Why AI doesn't work for the finance function

Everyone is talking about AI. Everyone is supposed to use AI. And the finance function is, of course, expected to get on board. But there is a problem very few people talk about: AI doesn't work particularly well when it meets the reality of a modern finance function.

It's easy to be impressed by AI. It can write text, analyse information, summarise documents and answer complex questions in seconds. But ask it to produce an accurate group report, analyse profitability per customer or explain why the gross margin changed between two periods – based on data from several different business systems – and it suddenly becomes much harder.

How ready is your finance data for AI? Take our free Finance Control Assessment in a few minutes, or book a demo and see how Exopen makes your data AI-ready.

And there are several fairly obvious reasons for that.

1. AI can't find the data

In many companies, financial and operational data is spread across a multitude of systems. One ERP here, another there. Perhaps several accounting systems after a number of acquisitions. Add CRM, payroll, production systems and a handful of Excel files.

The information is in there somewhere. But it isn't consolidated, it isn't easily accessible and it certainly isn't organised for an AI to work with.

An AI that can't access the data can't do much useful with it either.

2. AI doesn't understand what the data means

Even if AI gets access to the information, the next problem remains.

The same customer can have different customer numbers in different systems. Charts of accounts differ. Cost centres, projects, products and organisational units have different definitions and code structures.

What one system calls revenue doesn't necessarily mean exactly the same thing in another.

A person with deep business knowledge can often sort that out. But for an AI expected to analyse everything together, it's a significant problem.

Having access to data is not the same as understanding it.

3. Garbage in. Garbage out. Only much faster.

Then there is data quality.

Incorrect postings, incomplete dimensions, duplicate entries, missing values and inconsistent definitions are everyday reality in many organisations.

AI doesn't automatically solve these problems. On the contrary, it can make them even harder to spot by presenting wrong conclusions in a very convincing way.

Bad data doesn't get better by being analysed with AI. You just get the wrong answer faster.

4. Finance can't afford to be almost right

This is perhaps the biggest difference between the finance function and many other parts of the business.

In marketing, for example, an AI analysis that is roughly right can still be valuable. It can give an indication, identify trends or create a basis for further decisions.

But finance operates under completely different requirements.

A balance sheet must balance. A group consolidation must reconcile. An income statement must be traceable to correct transactions. And the numbers must be possible to review, explain and reproduce.

It isn't enough for AI to give an answer that seems reasonable.

It has to be right. To the cent. Every time.

And that is where a fundamental problem with general-purpose AI models appears. They are fantastic at language, patterns and reasoning, but they are not built to guarantee financial data integrity on their own.

5. When it really matters, we go back to Excel

So what happens in practice?

Finance teams experiment with AI. They use it to summarise reports, write commentary, analyse variances and streamline various tasks.

But when the month-end close is presented, the board report is delivered or the numbers are reported to the owners?

Then they go back to the old, proven systems.

To Excel. To the BI tools. To the reports they know are correct.

AI gets to help around the edges. But it isn't trusted with the numbers.

And then we haven't really changed very much.

The problem isn't AI. The problem is what we feed it.

For AI to truly transform the finance function, we have to start at the other end.

Not with the AI model. With the data.

All financial and operational data needs to be brought together in a common platform where it is automatically structured, harmonised and quality-assured.

Charts of accounts must be mapped to common structures. Customers, products, projects and organisational dimensions must have a consistent meaning. The information needs to be enriched with metadata describing what it actually represents and how it fits together.

And the data must be available to AI in a way that means the model doesn't have to guess.

Only then can AI start to truly understand the company's finances.

That is exactly why we built Exopen

Exopen is not yet another AI tool trying to make sense of unstructured data from a range of different systems.

We have built the very foundation that makes AI useful for the finance function.

With around 50 standardised integrations to different business systems, our platform automatically collects financial and operational data. The information is harmonised and structured in a common finance model, where accounts, dimensions, customers, products and organisational units get the right context.

The result is a shared, structured and AI-ready data platform.

On that foundation, companies can already build income statements, balance sheets, KPI reports, customer profitability analyses, product analyses and other advanced financial reports.

But the really interesting part begins when we put AI on top of this structure.

Exopen has a specialised AI solution tailored to the platform's financial and operational data model, rather than relying solely on a general language model.

We have also developed our own EXO functions for Excel, which AI can use to create reports directly in Excel from plain instructions.

This means users can describe the report or analysis they need and let AI build it using the platform's structured data and defined calculation functions.

AI doesn't need to make up numbers or try to interpret what different account codes mean. It can work against a common data model and retrieve and calculate values with the same logic as the company's regular reporting.

That lets us combine the flexibility of AI with the finance function's demands for accuracy, traceability and control.

The finance function of the future doesn't need more AI. It needs better conditions for AI.

It's tempting to think the next generation of AI models will solve every problem. That they just need to get a little smarter, a little faster and a little better at reasoning.

But no AI model in the world can conjure an accurate group report out of data it can't access, doesn't understand or that contains errors.

It takes a well-designed data platform, a common information model and an architecture where AI can work with reliable data and deterministic calculations.

Only then can AI go from being an impressive assistant to becoming a real working tool for the finance function.

So no, AI doesn't work for the finance function.

Not until we give it the right conditions.

And that is exactly what Exopen does.

Want to know where you stand today? Take the Finance Control Assessment and get a clear picture of how well your finance data supports reporting, decisions and AI. Or book a demo and we'll show you how Exopen collects, structures and makes your data AI-ready.