Making decisions with AI: from faster work to better decisions
Almost every finance leader uses AI today. Few see it mainly as a tool for better decisions. The difference is rarely the AI tool, but the data it gets to work on and how the work is set up.
Hypergene's survey Confessions of a Nordic CFO 2026 is based on answers from 489 senior finance managers in Sweden, Norway and Finland. The result on AI is clear: 95 percent use the technology day to day. But only 11 percent see AI mainly as a tool for better decision-making, and almost half say the biggest benefit is that they work faster 1.
So AI has made its way into the finance function, but as a way to write, summarise and calculate faster. Few have taken the step to AI helping with the decisions themselves. This guide covers why, and what it takes to get there.
Working faster is not the same as deciding better
When AI is used to work faster, the work stays the same, only quicker: a comment for the monthly report, an Excel formula, a summary of a contract. That is useful, but decisions are still made on the same basis as before.
A better decision needs something else. You need to see what has actually happened, understand why, and compare what different choices mean for profit and cash. That is where most finance functions still lack data, and where gut feeling takes over. In Hypergene's 2025 survey, two in three said they had made important decisions more on feeling than on data 2.
Three things that keep AI out of decisions
- The data is not fresh. The AI gets an export taken last week. The answer is about a snapshot, not today's position.
- The data is not complete. In a group the companies sit in different systems and charts of accounts. A chatbot sees what was pasted, not the whole group.
- The answer cannot be checked. A decision has to be explainable to management and the board. An AI answer with no link to account and voucher cannot be defended.
The expert panel in the survey points the same way: AI needs order in the source data and adapted ways of working before it can strengthen decisions in practice 1. The bottleneck is not the model but what it works on.
Four levels of AI in the finance function
| What AI does | Data | Effect on decisions | |
|---|---|---|---|
| Writing help | Texts, emails, summaries | What you type in | None |
| Analysis of an export | Explains numbers you paste in | A snapshot | Small, hard to check |
| Analysis on connected data | Answers questions about actuals, variances and trends | Current accounting for all companies | Better basis, traceable to the source |
| Decision support | Runs scenarios and shows the effect on profit and cash | Actuals, budget and forecast | Compares choices before they are made |
How to take the step to deciding with AI
- Start with the decisions, not the tool. Write down the five decisions that come up most often: hire, invest, price, cut back, finance. Ask what basis each decision needs.
- Connect the data instead of exporting it. Let accounting and bank be read directly, so the AI always works on the latest actuals for all companies.
- Make the numbers traceable. Every number in an AI answer should trace back to account and voucher. Then the decision can be explained.
- Run scenarios. A decision is a choice between alternatives. Let AI work out what each alternative means for profit, balance sheet and cash, with assumptions you set yourselves.
- Follow up afterwards. Compare the outcome with what you expected when the decision was made. That is how both the model and the gut feeling get better.
How it works in Fintower
Fintower connects accounting and bank for every company in the group, such as Fortnox, Visma, Hogia or Business Central, and keeps the data up to date. You ask in plain language and the AI works on the whole picture, with every number traceable to account and voucher.
The budget and the rolling forecast build on what you have already booked. Set up scenarios for different assumptions and see straight away what they mean for profit and cash, by company and for the group. Forward-looking views are marked as indicative and you set the assumptions yourselves. The data is stored within the EU and is not used to train AI models.