For customer experience leaders, adopting AI can feel like navigating a minefield of vendors, metrics and compliance risks. Success depends on operational discipline, governance, vendor alignment and careful integration into existing workflows.
US-based Chime, an app-first financial services company, has found that treating AI customer service agents like full-fledged team members can boost resolution rates and give buyers a clear roadmap for working with vendors to scale technology effectively.
As Janelle Sallenave, Chief Experience Officer at Chime, told CX Today in an interview:
“We made sure to design the ecosystem so that we could treat these bots the same way we treat agents.”
The fintech’s AI chatbot and voice bot now handle around 70 percent of its customer support. The quality of that experience, measured by CSAT, has increased by more than 50 percent in some channels. “That makes sense if you’re comparing a really good GenAI bot versus… the endless tree of the IVR. Of course, the support feels better. But for us, what is also really important is that the resolution rate is up almost 40 percent,” Sallenave said.
What does it mean in practice to treat these AI bots like human support teams?
Holding AI to the Same Standard as Human Agents
Chime applies the same quality assurance (QA) and governance frameworks to its AI chatbots that it uses for human agents, Sallenave explained.
“We don’t view talking to the chatbot—her name is Jade—any differently than talking to an agent named Sally. Those both produce transcripts from that live interaction, and we put them through the QA.”
“The data source that they come to is no different than an agent. The SOP and the underlying policy are all exactly the same. So the bot is as empowered as a human agent.”
This ensures that Chime’s AI-based interactions maintain quality, brand voice and compliance. By treating AI as a full-fledged agent, the fintech is creating measurable and auditable interactions, transforming AI from a black box into a partner in the support operation.
“It was a great way to jumpstart the program because we have spent so many years finding the right QA environment for a human agent.”
That approach is helping Chime “not just identify the areas where a live agent might need to be coached; it is giving our team data about where we might need to make adjustments in the model,” Sallenave said. There is value in “being able to get that insight and bring it back to our GenAI partners and say, we’ve looked at it and in this area the bot isn’t doing what we really intended.”
Aligning Vendors to Your North Star Metrics
One of the biggest challenges tech buyers face is ensuring that vendors are aligned with an organization’s goals. While vendors focus on metrics like containment or cost reduction, Chime prioritizes quality of resolution and customer satisfaction, Sallenave noted.
“We don’t view our work in AI as being about cost savings, ‘hey, let’s replace people.’ It’s really for us more about ‘how do we remove friction in the economy.’”
To that end, Chime established two “hero metrics,” automated resolution rate and CSAT. As Sallenave explained, “It’s not that hard to make one rise at the expense of the other. Our real challenge to them was we need both.”
Chime uses two different vendors for its chatbot and voice bot for “optionality”. The fintech views those relationships as strategic partnerships, rather than a straightforward customer-vendor dynamic, Sallenave said.
“We’ve invested the time and energy with both of those companies so that our goals are their goals and vice versa. They care as much about the CSAT as we do, not just about containment rate.”
That has taken some adjustments along the way for the vendors to internalize Chime’s priorities and rethink their usual instincts around optimization. “We’ve had a trip up here or there where we do our analysis and we’re saying, ‘why did that automated resolution rate up jump so quickly?’ And we’d discover some engineer somewhere, deep in the company that we work with, added a little bit of friction to get to a live agent,” Sallenave said.
“And we have to unwind that and use that as an opportunity to say, ‘wait a second, aren’t we strategically aligned about what we’re accomplishing?’ So that’s been for us a journey and it’s taken us probably six plus months to get to a good spot.”
Sallenave underscored how important it is for buyers to define a clear North Star and ensure the vendors they choose are committed to achieving it.
“What’s really important is to be very clear about what your North Star is. There’s nothing wrong with having a North Star of saving money… For some of our back office queues that don’t face the customer directly, that is our North Star. We want to automate and we want to save money so we can spend it somewhere else.”
“But for our chatbot and our voice bot, we have to, from day one, not lose sight of what our North Star metric is. And in our case, it is about resolution rate and quality of the experience.”
As voice and chatbots handle more of its customer interactions, with volumes to human agents down from around 50 percent to 20 percent, Chime has reduced its service headcount, although Sallenave noted that this has not been at the same rate as the handoff from human agents. The interactions that human agents continue to handle are the more complex issues that take longer to resolve, Sallenave said. “[T]hat is where agents really add their value.”
The “Art and Science” of Deciding When to Escalate to Human Agents
Deciding which interactions AI should handle, and which require human judgment, is a key consideration for tech buyers. Chime approached this as both “an art and a science,” starting with simpler interactions and gradually expanding.
“We came into it with a certain set of hypotheses of ‘these ones should be automated last because they are the most difficult, the most risky, the most emotional for our members.’” Sallenave said.
Having started with a set of tasks that were already being handled by Chime’s old IVR technology, the company has set a roadmap to add new features to the AI system each quarter since it started in late 2024.
“We are now sitting on a robust data set that is teaching us not just when should we automate or not… We are also learning which channel is better, because for certain issue types, we are just learning it is better when the bot handles it on the phone versus when the bot handles it through chat.”
Chime’s intensive preparation, consolidating content into a single source of truth, provided a foundation that not only allowed the AI to be integrated smoothly into its ecosystem but also helped prevent chatbot hallucinations early on, enabling the team to catch errors quickly during initial QA iterations before the system ever interacted with customers.
The data also helps the fintech gain detailed insights into the customer journey beyond a specific support interaction, Sallenave added. “Why did you fall off the happy path? How good of a job did the bot and/or the human do to get you back on? What did you then, as the member, do after we got you back on the happy path? The whole ecosystem of data now for us is really phenomenal and wildly insightful.”
Why the Crawl-Walk-Run Approach Sets the Stage for Scale
Introducing GenAI to customer interactions in sensitive industries like financial services carries risk, making it essential that organizations carry out controlled rollouts. Chime adopted a crawl-walk-run strategy, starting with internal employees who understood the member experience, then BPO agents and finally a small percentage of members before full deployment, Sallenave said.
“We spent five months doing that. But once we were ready to go and we put it in front of that first 1 percent of our customer base, from the time we did 1 percent of our customer base until we GA’d it to 100 percent, it was two months. That’s very, very fast. And it’s because of all of those crawling steps that got us there.”
This incremental approach allowed Chime to minimize risk, build confidence and ensure the system delivered consistent, high-quality support.
Chime’s experience shows that successful AI adoption hinges on clear vendor alignment and a rollout strategy that treats AI as an integrated member of the support team.
As Sallenave put it:
“The dream of what they’re selling is absolutely possible… You get the right partner and you have the right [commitment] from inside of your company… it absolutely can be realized.”
