapping entities that were convoluted across different sources. It did not replace the work, but it helped with the heavy lifting. It wrote the template. It flagged the edge cases I had missed. And at one point, when rows were dropping from the joins and the transformation, it helped me debug exactly where I had failed to understand my own logic. This is the real shift. Finance is no longer working just to churn out reports through these tools. We are building systems that can answer the major questions, support the leadership team, and sharpen the way they think about business goals. The same job, using only Excel and Power Query, would have taken me at least a week of work.

Second is the draft of the management commentary. From board-pack meeting notes to summaries that define the quarter, the month, and the weekly close, none of this needs to be innovative or creative. It just needs to follow a certain template, with accuracy, speed and consistency in tone. A trained and well-prompted AI will maintain fixation on the actuals and prior data and produce a draft for the analyst within a tenth of the time it would take the analyst to observe and write the commentary themselves. This is a win-win for both manager and analyst, who now have something to start working on. If deeper analysis is required, the analyst already knows where to start, because the quick management commentary has shown them the shape of the month.

Where It Still Falls Short

Vendors who claim to crack the code of FP&A would rather sell the optimism, just like AI itself. AI is, by default, trained to think in best-case scenarios, no matter what the actual numbers are, and that is what we call AI hallucination. It is pretty common to work with AI during the planning process and walk away with a false sense of accomplishment about certain numbers. But we need a dial. Not just best, likely, and worst case, but a sensitivity dial that the prompter or end-user can set themselves, based on how they read the pattern or the market shift that is actually happening.

Even a powerful AI will always have a limitation when it comes to forecasting a genuinely uncertain future, which human intelligence can often sense well before the wave of uncertainty hits. To give one example, AI will never understand the politics behind a variance. When the sales team has one projection and the finance team has another, that gap is not a system error. AI sees the numbers, but not the boardroom drama. An FP&A associate can walk into the cabin of the sales department and get the answer sorted in five minutes. AI cannot.

A junior analyst can sense a missing accrual or an incurred cost, simply because the size and shape of the P&L report feels wrong. AI has a more controlled environment, which means the sense of being wrong is highly unlikely until someone catches the mistake. Otherwise, it becomes a silent mistake, which is worse. The best part of AI's shortcomings, though, is the sense of ownership of the numbers. AI cannot carry the responsibility for its number the way a head of finance would, whose name is attached to the forecast model and who walks down to the CFO's office to explain the reality of the variance between plan, actuals, and forecast. Human looping is not optional. It is required, because someone has to take responsibility and own the numbers being produced.

Why the Gap Between Expectation and Reality Exists

The real difference between what AI is supposed to do in FP&A and what actually works is not a model problem. It is what sits underneath the model. In my current work, building a cloud-native FP&A stack at the organisation level, the real potential lies in the reporting layer, not in financial modelling or prompts. Most organisations run on fragmented inputs coming from different ERP systems, with each department having its own separate platform and a long tail of Excel files. Stitching the report takes most of the analyst's time, and creating harmony at the reporting layer is what allows AI to plug in. Until that layer is harmonised with consistent definitions across entities, AI cannot deliver end-to-end value on top of it. The model is only as good as the semantic layer it sits on. The other reason the gap exists is the business context. AI does not sit in the QBR. It does not hear the head of marketing explain a campaign was pulled forward, or the COO flag a supplier renegotiation. That context is what turns numbers into decisions, and right now it lives in human heads, not in any data feed.

The Line That Separates the Two

If I had to compress this article into one rule, it would be this. I am pro-AI where the work is about data and the constraint is throughput. Humans handle the judgement, the uncertainty, and the work where the constraint is accountability. Most FP&A analysts today are trying to create automated workflows with the help of AI that eliminate repetitive reporting work and support reports for the ELT and SLT teams in their meetings, where the numbers do not normally drift apart. The teams that struggle are the ones that rely on an AI platform to handle FP&A end-to-end and now own a tool that produces plausible answers nobody can vouch for.

What This Means for FP&A Leaders

If a CFO needs data quickly for an analysis tomorrow, they should not have to wait for an Excel report from the director. It should be readily available. I would suggest applying AI to the repetitive actions: data ingestion, mapping, ETL, calculations based on business drivers, and converting them into actionable KPIs with visualisations. This becomes the scope within the AI domain, helping the analyst become a powerful analyst focused on the business side rather than wasting time on information reconciliation. Automated forecasting and planning models can be a future thing, but they will always produce numbers that may not carry the tag of authenticity. The real investment is implementing an AI-ready reporting layer with consistent mapping, consistent dimensional definitions across entities, and a finance team that can prompt and interrogate AI output rather than just consume it. The teams that invest there first are the ones that get genuine leverage. The teams that skip it end up owning expensive tooling and the same number of analyst hours as before.

The Honest Outlook

AI in FP&A today is not a replacement for the function. It is a serious augmentation of the parts of the function that were never the most valuable parts to begin with. The analysts and finance leaders who treat it that way, as a way to speed up the plumbing, drafting, and document work, are getting weeks of capacity back every quarter. Those expecting a magic forecasting engine are quietly disappointed, and louder about it than they should be.

The job of finance has never really been to produce numbers. It has been to produce the right conversation around the numbers. AI is making the first part faster. The second part is still, stubbornly and reassuringly, ours.