Can an AI chatbot calculate your sales commissions? Here's why it shouldn't

It's the end of the month. Commissions are due. Someone exports the deals, pastes them into an AI chat and types: "Calculate the commissions for each rep."

Thirty seconds later there's a clean table with a number next to every name.

It looks like the problem is solved. It isn't. Here's why.

1. It only sees what you paste

A chatbot doesn't know your business. It knows the export you gave it, at the moment you gave it.

It doesn't know that a deal was lost last week, that ownership changed in the middle of the quarter, or that the plan was updated in March. It can't go and check, because it isn't connected to your CRM, your plans or your payment history.

So it calculates on a snapshot that was already out of date when you copied it. And when a rep asks a follow-up question, it can't look anything up. It can only guess from the same pasted table.

2. It doesn't give the same answer twice

General-purpose AI is probabilistic. Ask the same question twice and the answer can change. Long tables of numbers are exactly where small errors slip in: a skipped row, a rate applied to the wrong deal, a rounding that adds up.

For an email draft, that's fine. For someone's pay, an answer you can't reproduce is an answer you can't defend.

3. It has no memory of what was already paid

Commissions aren't a one-off calculation. They're a running record.

What was paid last month? Which deals were already counted? Who owes money back because a customer cancelled? A chat conversation holds none of that. Without a record of what was paid, you can't stop a deal from being paid twice, and you can't recover money from a deal that fell through.

4. "The AI said so" is not an audit trail

The hardest question in commissions isn't "how much?". It's "why?".

When finance asks how a number was calculated, or a rep disputes their statement, you need to show the sale, the rule that applied, and the version of the plan in force at the time. A chat window can produce a number. It can't produce the evidence behind it.

The audit trail matters as much as the calculation itself.

5. You may be sharing data you shouldn't

Commission data is pay data about named people. Pasting it into a general-purpose tool your company hasn't approved can mean sending personal information outside the systems meant to protect it. In Europe, that's a conversation your data protection officer will want to have before it happens, not after.

So when do you actually need a system?

Not at a magic number of reps. The real tipping point is this:

when checking the exceptions takes longer than running the commissions.

A spreadsheet, or a chatbot, can handle a surprisingly large team when plans are simple and the data is clean. Both start to break as soon as you add split deals, mid-period changes, cancellations after payment, or rates that depend on quota attainment.

Five questions before you trust any tool with commissions

1. Does it read your real data, or only what you paste into it?
2. Will it give the same result tomorrow from the same inputs?
3. Does it remember what was already paid, and stop it being paid twice?
4. Can you trace every number back to the sale and the rule that produced it?
5. Does your pay data stay in a system your company controls?

If the answer to any of them is no, you have a calculator, not a commission system.

Where AI does belong

None of this means AI has no place in commissions. It means it belongs on top of your data, not instead of it.

That's the approach we took with Evenq. The calculation is done by a deterministic engine: same inputs, same result, every time. Zeke, the AI assistant, never calculates money. It answers questions like "where does this amount come from?" using your real payouts, pay runs and balances, with the same permissions as the screens you already use. And when it can't answer, it says so instead of guessing.

AI for the questions. An engine for the money. A record for the proof.