Every sitcom has the character who knows everyone's business and never takes a day off. In greater Philadelphia, that character is named Gigi, and she isn't a person.

On November 12, SoundHound AI announced that Gigi, the voice and chat AI agent it built with American Heritage Credit Union, handled 598,000 member conversations in 11 months. That's not a pilot. That's not a demo at a trade show. That's a 325,000-member financial institution letting a bot be the first voice most members hear.

And banking, of all industries, is where people are least forgiving about getting it wrong.

What Gigi actually does all day

According to the announcement, Gigi works American Heritage's phone and chat lines around the clock and handles more than 30 types of member requests. The list reads like a front-desk job description: verifying members, reading balances and recent transactions, moving money between accounts, resetting online banking passwords, activating and replacing debit cards, quoting loan payoff amounts, and finding branches with their hours and services.

On the phone side, Gigi averaged 1,897 conversations a day between July 2025 and May 2026. Picture that many calls landing on a human team every single day, including at 2 a.m. on a Sunday when nobody is there to answer.

She also quietly screens out silent and dropped calls, which the release puts at about 5% of inbound voice traffic. That's a small, unglamorous win, and those are often the best kind. Nobody on staff has to pick up the phone just to hear nothing.

The numbers that matter

Here's where it gets interesting. Over that July-to-May stretch, Gigi resolved 88% of fraud and dispute chats without a representative getting involved. Account maintenance chats came in at 78%, and rate and certificate questions at 70%.

Fraud and disputes are the stressful conversations, the ones where a member is already anxious. The fact that those had the highest resolution rate says something: when the questions are predictable and the process is clear, an AI agent can handle them faster than a hold queue can.

The release also says Gigi has handled a bigger share of requests on her own every quarter since tracking began. In other words, she got better at the job over time, which is more than can be said for most of us and our inboxes.

The handoff is the real flex

The part most businesses overlook isn't what the AI handles. It's what happens when it can't.

When a member needs a human, Gigi passes the whole conversation to a representative in one click. The rep sees every question and answer plus the member's verification status. The member doesn't repeat their account number. The rep doesn't re-verify. Nobody starts over.

If you've ever explained your problem to three different people on one call, you know why that matters. A good AI agent doesn't replace your team; it hands them a fully briefed customer instead of a frustrated one.

One American Heritage digital analyst summed up the member experience simply: "You feel like you're talking to an actual person when you're talking to the bot."

What's next: from answering to anticipating

American Heritage isn't stopping at answering questions. The credit union says it's adding more complex workflows and building proactive alerts. After verifying a caller, Gigi will check the account for issues, starting with mail flagged as undeliverable, and confirm or update the address on the spot. Unactivated debit cards, delinquent loans and negative balances are next on the list.

That's the shift worth watching: the agent stops being a smarter voicemail and starts fixing problems the customer didn't even call about.

What this means for your business

You don't need 325,000 customers or a $5 billion balance sheet to take the lesson here. Most of Gigi's wins came from routine, repeatable requests: balances, passwords, card activations, branch hours. Every business has its own version of that list, whether it's booking appointments, quoting prices, confirming hours or checking order status. Start by writing down the ten questions your phone rings with most, figure out how many of those calls go unanswered after hours, and make sure any AI you put on the line can hand off to a human with the full context attached. That's where the time, and the revenue, is hiding.