Behind the Build: One agent, many brands at Imprint.

Imprint builds co-branded credit cards for brands like Booking.com, Shell, and Rakuten. Each partnership means a new set of customers who think of themselves not as Imprint cardholders, but as Booking.com cardholders, or Rakuten cardholders. The brand always comes first.

"If they just hear Imprint, they know they're an Imprint customer, but to really feel at home, they want to hear 'your Rakuten card, powered by Imprint,'" says Will Larson, Imprint's CTO. "That's where it feels right for the cardholder, but also for the brands we work with."

The expectation that every touchpoint reflects the partner's brand, not Imprint's, shapes everything the company builds. So when Imprint set out to deploy an AI agent for customer support, the challenge wasn't just automation. It was building one platform that could become a dozen different brands, each with its own voice and vocabulary, in an industry where sponsoring banks need to understand and approve every behavior.

Starting with experience, not cost

It would have been easy to frame the project as a cost play. That's where many companies start when they think about AI and customer service: same quality, lower cost.

Larson rejected that framing early. "Neither our cardholders nor our partners care about our cost structure," he says. "We really wanted to figure out how to make a great experience first. And then also do it at an affordable cost."

Before Sierra, Imprint's support tooling was throttled by the availability of customer support agents. When they were available, the service was thorough, but sometimes the queues would back up, and wasn’t always available outside of business hours. The support experience was branded, but sometimes felt surface level rather than deeply integrated.

Imprint needed an agent that could do real things, like processing payments, checking balances, and issuing replacement cards, while reflecting the specific voice, tone, and customer needs of each individual brand partner. And they needed to do it without building a bespoke system every time they onboarded a new brand.

Customizing across brands

Sierra's configuration layer, called their Agent Studio, gave Imprint a way to manage brand-specific preferences, like tone, pronunciation, how each brand refers to its cardholders, and how the brand refers to itself and experiment quickly.

The approach they landed on is overrides-driven: a strong agent baseline, with brand-specific adjustments layered on top. Use this term, not that one. Refer to cardholders this way. Match this tone.

"The platform is infinitely flexible, which is kind of a pro and a con," Larson says. "The freedom to customize anything can be overwhelming. So we've gotten more disciplined about focusing on what actually matters for each brand, and using overrides to handle specific issues as they come up."

Making it legible to regulators

Imprint operates in a highly regulated space. Their sponsoring banks need to understand how the system works. The way you typically document an IVR is essentially a tree diagram — here are the states, here are the transitions. Very legible to an auditor.

Imprint's agent doesn't work that way. Rather than following a rigid decision tree, the agent executes against an identified goal, intelligently selecting the tools needed to resolve whatever problem a cardholder is working on. That's a better experience, but it's harder to put in front of an auditor.

The solution was a mode-based architecture. Imprint defined three major modes: logged out, logging in, and logged in, plus a fourth for customers with multiple accounts. Within each mode, they mapped the available tools and documented what's possible in each state.

From there, they focused on the two things the banks care about most: authentication, and the specific places where exact regulatory language is required. Sierra's verbatim tags guarantee that precise legal language appears exactly as written, every time. And Sierra's simulation testing gave Imprint full logs of every test case — something they could put directly in front of auditors to demonstrate coverage.

"It is genuinely complicated to reason about how an intelligent agent selects tools," Larson says. "Credit to the banks. They worked through that with us rather than just defaulting to 'give us a tree diagram or we can't approve it.'"

What cardholders experience

For the most common tasks, like checking a balance, making a payment, or requesting a new card, Imprint's agent is faster than a human. No hold time, immediate verification, and then the cardholder does what they came to do and they're done.

The open question going in was whether cardholders would actually engage with an AI agent, or just force-escalate to a human. Some do. Imprint's agent respects that. But Larson has been encouraged by the willingness he's seen, especially when cardholders have a positive first interaction on their own terms.

Imprint's support team remains central for complex problems that require judgment. But the rate of improvement on the AI side is something Larson says couldn't happen with a purely human approach. Every interaction gets automatically classified — what worked, what didn't, what type of issue — and the team uses that to close gaps continuously.

What's ahead

Imprint is now expanding into a pre-authenticated in-app experience. Instead of requiring OTP verification over the phone, the agent can leverage the biometric auth cardholders have already used to log into their phone, acting on their behalf immediately with no duplicated authentication step.

Larson is also watching an unexpected inversion take shape: because new tools can be added to the agent so quickly, the voice experience could become the fastest place to ship new functionality — faster even than building it into the app.

"You could imagine demoing capabilities to a brand partner by just calling the phone number," he says. "That's a complete inversion of how IVRs have traditionally worked, where voice is always the most under-featured channel."

Imprint set out to build one agent that could represent many brands. What they ended up with is an agent that's not just keeping up with their growth. It's enabling it.

Will Larson is the CTO of Imprint. Previously, he served as CTO at Carta and Calm, and held engineering leadership roles at Stripe and Uber. He is the author of An Elegant Puzzle, Staff Engineer, The Engineering Executive's Primer, and Crafting Engineering Strategy.

[Behind the Build is a series where Sierra customers tell the story of how they designed, deployed, and scaled their AI agents.]