Building Chatbots that resolve problems, not just respond accurately

After reviewing 12 months of quantitative and qualitative feedback from real clients, if there was one thing apparent it's that clients are most frustrated when speaking to chatbots that simply don't understand their, seemingly, simple request.

A failure to understand a customer's intent is also often cited as the #1 reason why conversational experiences fail. So, when joining this project, I prioritised centring true client value when in developing the conversational standards and building the AI Chatbot.

Introducing the CRF (Conversation Response Framework)

As a result, I developed a framework, that could scale with future chatbot ambitions and capabilities, while simultaneously delivering consistent, controlled and compliant responses that earned and retained the trust of clients.

Prior to the framework, like many organisations the current experience had been developed on a "topic" based approach.

Topics don't tell you why someone's seeking support

The CRF is a framework designed to uplift the quality of an AI response in a controlled way that maximises the intuitiveness and engagement of a conversational response.

I developed 4 core intents that motivate the majority of client chatbot contact:

1. Fix a problem

2. Complete a task

3. Understand or learn

4. Get an update on a transaction or transfer

With these intents, I then developed conversation "goals" which help the AI Chatbot prioritise the details that matter most to the client. These ranged from understanding implications of actions or decisions, to simply comparing product options.

Delivering value beyond accuracy

The framework transforms the experience:

From: "the AI understands the question."
To: "the AI understands the support the client is seeking."

In collaboration with other design partners, I developed AI design principles that now shape every conversation decision being built into the system.

In financial services, a wrong or inconsistent answer risks becoming a compliance failure that compromises client trust. Designing response frameworks that hold up under regulatory scrutiny, without feeling like compliance to the client, is the harder and more valuable design problem.

What's changed?

And as a result, different combinations produce very different responses with different content, tone, structure and success criteria, allowing an ambitious organisation to offer intuitive, dynamic responses in a controlled way that soothes the anxieties of our stakeholders in compliance.

More concrete metrics

On the chatbot
The current projected reduction in unresolved conversations, with more consistent tone and depth across topic areas.

On the business

We're anticipating a 33% reduction in contact centre cost, modelled against industry benchmarks ahead of Q4 2026 launch.

On the organisation
CRF and the guardrail system are becoming the reference architecture for every new capability added to the chatbot.

Try the framework for yourself

Check out my illustrative interactive prototype