Harnessing AI to transform Confirmation of Advice letters

Posted June 19, 2026

Across the debt advice sector, there’s a long‑standing challenge with Confirmation of Advice (CoA) letters. 

They are a vital part of the advice journey; helping clients understand what was discussed and what they need to do next. However, feedback from across the sector, including through MaPS consultations, has been consistent: CoA letters can be too long, too complex, and too generic to truly support people in taking action. 

With support from the Money and Pensions Service (MaPS) Transformation Fund, we set out to explore how we could tackle this challenge using AI. 

A more personal, more engaging approach 

Our ambition was simple: to create CoA letters that genuinely reflect each individual National Debtline client’s conversation with their adviser. 

That meant moving away from standardised templates towards something more tailored, more human and more useful. We wanted letters that felt engaging and easy to follow, while clearly setting out prioritised, actionable next steps. 

At the same time, this project gave us an opportunity to explore how AI could be used responsibly within debt advice – enhancing outcomes for clients while supporting, not replacing, the role of skilled advisers. 

Building and refining the model 

Working with our technology partner, we developed an AI model trained on call transcripts, CRM data and ‘gold standard’ CoA examples, alongside regulatory and quality frameworks. 

The goal was to produce letters that: 

• Reflect the conversation in a clear and accurate way 

• Use plain English rather than jargon 

• Provide prioritised, actionable next steps 

• Maintain a reassuring, human tone 

We also put in place strong guardrails to ensure the accuracy and appropriateness of outputs. 

The model was refined through multiple feedback loops, with advisers playing a central role in shaping the outputs and ensuring they met the needs of clients. 

Testing in the real world 

We tested the tool through two adviser pilots, generating over 500 CoA letters across both phases. 

This gave us invaluable insight into how the model performed in real advice scenarios and, crucially, how it was received by advisers. 

The results were encouraging: 

• Over 45% of letters were ready to send to clients without changes 

• A further 47% required only minimal edits (less than 2–3 minutes) before being issued  

This means that in the vast majority of cases, the AI was able to produce outputs that were either immediately usable or very close to it. 

What we’ve learned 

Like any emerging technology, there have been important lessons along the way. 

We’ve seen that: 

• AI models require continuous training, testing and oversight 

• Outputs must always be reviewed to avoid inaccuracies or assumptions 

• A human element remains essential to delivering high‑quality advice 

But we’ve also seen the potential. 

This approach makes it possible to deliver a level of personalisation that would otherwise be difficult to achieve at scale, helping clients feel understood and more confident in taking the next steps. 

Looking ahead 

This work is just the beginning. 

We want to build on these early results by continuing to refine the model, testing its impact on client outcomes, and exploring how it could be applied more widely – including within Business Debtline. 

Above all, our focus remains on improving the experience and outcomes for people seeking debt advice. 

AI has a role to play in that future but it must be used thoughtfully, responsibly and alongside the human expertise that underpins our services. 

For more detail on our approach and findings, you can read the full report here.

---

---

---

---