Using AI without leaking your clients' data

A finance broker stopped me at BNI a couple of weeks ago with a question I keep hearing in different forms.

I’d love to use AI on the documents I deal with all day. But these are people’s financials. How do I do that without their private details ending up somewhere they shouldn’t?

It’s the right thing to worry about. And the answer surprised him, because it isn’t a tool. You don’t fix this by finding “the private AI.” You fix it with a setup. Three layers.

Layer one: use a tier that doesn’t train on your data.

The free, consumer versions of most AI tools can use what you type to improve their models. For anything with a client’s information in it, that’s the wrong door. The paid business tiers, Claude Team, ChatGPT Team, Microsoft Copilot, all commit to not training on your content. Start there. This one change rules out most of the risk on its own.

Layer two: redact before you paste.

Even on a safe tier, good habits matter. Before a document goes anywhere near AI, swap the identifying details for placeholders. The client’s name becomes [CLIENT]. The ABN becomes [ABN]. Dollar figures become [AMOUNT]. You get the AI to do its work on the de-identified version, then put the real details back yourself at the end. After a week it’s muscle memory, not a chore.

Layer three: separate your environments.

This is the one that turns careful habits into a system. Keep two spaces. One for general work, where client information never goes. A second, locked down, for client documents, where the setup itself carries an instruction never to repeat back a name, an ABN, or a dollar figure.

In practice that instruction lives in a small file the AI reads every time. A line or two is enough:

Never echo back client names, ABNs, or dollar amounts. If you need to refer to them, use placeholders.

Once that’s in place, privacy stops depending on whether you remembered to be careful today. It’s built into how the tool behaves.

Try this: pick the one document type you’d most like AI’s help with but have held back on. Set up those three layers for it, just that one. A safe tier, a redaction habit, and a separate space with the “never echo back” rule. See how it feels to hand that work over without holding your breath.

That’s the difference between using AI and operating it safely. Most finance practices haven’t made it yet. The ones that do will look, in a year or two, like they were early.

If you’d rather have the whole thing set up for you in one sitting than piece it together yourself, that’s what I do. Either way, the three layers are yours to use.