The bank is transforming how its people access the right answers at the right time by deploying two distinct AWS‑native AI solutions: a Contact Centre Knowledge Chatbot to support frontline agents, and a Policy Chatbot that provides fast, consistent answers to internal policy and procedure questions. The latter is particularly valuable during periods of change, such as merger-related activities.
Together, these solutions replace fragmented search tools and inconsistent guidance with a governed, conversational knowledge experience. By combining responsible AI with human oversight, the bank improves accuracy, reduces handling time, and strengthens quality and compliance.
From scattered information to two purpose‑built assistants.
Like most customer‑focused banks, this organisation has long invested in knowledge management to help contact centre agents support customers effectively. Over time, however, product expansion and ongoing policy changes meant critical information was spread across multiple systems and formats.
The result was familiar friction: keyword searches that didn't always surface the right guidance quickly, inconsistent phrasing of answers, and longer onboarding times for new staff who had to learn both the content and where to find it.
Rather than treating "knowledge" as a single problem with a single solution, the bank took a more practical approach: two assistants, each designed around a specific user context and job‑to‑be‑done.
1) Contact Centre Knowledge Chatbot: faster, more accurate customer conversations
The Contact Centre Knowledge Chatbot helps agents respond to customer questions quickly and accurately, without removing the human from the interaction. It fits naturally into agents' workflows, responds in plain language, and grounds every answer in approved source content.
The goal is not to automate customer service, but to make human service more consistent, more confident, and more efficient.
Importantly, the solution goes beyond chat. It also includes AI‑powered smart search, allowing agents to retrieve the most relevant guidance rapidly. This reduces time spent searching through documents and lowers the risk of missing important policy nuances during live customer interactions.
2) Policy Chatbot: internal policy clarity during change
In parallel, the bank deployed a Policy Chatbot for internal teams, designed to answer questions about policies, procedures, and related internal guidance with strong governance and traceability.
This capability becomes especially valuable during organisational change, such as merger activities, when employees need fast, reliable answers and policy owners need confidence that staff are working from the latest approved position.
Both solutions share a common philosophy: answers must be explainable, controlled, and grounded. This ensures staff can trust the guidance they receive, while leaders retain confidence in quality, compliance, and oversight.
AWS-native foundations, tailored experiences
Both assistants are built on AWS using a retrieval‑augmented generation (RAG) approach, turning approved knowledge sources into timely, traceable answers.
In production, Amazon Bedrock hosts the foundation model (Claude 3.x Sonnet), which synthesises responses after retrieving relevant source material. Content is prepared using embeddings and re‑ranking so the model reviews the most relevant information before responding. A self‑hosted vector database on Amazon EC2 supports retrieval, while Amazon S3 provides a secure staging layer for content ingestion.
Amazon DynamoDB captures chat logs to support auditability and continuous improvement, and AWS Lambda, orchestrated with Amazon EventBridge, manages scheduled content updates so both assistants remain current without manual intervention.
Where the solutions differ is in the last mile: each assistant is tailored to its audience, workflow, and governance requirements.
Contact Centre Knowledge Chatbot: agent assist + smart search
Agents authenticate via single sign‑on, access a simple interface, and ask questions in natural language. Responses are delivered with clear answers and linked citations back to source articles, allowing agents to validate information quickly and follow through to the underlying guidance when needed.
Alongside conversational responses, the AI‑powered smart search capability surfaces the most relevant content at speed, supporting accuracy and confidence under real contact‑centre time pressure.
Policy Chatbot: policy navigation with strong controls
The Policy Chatbot applies the same AWS‑native architecture and guardrails to internal policy and procedural content. It helps employees quickly navigate questions such as "How do we do this?" or "What's the current policy position?", which is particularly important when policies, processes, and responsibilities are evolving during merger‑related work.
Identity integration ensures access remains controlled, versioning and update processes keep answers aligned to the latest approved position, and monitoring and reporting provide ongoing operational confidence.
Across both solutions, the bank's AWS environment establishes a secure boundary, including data residency in the Sydney Region, encryption at rest and in transit, and role‑based access control via Identity and Access Management (IAM). This ensures the capability aligns with existing operational and compliance standards.
Measuring accuracy as a capability, not a configuration
Accuracy is treated as a capability, not a one‑off configuration. Both solutions are measured and improved continuously.
Retrieval quality is tracked using metrics such as Recall@K, Mean Reciprocal Rank, and Mean Average Precision. The chat experience collects lightweight user feedback through thumbs up/down ratings and optional comments, while service telemetry monitors availability, latency, and usage patterns.
This closed feedback loop informs ongoing tuning, including refinements to chunking strategies, re‑ranking thresholds, and prompt design. Over time, real‑world usage helps the assistants become increasingly precise and reliable.
Sharper conversations today, clearer policy tomorrow
The first change people notice is momentum. Contact centre agents spend less time searching and more time helping customers, supported by fast, reliable answers and smart search that surfaces the right guidance when it matters. New hires ramp up faster as the assistant reduces the cognitive load of knowing where to look, while experienced staff reclaim minutes throughout the day that add up at scale.
The second change is confidence. With both solutions operating within the bank's AWS environment, security and compliance requirements are reinforced, and detailed logging supports auditability. Keeping humans in the loop—reviewing sources, flagging improvements, and providing feedback—ensures every interaction contributes to improving accuracy.
The third change is resilience during change. The Policy Chatbot gives employees a dependable way to navigate internal policies and procedures, particularly during merger activities, when clarity, speed, and consistency are critical. Policy owners gain a clear path to update guidance and see those changes reflected quickly, narrowing the gap between intent and execution.
In practice, this is not a story about deploying "a chatbot". It is about building two durable, governed AI knowledge capabilities: one that strengthens frontline customer conversations, and one that improves internal policy navigation when the organisation needs it most. AWS provides the foundation. The bank's people provide the judgement. Together, they turn institutional knowledge into a strategic asset—one answer at a time.
