AI can add up your net worth with mathematical precision and still lie to you by 1.6 million euros. Not because the model gets the arithmetic wrong, but because it adds up exactly what you feed it, and if what you feed it is already wrong, the result comes out wrong with all the confidence of a proper analysis. This happened to us, with our own net worth dashboard, and it's worth telling because the bug is nothing exotic: it can happen to any business or person centralizing accounts in several currencies.
The problem: a net worth that didn't match the bank
We built a dashboard that centralizes bank accounts across several countries and currencies to show, at a glance, total net worth. One day, that dashboard displayed €1,628,131 in personal net worth. The real balance in the bank was €1,150.
This wasn't a 10% or 20% error, the kind you notice by eye and investigate as routine. It was a figure a thousand times larger than reality, and the page didn't throw any error: the number showed up clean, with two decimals, looking exactly like a number you could trust.
How it happened: adding up accounts, not movements
The cause was very specific. The dashboard grouped balances by the currency each bank account was registered with, not by the actual currency of each movement inside that account. And some accounts, especially the ones meant for international use, are opened "in euros" while moving money in other currencies underneath.
The result: one account with 1,649,434 Indonesian rupiah in movements and another with -22,580 South Korean won got added to the total as-is, as if those figures were already in euros. A single account with rupiah was enough to send the total flying way above what was actually in the bank.
The fix wasn't complicated once found: convert every movement to its real currency before adding it up, instead of trusting the account's "headline" currency. But finding it took a while, because nothing in the system flagged that anything was wrong.
Why this matters more once AI is in the loop
This is where the problem stops being a dashboard bug and becomes a lesson about how AI gets used in any business. It's increasingly common to ask an AI assistant to read a dashboard and "tell you how things are going", whether that's your business, your cash flow, or, as in this case, your net worth.
An AI you ask "how much am I worth?", fed (or connected to) a dashboard with this bug inside, is not going to question the €1,628,131. It will read it, round it neatly, maybe compare it to last month, and answer you with total confidence. It's not that the AI "gets it wrong" doing the math: it inherits the error from the source data and hands it back with the confident tone of something that has "analyzed" it. That's what makes it dangerous: a wrong number shown on a screen invites someone to double-check it if something feels off. The same number, said by an AI with the air of an analysis, invites you to believe it.
| What you see | What's behind it | Lesson |
|---|---|---|
| €1,628,131 net worth | Rupiah and won added up as if they were euros | A clean number isn't the same as a correct number |
| No error on screen | A grouping-logic bug, not a calculation bug | The system doesn't always flag badly framed data |
| An AI reading that dashboard | Confident answer, with the error baked in | AI doesn't question the data you give it, it processes it |
The decision rule
If your system groups balances, sales, or any figure by a fixed label on the account or record (its registration currency, its default category) instead of by the real data behind each movement, the moment you introduce real variety (several currencies, several rates, several types), the final number can drift far from reality without anything warning you.
And if you're also planning to have an AI reason over that number (to answer questions, generate reports, or recommend something), that discipline stops being optional: it's exactly what we apply in the AI data platform we built for a pharmacy group (see the AI data platform case), where AI never reasons over unvalidated data. The data gets ordered and validated to the cent first, and only then does AI get to draw conclusions. Do it the other way around, and it looks nice and comes out wrong.
Checklist before trusting a multi-currency number
- Am I grouping by the real currency of each movement, or by the account's or record's "headline" currency?
- Do I have any account or process that moves several currencies internally even though it's labeled with just one?
- Does the system convert each movement to a common currency before adding, or does it add first and trust later?
- If a total surprises me a lot (a thousand times more than expected, not 10-20% more), do I check the data before acting on it?
- If I'm going to hand this number to an AI, am I giving it how the number was calculated too, or just the final result?
The lesson for any business
You don't need to operate in 47 countries to run into this. One "international use" bank account, one payment gateway that settles in several currencies, or one supplier that invoices in dollars while your books run in euros is enough. The moment you group by a label instead of by the real data, the risk is sitting there, waiting for your business's real variety to expose it.
It's the same discipline we lay out in why data comes before AI: AI isn't the step that fixes badly framed data, it's the step that amplifies it, for better when the data is trustworthy and for worse when it isn't.
If you're centralizing accounts, sales, or any figure from several sources and want the final number to be trustworthy before letting any AI reason over it, at AutoBoost we build that kind of custom data platform, integrated into your real system. Get in touch and we'll look together at where your number comes from.
