Yonder resolves disputes 150% faster with AI agents

From frontline support to AI disputes automation

In order to understand how Yonder successfully automated its disputes cycle, it's helpful to know where this initiative fell in the team's broader AI automation journey for customer operations. After signing with Gradient Labs in mid-2025, Yonder implemented Gradient Labs' AI customer service agent for frontline chat and email. Based on this success, in 2026 the team implemented Gradient Labs' AI lending agent for collections calls on accounts in arrears. After building internal confidence in the AI agents through these initiatives, the team was able to tackle complex disputes work, the kind of back-office automation that truly transforms processes, boosts cost savings, and makes a difference in CSAT. 

The foundation: 90% CSAT on AI customer service 

Yonder began its programme by deploying Gradient Labs' AI agent on frontline chat and email. During onboarding, the agent ingested Yonder's help centre articles and reviewed thousands of historic human agent conversations, capturing institutional knowledge that isn't always written down in a knowledge base. Yonder's team then configured a custom tone-of-voice profile and tested the agent against real member scenarios in a sandbox before going live.

Beyond general FAQs, the AI agent handles genuine account matters: eligibility for promotions, plan adjustments, credit limit increases, upgrades, and resolving account suspensions. These cases involve identity verification, deep knowledge of Yonder's policies, and checks across several sources to give the member the clearest picture of what they qualify for. Yonder's inbound AI customer support agent consistently scores an 80–90% CSAT rating, reaching heights of 69% in resolution rate.

AI lending agent for accounts in arrears 

The success of chat and email encouraged Yonder's team to expand automation to new channels. Yonder implemented Gradient Labs' AI lending agent to call customers in arrears and communicate payment deadlines, secure commitments, and handle sensitive conversations with empathy. The AI agent proved itself by handling complex calls without latency, maintaining a natural tone, and adhering to strict compliance standards with the same guardrails baked into the frontline agents already handling chat, email, and inbound calls. 

The lending agent performs identity verification checks, secures promises to pay, troubleshoots payment-related app issues, and offers a live handoff to a collections specialist when needed. Yonder contacts 80–85% of its collections book each week, and the team optimised for each member's preferred call windows, sending a text notification before some calls to improve contact rates and keep calls convenient.

Pushing the boundaries of AI automation 

With AI customer service agents deployed on voice, email, and chat, and outbound collections calls operating at scale each week, most teams turn their focus to a backlog of back-office use cases. Yonder's team knows that while most customer support automation programmes tend to peak at 60–65% resolution, the most significant cost savings occur past that plateau. 

The 40% of customer operations work that stays manual is often blocked by cases that go beyond frontline conversations and dig into back-office systems. Horizontal agents aren't built for the back-office work beneath the ticket. Gradient Labs' AI agents are: they work back-office cases, communicate with the frontline agents, and share context across conversations.

Yonder saw enormous potential for automation with these agents at their disposal, so the team set a goal to automate use cases that bridge this gap, starting with Gradient Labs' AI disputes agent.

Without the right agent, dispute automation hits a ceiling

Prior to Gradient Labs, Yonder had attempted some disputes automation with AI, but the team had hit the ceiling of what was possible with a single agent that couldn't retain context across frontline and back office systems. Yonder's workflow after that initial optimisation was as follows: 

When a customer raised a dispute, it generated a ticket that the frontline agent identified and passed on to the human disputes team. The ticket often arrived with incomplete evidence and required follow up. So, the human specialist copied the conversation transcript, pasted it into an AI co-pilot with a homemade prompt, and waited for it to classify the dispute, flag missing evidence, and draft an internal note and customer-facing message. The human specialist then copied the internal note back into the ticket, pasted the customer-facing message into Intercom, and sent it. When the customer replied with evidence or questions, the whole copy-paste cycle ran again. 

Only after days of back and forth was a case fully evidenced and ready for a human to decide and submit. While already an improvement on the previous process, the target turnaround time was 5 days, but the cycle averaged 5–7 days, stretching to 7–9 during volume spikes.

What disputes automation looks like with connected agents

Gradient Labs' Disputes AI agent operates on the back office layer that its frontline agent doesn't touch, but the two agents can talk to each other, retain context, and act reliably and independently without human oversight. Yonder's team immediately saw how this capability could transform their disputes automation workflow. 

With Gradient Labs, the workflow is streamlined into the following: 

Now, when a customer raises a dispute, the frontline agent asks the right intake questions and hands the case straight to the AI Disputes Agent, which classifies the dispute type, reviews evidence against scheme requirements, and checks for gaps. If something is missing, the frontline agent can contact the customer in the same conversation, in the moment: "your screenshot doesn't show the delivery date, can you send that over?" The investigation resumes as soon as the customer replies. Once there is enough evidence, the Disputes agent records the full case detail and a recommended outcome on the ticket, then passes it to the human specialist team for one job: approve or override, and submit.

The results were immediate. Because evidence gaps are flagged at intake, most intake conversations now wrap up complete evidencing within a single day. Cases reach the human specialist team fully evidenced, so the human review step that once took 5 days now takes 24 hours on average: the team rules on a case and submits it 5x faster than before.

With Gradient Labs, Yonder's entire disputes cycle is 150% faster: cases that took 5-7 days now close in 2. And this timeline holds steady even through volume spikes, which used to cause a 33% increase in processing time. Most importantly, since implementing this new workflow, Yonder has seen customer satisfaction rates climb. Their CSAT rating for disputes has held steady at 90% or higher.