For years, a bank's idea of artificial intelligence was a chat window that answered questions about branch hours. That's changing. Some banks are building tools that can take actions for their business customers. Depending on the bank, that might mean sorting and flagging transactions, lining up payments, moving money between your own accounts, or working an exception, all inside limits you set.
I think some of this will be genuinely useful. I also think a lot of companies will switch these tools on without asking the questions they'd ask of a new employee with the same access.
Nicole is the controller at a food distribution company with about sixty people. Her bank invited her to try a new assistant inside its online banking platform. The demo was impressive. It could look at the company's payables, suggest which bills to pay and when to protect cash, and line up the payments for approval. She liked it, and she was ready to say yes.
Then she asked the one question I wish every controller would ask. "When this thing lines up a payment, does it go through the same approval rules my people do?"
Nobody on the call was sure. It took two weeks and a conversation with the bank's treasury team to get a clear answer, and the answer turned out to depend on how her company set it up. Some of her existing approval limits wouldn't have carried over on their own.
Nothing bad happened. But if she'd switched it on that afternoon, the dual approvals she'd worked hard to build would have had a side door.
Here are the questions I'd want answered before your company uses any AI tool from your bank.
What can it do on its own, and what can it only suggest? Get this in writing. "It helps with payments" isn't an answer.
Do your existing controls apply to it? Dual approval, payment limits, user permissions. If a person needs a second sign-off for a wire, the tool should too.
Who in your company can turn it on, change its limits or connect it to other systems? That should be a short list, and it shouldn't be the same person who approves payments.
What record does it keep? You want a log of everything it did or suggested, who approved it, and when. If you can't review what it did, you can't control it.
Who's responsible when it gets something wrong? Read the agreement or addendum that comes with it. If the tool acts on a bad instruction, or a fraudster finds a way to talk it into something, you want to know where that loss lands before it happens.
Do your people understand it? The tool is only as careful as the person reading its suggestions. Someone on your team needs to know what it's doing well enough to say "that doesn't look right."
My advice is to start small. Turn on the parts that read and report first, and watch them for a couple of months. Add the parts that move money only after you're comfortable, and only with your approvals standing in front of them.
Your bank's AI might turn out to be the hardest worker in your finance department. Just give it the same onboarding you'd give anyone else with access to your money.

