AI is only as good as the data behind it

Everyone is talking about how AI will transform the finance function. Analyse the general ledger, explain variances, update the forecast and give management answers in seconds.

But one question is often overlooked:

Can AI trust the data it analyses?

It has never been easier to start using AI in finance. A controller can export a report, upload it to an AI tool and ask questions about costs, margins or cash flow.

The answers come quickly. They are well written and often look convincing.

But AI does not automatically know whether the numbers are accurate, complete or comparable. It does not know if an account has been mapped differently in two companies, if a dimension is missing or if the file contains outdated figures.

It can analyse the information it receives.

It cannot guarantee that the information reflects reality.

AI can find patterns, but cannot always understand them

Imagine a group with several companies and business systems. One company uses Microsoft Dynamics, another Visma and a third Fortnox.

In one system, a particular expense is posted to one account. In another, the same expense is split across several accounts. One company uses cost centres consistently, while another lacks equivalent dimensions.

A person who knows the organisation may be able to explain these differences.

To an AI model, it may look as though the business has changed.

The model might identify a rise in costs, but lack the context needed to determine whether the increase is real, due to a reclassification, or caused by different ways of recording the same information.

This illustrates an important distinction:

Having a lot of data is not the same as having usable data.

A general ledger is not a ready-made data model

AI can be used to analyse a general ledger or a spreadsheet. For a specific question, that can be both quick and valuable.

But a general ledger is not a complete picture of the business.

To compare companies, periods and business areas, the information needs to be placed in a common context. Accounts need to be mapped, dimensions harmonised, and the relationships between the income statement, balance sheet and cash flow understood.

The organisation also needs to agree on what the information means.

What counts as recurring revenue? How is gross margin defined? Which costs should be attributed to a particular business area? How are intercompany transactions handled?

AI can calculate a metric in seconds.

But if the business has no shared definition of that metric, the answer is still uncertain.

The risk is not just that AI gets the answer wrong

When a traditional spreadsheet contains an error, it is often possible to trace the calculation and find where the problem arose.

An AI-generated answer can be harder to assess. It is presented in clear language with a logical explanation, even when the underlying data has gaps or contradictions.

The greatest risk is therefore not always an obviously incorrect answer.

It is a credible answer built on incorrect or incomplete data.

For finance teams, traceability is therefore essential. You need to be able to see where a number came from, when it was updated and how it was processed.

Otherwise, it is difficult to tell whether AI has identified a real business insight or simply amplified a problem already present in the data.

AI makes an old problem more visible

Fragmented data is not a new problem.

Finance teams have long spent time exporting information from different systems, mapping accounts, checking files and compiling reports. Excel has often become the place where the information is finally pieced together.

That works as long as someone knows which file is the right one, how the mapping should be done and why two systems show different results.

When AI is introduced, the dependence on that knowledge becomes even clearer.

An AI model can process large amounts of data, but it cannot automatically replace the context embedded in manual processes or held by individual employees.

For AI to understand the business, that context must also become part of the data foundation.

That means shared definitions, consistent dimensions, quality-assured integrations and a structure showing how information from different parts of the organisation fits together.

The AI journey does not begin with choosing an AI tool

Many organisations begin their AI initiative by comparing models, tools and use cases.

Which solution can analyse fastest? Which can produce the best forecasts? Which can give management immediate answers?

But for the CFO, there is a more fundamental question:

Do we have the data foundation required to trust the answers?

Only when financial and operational data is collected, structured and harmonised can AI be used with control. Then the model can analyse the whole business using the same definitions and the same financial logic.

This creates the conditions for detecting variances earlier, explaining why margins change, identifying risks and producing more relevant forecasts.

AI then moves from being a standalone experiment to becoming part of performance management.

The goal is not better AI. The goal is better decisions.

Exopen brings financial and operational data from multiple business systems together in a common structure. The information is harmonised and made available for reporting, analysis, planning and AI.

Finance teams can continue to work in tools such as Excel and Power BI, but with a shared data foundation that makes figures comparable and traceable across the organisation.

That is the foundation for AI that delivers not just quick answers, but answers you can actually use.

Because better AI does not start with a better prompt.

It starts with better data.

Is your data foundation ready?

Take Exopen’s quick assessment to find out whether your organisation has a reporting problem or a data problem.

Start the assessment