Customer service in banking is becoming a field where artificial intelligence is taking over more and more of the daily work. The British company Gradient Labs is positioning itself at the forefront of this development with purpose-built AI agents for the financial sector – and the results they present are strikingly high.
Built for Banking – Not Just Adapted
Where many AI solutions are general tools adapted to various industries, Gradient Labs emphasizes that their platform is developed from the ground up for financial services. This includes support for complex workflows, API integrations with banking systems, and built-in regulatory compliance adapted to requirements such as the UK financial regulator FCA's Consumer Duty standard.
According to the OpenAI blog, the company uses GPT-4.1 as well as the lighter variants mini and nano to maintain low response times and high operational stability – qualities that are critical in a customer context.
"Agentic banking" is no longer a thing of the future – Gradient Labs claims to be running it in production for real bank clients today.
Impressive Figures – But From the Company Itself
Gradient Labs states that its AI agents resolve between 40 and 60 percent of incoming cases out of the box, and that the proportion rises to over 80 percent after customization. Insurtech company Zego is said to have achieved a customer satisfaction score (CSAT) of 77 percent with the AI agent, compared to 61 percent for human agents. The company's flagship agent, launched in 2024, is reported to have reached CSAT scores of up to 98 percent.
It is important to note that these figures originate from Gradient Labs itself and from the company's own case studies – they are not independently verified.
Nine Million Security Checks
In an industry where errors can have legal and financial consequences, Gradient Labs places great emphasis on its security architecture. The platform is designed to screen both AI-generated and human responses in real-time, with controls for prompt injection, vulnerability detection, prevention of unauthorized financial advice, and handling of sensitive customer information, among other things.
According to the company's own figures, a single bank client performed almost nine million such security checks during 2025 – a number that illustrates the scale of the activity, but which cannot be externally confirmed.
Privacy and Data Usage: Clear Promises
In the financial sector, data privacy is a key obstacle to AI adoption. Gradient Labs seeks to address this by promising that neither the company nor its subcontractors store or use customer data for purposes other than the service delivery itself – including an explicit prohibition against using data to train AI models.
The company states that it has SOC 2 certification and GDPR compliance, and claims to use zero data storage where possible, including with third-party providers such as OpenAI and Anthropic.
In an ongoing proof-of-concept with LHV Bank, the solution is being tested with particular emphasis on explainability, auditability, and clear division of responsibility. Kris Brewster, interim CEO of LHV Bank, states that the collaboration will make it possible to improve services without foregoing human judgment and customer protection, according to Gradient Labs.
From Support to "Agentic Banking"
Dimitri Masin, CEO of Gradient Labs, describes the development as a shift towards what he calls "agentic banking" – where AI agents take over complex processes traditionally handled by humans, including fraud investigation and money laundering cases.
It is an ambitious vision, and the industry will closely monitor whether the figures hold up in independent tests. For now, Gradient Labs itself is setting the agenda – and it is clear: artificial intelligence as a customer service agent is no longer an experiment, it's a product.