For most companies, the decision to deploy AI in customer support is born out of necessity. Queues are too long. Costs are too high. Something needs fixing. Capital on Tap's story is different.
The UK's leading small business credit card provider had already built something rare: 15-second phone answers, sub-minute chat response times and CSAT scores their competitors could only aspire to. For the small business owners who rely on Capital on Tap for credit cards, current accounts and savings products, support was one of the reasons they stayed.
So when AI Operations Lead Tilly James began exploring AI agents, the question was not how to fix a broken experience. "We were not motivated by wanting to cut costs," she explains. "The motivation was making sure we are building in a scalable way, so we can maintain our excellent customer experience as our business expands its offerings and enters new markets.”
The pyramid problem
Capital on Tap's support team, made up of 350 people based in Cardiff, UK, running a 24-hour operation, was performing well. But Tilly could see where the pressure points would emerge. "There is a clear pyramid of customer queries," she says. "At the bottom is a vast volume of questions that are straightforward to resolve, but not the best use of our team's time." The goal was not to replace human agents but to reserve them for conversations that genuinely needed their expertise, while every other customer still received a fast, personalized answer.
Knowing when to stop building
Capital on Tap initially explored building an AI platform on their own. A small internal team spent months building a RAG-based system from scratch, but the finish line kept moving. "The horizon of actually shipping was never getting close," Tilly recalls. "Authentication, guardrails, abuse detection - the less visible parts of the solution kept adding complexity and pushing it further out."
After a competitive evaluation, Sierra stood apart. Sierra's technical depth made the full journey feel achievable from day one. "It felt like it was built for AI people," Tilly says, "rather than a customer service platform moving into AI."
What the agent does for customers
Capital on Tap's AI agent is built around one principle: be more helpful than necessary. Before a conversation starts, the AI agent calls multiple tools to build a complete picture of the customer, grounding every response in real, current data rather than generic guidance. "We did not want to push customers towards a knowledge article," Tilly says. "What we try to do is say, for example, your credit limit is this. And if you want to find it yourself next time, here is where to go."
Today Capital on Tap’s AI chat agent handles over 60% of all conversations across 12 journeys. Key things customers can now do include:
Credit limit reviews. The agent checks eligibility, underwriting queue capacity and open banking status across three tools, delivering an instant answer to one of Capital on Tap's most common and complex contact drivers.
Card declines. Handled with care and the right level of disclosure, giving customers what they need to move forward without overstepping on sensitive information.
Statement balances and repayment guidance. Real account data contextualised in a clear explanation, helping customers build financial literacy rather than just resolving a ticket.
Bringing the team along
Introducing AI to a high-performing team that had already seen one internal build fail required trust, not just technology. "There was a healthy amount of scepticism," Tilly says. "We owned that we had tried this before and it did not work."
Capital on Tap built agent feedback loops from day one, letting support staff flag conversations for review. When simpler queries moved to AI and harder ones stayed with the team, bonus structures were adjusted so agents were not penalised for handling more complex work. "We are involving the team in the parts of the work that make it interesting," Tilly explains. "They are driving the continuous improvements of our AI agents."
Results that kept beating the target
Capital on Tap launched with a containment target of 30%. On the day they went live, they hit 35%. "Every time we have set targets, we have had to increase them midway through the quarter," Tilly says.
Today they sit at 60%, on track for 65% by the end of the quarter. The agent is consistently outperforming both human-only and human-escalated CSAT scores.
Customer readiness also exceeded all expectations: Capital on Tap anticipated 15% of customers would immediately request a human agent. The real figure was 4%. "If you had told me that at the start, I would not have believed it," Tilly admits.
The pace was made possible by a genuine partnership. Capital on Tap launched with one engineer and Tilly on the project. "We have been able to do the work of ten people," she says, "because we have had experts from Sierra working alongside us from day one."
What comes next
Capital on Tap is expanding the agent into voice, web, WhatsApp and the US market. But the ambition goes further. "We are reimagining every part of the business as if we were building it from scratch," Tilly says. "What would a truly AI-native company look like?" For Capital on Tap, that means not just automating support but helping small business owners understand their finances more deeply, make better decisions and navigate the pressures of running a company without a dedicated finance team. AI is not just a way to scale their support. It is a way to deepen it.
About Capital on Tap
Capital on Tap is the UK's leading small business credit card provider, on a mission to be the one-stop shop for small and medium business finances. Offering credit cards, current accounts and savings products, the company serves customers across the UK and US from its base in Cardiff.
