Self exists to help people build credit. Some of its customers are 18 with no credit history. Others missed a payment and cannot get out from under it.

"It can be tough for people to get out of a slump if they missed a payment or have poor credit," says Levi Benotti, Manager of AI Quality and Governance at Self Financial. "That's what Self is for, and what Self has built: that person who picks you back up."

Levi has worked at Self for six years. He has watched customers start a Credit Builder Account, make their payments as Self reports them to the credit bureaus, and move on to buying a car or a home. "That's one of the biggest perks of being here," he says.

Before his current role, Levi worked in quality assurance. The work was manual. "I would go in and check every single ticket, check every single call from start to finish manually," he says. "With doing that manual, you can only do a certain amount per day."

The rest of Self worked the same way. "We have always been a manual company where we rely on analysts and agents, for back office, for CS," Levi says.

Today his title is Manager of AI Quality and Governance. He tests the workflows Self builds with Lorikeet, and he builds automations for quality work. He is also candid about the job he left behind. "Before I got into AI, before my role focused on it, the job was repetitive," he says.

Much of Levi's work is on the voice line. When a customer calls Self about their Credit Builder Account or credit card, Lorikeet's voice agent answers. Callers do not always know how to describe the problem. The agent pulls the customer's account data from Self's internal system of records and matches it to what the customer says.

"AI and Lorikeet's really able to pick up and understand where the customer's coming from from pulling all their data and getting the information that they're giving," Levi says, "and to be able to pull the exact kind of situation, the problem that the customer's talking about, and go in there and fix it."

The agent then acts on the account. It can replace a card, close a Credit Builder Account, and explain a block on a card. Each of these once needed a person. Each action runs with oversight, escalation paths, and compliance guardrails. "These are things that a agent would have to go into [Self's internal system of records] and go through a lot of manual processing to action that item," he says, "where now Lorikeet can go in there in the conversation and automate the CBA closure or automate a card replacement."

Answers that a human agent had to find in Self's internal system of records now come back in the same conversation. "We've been able to build where it pulls all the API calls from [Self's internal system of records] and is able to give a real answer back to the customer within seconds," Levi says.

Each workflow goes through rounds of testing before it goes live. Levi works on them with the Lorikeet team. "Is this working? Is this not working?" he says. Then comes "the happiness that comes out of it once it's finally in production and it's working correctly."

He describes the Lorikeet team as "super smart, big problem solvers." Self brings the problem. "We will give them the data, the information, the idea of what we want, and they're really quick and really smart to come back with a workflow that fixes the problems we've had."

Levi also rebuilt his own QA work. Starting from what the two teams had already built together, he used Claude to build an automation that grades tickets against Self's rubric, split his own way, for example by agent action. A second automation logs each issue it finds in Self's issue tracker.

"It can go through hundreds of cases, hundreds of tickets," he says. "Just like I would grade it personally, it would do that through the automation."

Levi's advice is for people in roles like his old one. "I think it's a great opportunity, and you run with the opportunity," he says. "So I would say just run with it, keep learning from AI, keep pushing yourself, keep testing yourself."

Self is now extending the agent's capabilities. Levi expects the agent's work to keep growing. "The learning opportunities is literally limitless," he says. "Every single day, there's something new that's launching with AI, something new that you can perform with AI, something that you can learn from AI."