Someone is standing at a gas pump with a card that will not go through. They need to get to work, or pick up a kid, or buy groceries in the next twenty minutes. In these moments, they rely on Tilt.
“They’re trying to pay for food. They need gas so that they can go to work or pick up their kid,” says Delan Diaz, Senior Manager of AI-Enabled Operations at Tilt. “Oftentimes, these people need access to cash right away. ‘Hey, I’m at the gas station. I need money. What’s happening here? Is there a delay?’”
Tilt exists to serve exactly that moment. Tilt is a better way to access money, helping more people get cash and credit by looking at real-time income and spending, not just credit scores, to see their potential. This opens the door for people the traditional financial system tends to leave outside it. “Tilt is the brand that believes in your potential before your credit score does,” Diaz says. “By embracing a philosophy of making access to cash and credit fair, fast, and flexible, we’re opening a door to opportunity that isn’t traditionally available.”
That mission puts unusual weight on customer service, because at Tilt the support conversation sits directly on top of the transaction. A member who cannot log in cannot take the advance. Every one of those calls decides whether the product works for that member or doesn’t, which means the operations team is running a growth function as much as a service function.
“If a member can’t even get into their account, nothing else matters until that’s solved,” Diaz says.
The reframe: prove it on voice, or don’t bother
Most companies deploying AI agents start with chat. It is the friendliest channel, the most forgiving of latency, and the easiest place to show an early win. Diaz went the other direction on purpose.
“Everyone seems to be able to handle chat. That’s where everybody starts. It’s sort of the easiest,” he says. “People don’t want to talk to an AI agent. You really feel the latency there more than other channels. I think voice is a really good measure for how well-built an AI agent is.”
The logic was that voice would tell Tilt the truth fastest, about the technology and about the vendor behind it. “If you can get the hardest thing done the best of all the competitors, you’re going to get the easy thing done as well.” Voice was also entirely un-automated at Tilt, so anything built there was net new capacity rather than a migration of work that already ran.
The bar Diaz set was not a containment target. “It just needs to be better than a human. That’s the bar we set for AI at Tilt.”
The blind taste test
Tilt evaluated several vendors, took several calls with each, narrowed to three, then cut to two once it became clear that no team had the bandwidth to run three production-grade pilots at once. Over two weeks, Tilt’s biggest contact driver was implemented on each finalist platform.
Then Diaz ran the comparison the way a product team runs a taste test. Employees got three phone numbers — including Sierra and two other competitors — no labels, a snapshot of the test data, and a survey at the end.
Employees across the business called, including several members of the senior leadership team. Everyone rated what they heard and left written feedback, which Diaz scored and parsed.
“That was really fun, really engaging for the company,” he says. “Creates a lot of excitement about the AI agent. That’s a good way to get everybody involved.”
Sierra won on the things voice punishes: latency, interruptions, ambiguous requests, and members raising two problems in one breath. “The voice sounded great. It did really well at handling interruptions, disambiguation, and multi-intents, and just navigated that whole process really smoothly.”
The other half of the evaluation had nothing to do with the demo. Tilt runs hard on its vendors by design. “We’re always using the latest features. We’re always asking first for the next thing that’s coming.” Diaz wanted to know whether the team on the other side would keep up. “Are we going to push each other in a healthy manner that allows both organizations to be more successful?”
What Tilt built, and where it stops
Tilt’s agent handles the work that makes up the bulk of member contact, and nearly all of it maps to something the business runs on. It reviews recent payments and cash advance details. It checks advance eligibility and adjusts repayment dates, which lets a member move a payment before it fails rather than after. It cancels subscriptions and processes eligible fee refunds, the conversation that decides whether a member stays. And it clears the access problems, card activation, card closure, login, and profile updates, that otherwise stand between a member and the cash or credit they came for.
Diaz picked that scope deliberately. “We really try to look at where users have the most friction. What are our high volume drivers and what are our highest friction support points? Where can we strike the right balance to ensure we’re expanding access to cash and credit for as many folks as we can?”
Just as deliberate is how the agent is governed. Compliance was designed into it rather than bolted on afterward, and the routing rules are part of that design: regulatory and compliance-sensitive matters go straight to a specialist trained for them, by policy. Verification is handled the same way, calibrated to which products a member holds and where they are in their journey, because the requirements differ. Diaz built those rules with the people who own the risk. “We’re collaborating with product, engineering, compliance and legal.”
The result is an agent that can act confidently inside a regulated business, because the boundaries it operates within were drawn by the teams accountable for them.
Tilt also staffs the agent like a product. Two people work full time on the agent experience, and several hundred simulations run continuously against it. Scope expanded only after containment and quality held up in the real world. “We started narrow, not ‘handle everything.’ That combination is what let us feel good handing over our highest-stakes channel.”
Tilt’s team also works alongside Ghostwriter, Sierra’s AI agent for proactively building and improving agents, which can now join the team wherever they work, like in Slack. For Diaz, it means the agent’s expertise is never more than a message away. “As a program owner, I have 24/7 Slack access to an expert in my team’s AI agent,” he says. When a conversation he was reviewing pointed to a possible issue in one of Tilt’s journeys, he asked Ghostwriter from his phone whether the behavior was built into the journey and got an answer right away. “By the time my specialist logged on and reviewed the thread, Ghostwriter and I had already diagnosed the issue, found the root cause, and had a summary ready to go.” Having Ghostwriter in Slack has already saved the team a significant amount of time, showing up as a trusted AI teammate.
What the team does with the room it got back
The result Tilt points to is how much more the business can take on, and how much better its people are at the part only people can do.
New products used to arrive with a service tax attached. Every launch meant new contact drivers, new SOPs, and a team stretched thinner across products it had only just learned. That math has changed. “We have so many new product offerings that our team is now able to dive into,” Diaz says. “It’s allowed us to really operate completely differently and properly skill out our team to develop real subject matter experts across our products.”
The old model spread agents thin by necessity, which is how members ended up transferred between skill-based queues to find someone who knew their product. Now the routine access and payment questions resolve without a person, and Tilt’s specialists go deep instead of wide. The team also got something it never had, which is time to work on the program rather than in it. “We have had bandwidth to really analyze how we want to run our customer service program and make the decisions to help us get there.”
Members feel it at the front end. Wait times are significantly reduced, and the transfer loop has largely gone away, which matters most for the member who needs an answer before they leave the gas station. When someone does reach a person, they reach a better prepared one. “When customers need advanced service, you get better white glove treatment, because the support team isn’t bombarded. We have the capacity to actually go in and resolve that issue.”
What’s next
Tilt is onboarding in-app chat and expects to be fully omni-channel in the coming months. International expansions are the next frontier. Underneath the roadmap is the same test Diaz applied on day one. “It’s always about the customer experience first. Is this a quality experience, or is this not? Does it actually do useful things for them? Or are we just throwing AI in front of it to throw AI in front of it?”
For someone standing at a gas pump, that is the only question that matters.
