Key Points
- LHV Bank has launched an agentic artificial intelligence (AI) system for UK retail customer email support in partnership with Gradient Labs, enabling automated responses to routine banking enquiries around the clock.
- The launch follows a proof of concept announced in March 2026, moving the collaboration from an exploratory phase into live customer service.
- The AI agent can handle defined customer journeys from start to finish, including lost-card enquiries, payment explanations, bank transfer questions and updates on existing requests.
- Sensitive, complex or out-of-scope cases are escalated to LHV Bank’s human customer support team.
- The system combines large language models with deterministic guardrails to control data access, permitted actions and escalation procedures.
- LHV Bank says customer data is accessed only during live interactions and is not used to train the underlying AI models or build the agent.
- The bank is using Gradient Labs’ Collaborate platform to let operations, product, engineering, risk and compliance teams test, review and approve changes before deployment.
- Kris Brewster, Interim CEO of LHV Bank, said the objective was to provide customers with immediate answers while keeping employees in control of how the service operates.
- Dimitri Masin, CEO of Gradient Labs, said autonomous agents could take responsibility for defined tasks when supported by clear controls, escalation points and human judgement.
- The companies intend to expand the agent’s capabilities across additional email-support procedures and explore potential future applications in live chat, voice, back-office operations, financial crime workflows and LHV’s Estonian business.
What Has LHV Bank Announced About Its Agentic AI Customer Support System?
LHV Bank has introduced an agentic artificial intelligence system for customer email support in the UK, working with Gradient Labs to automate routine banking enquiries from beginning to end. Announced on 20 October 2026, the deployment is intended to give retail customers quicker answers while allowing the bank’s human support team to concentrate on complex, sensitive and exceptional cases.
As reported by Lauren Towner of FF News in LHV Bank Launches Agentic AI for Customer Support in Partnership with Gradient Labs, the new system moves beyond conventional AI assistants that draft suggested email replies for employees to review. Instead, the agent can complete approved customer-service workflows independently within defined operational limits.
The launch follows an earlier proof of concept between the two companies, announced in March 2026. That initial initiative examined how agentic AI could support retail banking communications while maintaining human oversight, customer protection and alignment with the bank’s regulatory obligations.
The development represents a practical application of Artificial Intelligence (AI) in UK financial services, where institutions must balance the potential benefits of automation against the risks associated with customer data, financial transactions and regulatory compliance.
According to the launch announcement, the service is designed to provide immediate responses to everyday enquiries around the clock. Its operation remains limited to approved procedures, with the bank retaining responsibility for the rules governing the agent and the circumstances in which employees must intervene.
How Does LHV Bank’s AI Agent Handle Customer Enquiries?
The system developed with Gradient Labs is designed to manage specific customer-service journeys rather than simply generate text for a human employee.
According to the announcement carried by FF News and the accompanying company statement published on 20 October, the agent can handle several routine banking enquiries.
These include:
- Lost-card enquiries: Providing assistance with questions about what customers should do when a card is lost.
- Payment explanations: Helping customers understand payment details.
- Bank transfer queries: Responding to questions concerning transfers.
- Existing requests: Providing updates on the progress or status of previously raised enquiries.
The agent can manage these defined workflows from initiation to completion, subject to its approved procedures and available tools.
The approach is intended to address a limitation of conventional customer-service queues. Straightforward questions can wait alongside complicated cases, even when they require substantially different levels of investigation and judgement.
By handling routine enquiries automatically, LHV Bank aims to provide quicker answers without requiring every interaction to pass through a human employee before a response can be issued.
However, the system is not presented as a replacement for human support across all customer interactions. Cases involving sensitive circumstances, complaints, vulnerable customers or requests outside the agent’s authorised scope can be escalated to the bank’s support team.
Jessica O’Connor, Deputy Editor of The Intermediary, reported on 20 October that the service would allow employees to focus on enquiries requiring human judgement rather than processing routine questions through traditional queues.
The operational distinction is important: automation is being applied to specific, approved tasks rather than unrestricted decision-making across the bank’s retail customer service.
Why Is LHV Bank Moving Beyond Traditional AI Co-Pilots?
The launch reflects a difference between AI co-pilots and agentic AI systems.
A conventional AI co-pilot can draft a reply, summarise a customer’s message or suggest a possible response. A human employee typically remains responsible for reviewing and approving the output before it is sent.
An agentic system, by contrast, can undertake a defined sequence of actions and complete an authorised workflow without requiring a person to approve every individual step.
Kris Brewster, Interim CEO of LHV Bank, explained the reasoning behind the decision in a statement quoted by Lauren Towner of FF News.
“As a digital-only bank, we don’t have to accept the way customer service has always been done,” Brewster said.
He argued that AI co-pilots could accelerate individual tasks without necessarily resolving the underlying problem of customer enquiries remaining in the same queue.
Brewster also highlighted the risk that employees might begin approving AI-generated suggestions without checking them closely when the technology is usually correct.
He continued that LHV Bank wanted to provide customers with an answer immediately while ensuring that its employees remained in control of the service and could intervene whenever a case required them.
“Getting that balance right is what responsible innovation looks like to us,” he said.
The comments establish the bank’s stated rationale for the deployment: the objective is not simply to make employees faster at handling emails, but to automate suitable tasks while preserving human oversight of the overall service.
The distinction also creates a different set of operational requirements. An AI agent must follow approved procedures, recognise when it has reached the limits of its authority and transfer cases to employees when necessary.
What Safeguards Protect Customers When the AI Agent Is Used?
Operating AI in banking involves risks that go beyond producing an inaccurate response. Systems may have access to sensitive information, interact with financial-service processes and encounter customers who require additional assistance.
LHV Bank and Gradient Labs have therefore designed the deployment around defined controls and escalation procedures.
According to the launch announcement, the system combines large language models with deterministic guardrails. These controls establish which information and tools the AI agent can access, which actions it is permitted to take and when a case must be blocked or referred to a human employee.
The safeguards also govern the handling of complaints and interactions involving vulnerable customers.
The model allows the AI to interpret customer enquiries and manage approved workflows while keeping its operational authority within boundaries established by the bank.
LHV Bank’s teams remain responsible for defining the procedures, setting their limits and approving changes before they reach customers. The bank also monitors the system’s performance and handles exceptions escalated by the agent.
According to LHV Bank’s published statement, customer data is accessed only during live interactions and is not used to train the underlying models or build the agent.
That is an important distinction in the bank’s description of its data practices. It does not mean that the system operates without accessing customer information; rather, the company states that such access takes place during live service interactions and that the data is not used for model training or agent construction.
The deployment’s effectiveness will depend on whether these controls work consistently in real customer interactions. In particular, the system must recognise when an enquiry falls outside its authorised procedures and when human intervention is necessary.
The announcement sets out the intended safeguards but does not provide independent performance figures demonstrating the system’s accuracy, escalation rate or customer satisfaction since launch.
How Does Gradient Labs’ Collaborate Platform Support AI Governance?
Alongside the agent itself, LHV Bank is using Collaborate, a feature of the Gradient Labs platform designed to apply software-engineering practices to the management of customer-facing AI.
As reported in the company statement reproduced by Cherry and covered by FF News, Collaborate enables LHV Bank’s operations, product, engineering, risk and compliance teams to prepare and test changes to the agent’s procedures in a controlled environment.
The process includes drafting changes, testing them against representative customer cases, reviewing the results and obtaining approval before deployment.
Changes can also be rolled back when necessary.
The platform maintains records of what changed, who approved the change and when it was made. This provides a documented history of modifications to the agent’s operating procedures.
The approach is intended to reduce the risk of untested changes reaching customers and to make the system’s development more manageable for teams responsible for operational performance and regulatory compliance.
Rather than treating the AI agent as a standalone model that is simply switched on and left to operate, the process places its procedures within a structured cycle of development, testing, approval and monitoring.
This is particularly relevant when AI is used in financial services, where changes to customer-facing processes may affect the handling of complaints, access to information and the resolution of service requests.
The bank’s operations, risk and compliance teams consequently have a role in governing how the technology behaves, alongside the engineering and product teams responsible for its implementation.
What Did Gradient Labs Say About the Partnership?
Dimitri Masin, CEO of Gradient Labs, described the deployment as an example of how AI agents could take responsibility for defined tasks while operating within clear boundaries.
In a statement quoted by The Intermediary and reproduced in the 20 October announcement, Masin said: “The future of AI in financial services is not a faster human approving AI suggestions at every step.”
He added that trusted agents should be able to take responsibility for defined work, supported by deterministic controls, clear escalation points and human judgement where necessary.
Masin concluded that LHV Bank was demonstrating what this approach could look like in practice.
Gradient Labs specialises in AI systems for financial-services operations. Its technology is designed to automate customer support and other defined workflows while addressing the control and accountability requirements associated with regulated activities.
In a September report for FinTech Futures, Senior Reporter Tyler Pathe also covered the rollout and discussed Gradient Labs’ broader approach to AI implementation with Masin.
Pathe reported Masin’s view that the value of AI in financial services should not be considered solely as a cost-reduction exercise. Masin said the experience of working with partners had led the company to place greater emphasis on regulatory compliance and customer experience.
That perspective provides context for the LHV Bank deployment. While automated handling of routine enquiries may reduce the amount of manual work required for each case, the companies have publicly emphasised service quality, governance and the ability to resolve defined tasks safely.
The launch announcement does not disclose specific savings, productivity improvements or customer-satisfaction results attributable to the new system.
How Did the LHV Bank and Gradient Labs Partnership Develop?
The collaboration began with an exploratory initiative announced on 29 March 2026.
In LHV Bank to Explore AI Usage With Gradient Labs to Support Retail Customers, Lauren Towner of FF News reported that the bank would conduct a proof of concept focused on agentic AI for email-based retail customer service.
The trial was intended to assess whether AI systems capable of planning, reasoning and taking controlled actions across defined workflows could improve the efficiency and consistency of customer support.
At that stage, the project was exploratory. The companies planned to examine response quality and turnaround times while ensuring appropriate human oversight and alignment with the bank’s governance, risk and regulatory requirements.
Brewster described customer support as a critical point of contact in retail banking and said the proof of concept would allow the bank to explore the technology while maintaining customer protection and human judgement.
Masin also welcomed the initial collaboration, highlighting the importance of speed, accuracy and empathy in individual customer interactions.
The October launch represents the next stage of that process, moving the technology into live UK retail email support.
In a separate report published on 28 September 2026, Tyler Pathe of FinTech Futures had already covered the deployment, reporting that Brewster confirmed the system was live and emphasised that the bank had prioritised safeguards over speed.
The subsequent announcement on 20 October provides further detail about the system’s permitted workflows, data-access arrangements, governance processes and potential future development.
What Is Gradient Labs’ Position in the Financial Services AI Market?
Gradient Labs is developing AI agents for financial-services operations beyond customer email support.
In his September reporting, Tyler Pathe of FinTech Futures noted that the company’s technology could also be applied to disputes, lending and customer onboarding. The report identified Wise, Zego, Plum, Stash, Yonder, Pockit, Penfold, Morse, SteadyPay and Bondora among Gradient Labs’ clients at the time of publication.
Pathe also reported that Gradient Labs was founded in 2023 by Dimitri Masin and fellow Monzo alumni Danai Antoniou and Neal Lathia.
The company announced in June 2026 that its Series A funding had increased to $26 million. According to Gradient Labs’ funding announcement, Octopus Ventures and CommerzVentures led the expanded round, with continued backing from Redpoint Ventures and Exceptional Capital.
These details provide background to the company’s development and its focus on automating operational work in financial services. They are separate from the LHV Bank deployment, and the funding announcement does not establish the financial terms of the partnership between the two companies.
For LHV Bank, the collaboration is part of a broader effort to apply technology to banking operations. Pathe reported that the bank also uses Project SARathon for regulatory compliance work involving Suspicious Activity Reports and Luminance for automated contract analysis.
The UK business operates as the regional subsidiary of Estonia’s LHV Group and provides retail banking, SME lending and banking services for fintech businesses.
The Gradient Labs deployment therefore sits within a wider technology-led operating model, although the announcement does not quantify its contribution to the bank’s financial performance.
What Other Customer-Service Channels Could LHV Bank Automate Next?
LHV Bank’s immediate priority is to extend the AI agent to additional customer-service procedures within UK email support.
According to the 20 October company announcement, the longer-term possibilities include exploring additional communication channels, such as live chat and voice.
The companies also intend to explore potential applications in back-office operations and financial crime workflows. Areas identified in the announcement include onboarding, know-your-customer (KYC) processes and anti-money-laundering (AML) activities.
Another possibility is extending the approach to LHV’s Estonian business.
These are future areas for exploration, rather than confirmed completed deployments. The announcement does not specify launch dates, implementation budgets or a timetable for bringing these additional capabilities into service.
Moving from email support into voice, onboarding or financial crime processes would also involve different operational requirements. Each use case would need to be assessed against its specific risks, data-access needs, escalation arrangements and applicable regulatory obligations.
For the immediate UK rollout, the focus remains on routine customer email enquiries and the gradual expansion of approved procedures.
What Could the Launch Mean for Customer Service in UK Banking?
LHV Bank’s deployment illustrates a shift in how financial institutions are testing AI in customer-facing operations.
Traditional automation has often focused on helping employees complete individual tasks, such as drafting responses or summarising customer messages. Agentic AI seeks to automate an entire defined workflow, provided the system has the authority and information needed to complete it.
For customers, the intended benefit is faster responses to straightforward questions, including enquiries about payments, transfers and lost cards. For the bank, the expected operational benefit is the ability to direct human attention towards cases requiring judgement, investigation or additional care.
However, the launch alone does not establish whether the approach will deliver better outcomes than existing customer-service systems across all measures.
Relevant indicators would include response times, resolution rates, accuracy, the frequency of unnecessary escalations, the handling of vulnerable customers and customer satisfaction. The bank has not published detailed results for these measures in the announcement.
The reliability of the controls will also matter. An agent that handles routine questions efficiently must still recognise unusual requests, avoid actions outside its authority and transfer cases when a customer needs human assistance.
The approach described by LHV Bank places governance and operational oversight at the centre of the deployment. Its longer-term significance will depend on how consistently the system performs in practice and whether the bank can extend its capabilities without weakening those safeguards.
For organisations examining the relationship between technology, operational governance and customer-facing processes, the development is relevant to Artificial Intelligence (AI) and Customer Service, particularly where automation must operate alongside human accountability.
What Are the Main Facts to Watch Following the Launch?
The next stage of the rollout will provide a clearer indication of how the system performs beyond its initial deployment.
Several developments will be relevant:
- Expansion of email procedures: Whether LHV Bank introduces additional approved customer-service workflows.
- Operational performance: Whether the bank publishes measurable evidence of faster responses or improved resolution rates.
- Human escalation: How effectively the agent identifies complex, sensitive or out-of-scope cases.
- Governance and testing: How the bank uses Collaborate to approve, monitor and, where necessary, reverse changes.
- Customer data protection: Whether the stated limits on data access and model training remain central as capabilities expand.
- New channels: Whether the companies progress from exploring live chat and voice to announcing confirmed deployments.
- Further applications: Whether onboarding, KYC, AML and other back-office processes become part of a defined implementation plan.
At present, the confirmed development is the launch of agentic AI for UK retail customer email support. The additional channels and operational applications remain areas the companies intend to explore.
The deployment provides a concrete example of a bank moving from an AI proof of concept to live automation of defined customer-service tasks. Its longer-term impact will depend on operational results, customer outcomes and the continued effectiveness of the safeguards governing the technology.