Tools

What to Check Before Trusting a New AI Tool With Client Data

New AI tools launch constantly, and it's tempting to adopt whatever looks impressive in a demo. Here's a practical checklist to run through before trusting a new tool with anything client-facing.

Test on low-stakes work first, always

Before using any new tool on active client work, test it on something low-stakes first, a draft with no real deadline, or a task only you will see the output of. Skipping this step because a tool looks polished enough to trust immediately is a common and avoidable mistake.

Check what happens when it's wrong, not just when it's right

Most demos show a tool's best-case output. What matters more is how it fails. Does it clearly signal uncertainty, or does it confidently present wrong information the same way it presents correct information? Tools that fail loudly are safer to use than tools that fail silently and convincingly.

Consider what happens to the client's data

Before connecting any tool to a client's email, calendar, or documents, check the tool's actual data policy, not just its marketing claims about security. This matters especially for any work involving direct access to a client's private business communication.

Ask whether it solves a real bottleneck, or just looks impressive

Some tools are genuinely capable but don't address an actual bottleneck in a specific workflow, they're solutions looking for a problem. Tools that stick long-term tend to be the ones addressing a specific, recurring friction point that already existed, not the ones with the most impressive individual feature.

The checklist, condensed

This slows down adoption of new tools slightly, but tends to prevent mistakes that cost more time and trust than the delay itself.