Blog

It’s one thing to be ahead of the curve for one technological trend, but when you are ahead for two, you can be forgiven for suspecting you are very much onto a good thing. 

The potential of agentic AI is well-known, with many industries signalling a fundamental transformation of their processes. IDC notes that agentic AI is emerging as a ‘strategic inflection point’. In the latest FutureScape 2026 research, the analyst forecasts that by 2030, almost half (45%) of organisations will orchestrate AI agents at scale and embed them across business functions. 

Alongside this is the rise of ‘vertical AI’. In September 2024, the brains trust at Bessemer Venture Partners – no fewer than 10 contributors are cited – explored in an article how a new class of LLM-native applications were unlocking markets previously out of bounds for legacy SaaS. “Unlike their predecessors, these applications can target the high-cost repetitive language-based tasks that dominate numerous verticals and large sectors of the economy – such as legal, healthcare, and finance – that were largely out of bounds for legacy vertical software,” the Bessemer article notes. 

For Gradient Labs, a startup building a best-in-class AI support agent for financial services, the vision of the company involved these two concepts without people even realising. Dimitri Masin, co-founder and CEO of Gradient Labs, explains the rationale. 

“[When] we started, we were thinking ‘we would rather build something super, super high quality for one vertical, rather than spreading ourselves thinly,’” Masin explains. “There were a lot of horizontal players already back then, so it was key to actually stand out.” 

“For us, back then, it was already very clear. It’s only by focusing on one vertical that you can achieve high-quality results,” adds Masin. “So intuitively we went in that direction, and now [it’s] got a name.” 

It was the release of OpenAI’s GPT-4 in March 2023, which became the catalyst for the company. Masin, along with fellow co-founders Danai Antoniou and Neal Lathia, found after weekends spent hacking with new LLMs that delivering customer experiences in finance which were autonomous, yet safe, compliant, and exceptional, was now possible. While organisations had learned to scale customer operations through SOPs (standard operating procedures), the goal was to build an intelligence platform which followed SOPs like humans would – but at much greater quality.  

“When GPT-4 came out, it dawned on us that over the next five to 10 years, this area will be completely transformed based on the technology that came around,” explains Masin. “And we believe, ultimately, that within the next five to 10 years, our vision is to establish ourselves almost like an operating system for customer operations, and how customer operations scale in companies like banks or financial services.” 

Customer operations is naturally a large bucket. Dimitri estimates that a bank would have somewhere in the region of 900 different SOPs. Customer support was therefore chosen as the entry point into the market, being SOP-based. “So we essentially could pick an area which allowed us to build the core technology that we needed, that we had vision to develop, but equally where we can serve a well-defined, specific, very valuable use case – and that’s customer support,” says Masin. 

The spend potential is there for those who get it right. KPMG’s 2025 CEO Outlook survey finds that nearly three-quarters of chief executives plan to invest 20% of their entire budget on AI. Writing for Fortune, KPMG head of AI and data labs Swami Chandrasekaran, notes organisations are investing in the emergence of what he calls the ‘superhuman’ employee, and AI agents are the ‘catalyst’ for this. 

It is no coincidence that the word ‘superhuman’ also appears in the top line of Gradient Labs’ mission statement – and with good reason. The first live deployments began in January, yet the numbers make for interesting reading. Motor insurance provider Zego is a good example. The company launched an AI agent, named Alex, which can handle multiple, technically unlimited customer conversations at once but, crucially, with a human-like touch. The end customers have rated Alex as the best ‘human’ on the team.  

Customer satisfaction (CSAT) rates went up from 61% to 77%, while call volumes dropped by 25%. Plum, a financial app, achieved 80% CSAT on day one. Overall, Gradient Labs claims its AI agent delivers up to 98% CSAT in optimal implementations – even in complex, sensitive financial conversations.  

Ian Kershaw, Zego’s VP of Customer Service, Claims and Fraud, says what sets Gradient Labs apart is how they work – they ‘understand fintechs.’ That deep knowledge, as Masin explains, means the front end is only the beginning – and the first step on expanding from customer support to customer operations.  

“Customer support is just the tip of the iceberg,” says Masin. “If the customer gets in touch and says [their] payment has failed, more often than not, there will be some other underlying issues. A lot of work is happening behind the scenes in financial services. In order to enable an amazing or exceptional customer experience, it’s not enough just to care about the customer support. 

“Our procedure-based AI agent technology allows you to also automate back-office operations, as long as they’re procedure-based, and it allows you to do outbound outreach to customers,” adds Masin. “You can define a process of how to review a particular situation and what to do on the back of this, still based on the same technology, so it’s still procedures. But the interaction doesn’t happen in chat – it happens somewhere in the back end.” 

Masin is speaking at Fintech Connect on December 2-3 in London, and his session will outline the rise of agentic AI in customer experience. He will give an example of a large, regulated UK bank deploying agentic AI at scale and how Gradient Labs is powering that use case for them. “There’s a lot of hype around AIs. Many people are still sceptical, like ‘it’s just another chatbot’,” says Masin. The talk will be illustrated with statistics, including human versus AI CSAT, as well as resolution rates. “I think that will be interesting for people to see as well,” adds Masin, hoping to provide a ‘real-world flavour.’  

“It’s possible already now – a very high bar for quality. Large, regulated banks are already doing it.”