A few weeks ago, my team spent a meeting arguing about whether AI should ever be the first voice a client hears when something breaks. We didn’t fully agree, and I think that disagreement is worth more to you than a tidy answer would have been.

Where AI Helps

Voice AI has gotten good enough that it’s mostly stopped being annoying, which is a bigger deal than it sounds. Someone on my team described calling her pharmacy recently and getting routed by a system that actually understood “I have a problem with my foot and need to talk to the doctor,” and got her there without the usual maze of numbered menus. That’s a real shift from the chatbot era, when the whole point of the experience seemed to be making you beg for a human before it finally gave up and connected you.

There’s a real case for using something like that on simple, well-defined requests: resetting a password, answering a basic account question, routing a call to the right person faster than a receptionist juggling four lines, that kind of thing. People are getting more comfortable with it too, the more they run into versions that actually work on the first try.

Where We Draw the Line

This is what stopped us from just rolling it out. To answer even a simple question like “I forgot my password,” an AI system needs some level of access to our password management tools and parts of our clients’ network security. That’s not a small ask. We manage the systems that protect law firms, accounting practices, and property managers for a living, and handing any part of that access to a system we can’t fully audit in real time isn’t a convenience trade we’re willing to make yet.

There’s a simpler truth underneath it all, too. When someone’s technology breaks, the last thing they want to talk to is more technology. They want a person who understands what’s actually going on and can tell them, in plain language, what happens next. That’s not a preference we’re trying to design around. It’s the job.

This isn’t just a customer experience question either. Many of our clients deal with compliance requirements around who can access what, and when. An AI system sitting between a client and their own account access needs the same audit trail a human technician would need: a clear record of what it looked at, what it changed, and who is accountable if something goes wrong. Most of the AI support tools I’ve looked at aren’t built with that level of accountability in mind yet, because speed and convenience were the selling point, not oversight.

The Real Question Isn’t Whether AI Can Do It

The honest version of this decision isn’t “can AI handle support calls.” Some of it, clearly, it can. The real question is what you’re willing to hand over for that convenience, and whether the people on the other end of the call get an actual choice. My own answer, for now, is that anyone who calls us gets a real person. If we ever build in an AI option down the line, it will be exactly that: an option, clearly labeled, with a human always one step behind it.

If your own firm is weighing something similar, whether it’s a phone system, a chatbot, or anything that touches client data, ask the vendor pitching it one specific question: what does this thing have access to, and who is watching what it does with that access? If they can’t answer clearly, you already have your answer too.

Quick and Easy: AI voice systems have gotten good enough to genuinely help with simple requests like password resets and call routing. The trade off is access, since even basic AI support usually needs some reach into account and security systems. Before adopting one, ask exactly what it can touch and who is accountable for it.