)
Thinkmoney improves customer experience with
AI banking assistant
thinkmoney partnered with Data Reply UK to deploy a governed multi-agent AI banking assistant, with 63% of customers now resolving their query without reaching a contact-centre agent.
The CHALLENGE
Reduce pressure on contact centre teams and improve first-contact resolution for customer while meeting FCA Consumer Duty and UK GDPR requirements.
THE SCENARIO
From fragmented service journeys to governed customer-facing AI
Thinkmoney is a UK challenger digital bank focused on helping customers manage everyday money through app-based banking and customer support.
Routine customer journeys, such as card support, transaction queries and account support, can create friction when customers have to move between app screens or support channels to complete a task. thinkmoney wanted to build an AI banking assistant to improve customer experience, increase first-time resolution and help keep contact centre costs down across chat and agent-assisted communications.
But customer-facing AI in banking cannot be treated as a simple chatbot deployment. In a regulated banking environment, any move towards agentic customer service has to balance speed, trust, safety and operational control. Once an assistant can understand intent, route requests and support tool-enabled journeys, the architecture needs authentication boundaries, auditability, guardrails, escalation paths, testing and live monitoring.
The Solution
A smart banking assistant that resolves queries in minutes
Data Reply built a smart banking assistant for thinkmoney's mobile app that lets customers resolve routine service requests through conversation rather than navigating multiple screens or waiting for a human agent. Customers describe what they need in plain language and the assistant handles it — activating or freezing a card, ordering a replacement, explaining a declined transaction, or answering common account questions.
The assistant follows thinkmoney's "Capable with Consent" model. it can resolve queries autonomously, but only within pathways the customer has explicitly approved. When a request falls outside its scope or involves a vulnerable customer, it escalates to a human agent. Guardrails, PII masking, audit logging and human escalation are built into the operating model, not layered on afterwards — so the system met FCA and UK GDPR expectations from the point of deployment.
How we did it
Combining specialist agents, controlled release and continuous testing
The solution uses Amazon Bedrock, Claude Sonnet and Amazon AgentCore to coordinate specialist agents across thinkmoney’s service journeys, with a governance-first architecture designed for regulated banking. The delivery team built it around a custom agentic graph and controlled operational model, rather than relying on generic chatbot patterns.
Foundation triage
A triage layer interprets customer intent and routes requests to the right specialist journey, improving consistency and reducing manual effort.
Specialist service agents
Focused agents support card management, transaction queries, onboarding and complaints. This keeps each agent's scope bounded.
Data minimisation
Agents are given access only to the information required to complete the task, reducing exposure and supporting regulated operational control.
Auditable decisions
Reasoning, tool use and state information are visible so the team can review how the system arrived at an action or recommendation.
Dual-track testing
Dual-track testing generates realistic conversations from the agent graph, then runs them to check the assistant stays within bounds.
Staged production readiness
Automated pre-deployment testing, golden datasets, gradual rollout and live observability help build operational evidence before capabilities are scaled.
Queries solved 4x faster
Routine queries that previously took 10–15 minutes across fragmented screens and support channels can now be resolved in 2.2 minutes through a single conversational interface.
63% of queries resolved without
a human agent
The Q&A agent resolves 63% of customer queries autonomously, reducing pressure on contact-centre teams and freeing human agents to focus on complex, sensitive or high-value interactions.
98%+ intent-routing accuracy
The triage agent correctly identifies customer intent and routes to the right specialist agent in more than 98% of interactions, ensuring customer satisfaction and faster resolution times.
A reusable platform for regulated AI
The programme established a governed, reusable agentic platform — with guardrails, PII masking, audit logging, human escalation and lifecycle monitoring designed in from the start —giving thinkmoney a foundation for extending AI into additional service and operational journeys.
)
thinkmoney is a UK-based digital banking provider with 20+ years' experience helping customers manage everyday money through simple current-account services, budgeting tools, card management, payments and digital support.
Data Reply is the Reply Group company specialising in modern data platforms, advanced analytics and production-ready AI on Amazon Web Services (AWS). As an AWS Premier Consulting Partner since 2014, it designs, builds and operates data architectures and products spanning big-data engineering, streaming, migration, governance, business intelligence and self-service analytics. Its teams combine data science with machine learning, generative AI and agentic workflows, supported by DataOps and MLOps practices that automate deployment, monitoring and continuous improvement in production. This combination helps organisations target customers and grow revenue, optimise operations and supply chains, and strengthen security, risk management and compliance.