Shuweib Abdulrehman from Regulaitor discusses the evolution of AI in quality assurance, compliance monitoring and customer journey oversight in financial services.
Firms are adopting AI at different speeds and where traditional manual sample checking continues it is constraining oversight. The discussion explores how AI can move QA beyond transcription and basic sentiment analysis towards policy-aware, product-aware and regulation-aware assessment.
AI can help firms review more calls, identify vulnerability markers, assess affordability discussions, support Consumer Duty reporting and provide faster insight into adviser performance.
Find out more about Regulaitor -> Here.
Key Take Aways
- AI adoption in quality assurance remained uneven, with some firms adopting new tools while others retained manual checking.
- Telephony platforms increasingly embedded basic AI capabilities, including script checks and sentiment scoring.
- Consumer Duty increased the need for more rigorous end-to-end customer journey assessment.
- Manual sample checking continued to limit firms’ ability to identify systemic issues at scale.
- Generic transcription and sentiment tools were described as too limited for complex financial services use cases.
- Bespoke AI models were positioned as necessary where firms needed policy, product and regulatory understanding.
- AI-enabled QA could shift firms from reviewing a sample of calls to reviewing 100% of calls.
- The discussion positioned AI as a productivity enhancer rather than simply a headcount reduction tool.
- Smaller language models were presented as useful for targeted tasks, including vulnerability detection.
- AI governance, benchmarking and model control were highlighted as critical for regulated environments.
- Back-office AI use cases were seen as more viable than customer-facing advice applications.
- AI quality monitoring could strengthen Consumer Duty reporting by identifying trends, failures and training needs more quickly.
Innovation
- Use of AI overlays on call transcription to assess policy, regulation, advice quality and product suitability.
- Deployment of bespoke knowledge bases and retrieval-based systems to reflect company procedures.
- Use of smaller, task-specific language models for vulnerability detection.
- Decision-tree logic combined with AI interpretation to assess customer journeys and suitability.
- AI-generated call notes for agents, with human approval before submission.
- AI-enabled full population monitoring, moving beyond traditional sample-based QA.
- Feedback-loop mechanisms allowing compliance or QA teams to challenge outputs and improve the system.
- Layered assurance models where AI performs first-line QA, QA teams oversee the AI, and compliance oversees the overall control environment.
- Integration of call recordings, customer data, affordability assessments, open banking data and credit information to build richer case-level assessments.
- Potential use of agentic AI to investigate suspicious customer behaviour across multiple systems.
- Application of AI-derived thematic reviews to adviser training, Consumer Duty reporting and management oversight.
- Emergence of hybrid “forward deployed” roles combining technical knowledge with financial services process expertise.
Key Statistics
- The AI solution discussed was described as 20 times quicker than listening to a call.
- It was described as 10 times cheaper than the manual alternative.
- One client compared QA labour costs of £17 to £25 per hour with AI unit costs of £2 to £3 per call-recording hour.
- One firm moved from reviewing 10% of calls to reviewing 100% of calls because AI made full coverage affordable.
- Customer acquisition spend in the debt sector was described as reaching £40,000 to £50,000 per month for some firms.
- The speaker said Regulaitor had interviewed over 50 people across performance, coaching, learning and development, compliance, QA and team leadership roles.
- A future adoption horizon of around 18 months was discussed for broader catch-up by firms currently behind early adopters.
- Legacy systems were described as having been embedded for 20 to 30 years in some organisations.
- A customer outcome example identified five failures between question 10 and question 17 in one adviser review.
- The discussion referenced firms looking at KPI horizons of around one to two years, rather than ten-year transformation decisions.
- The transcript referred to the debt sector using markers such as income, debt level and asset value in affordability and suitability decision trees.
- The discussion referenced 100% call review as a practical capability when unit economics changed.
Key Discussion Points
- The market split between firms adopting AI-enabled QA and firms continuing with manual review.
- The limits of basic telephony-led AI tools such as script checking and sentiment scoring.
- The difficulty of using generic sentiment tools across regional accents and speech patterns.
- The need for AI systems to understand policy, advice, product rules and regulatory nuance.
- The role of Consumer Duty in increasing accountability for end-to-end customer outcomes.
- The importance of bespoke AI and internal knowledge bases rather than generic out-of-the-box models.
- The use of smaller language models for specific compliance and vulnerability tasks.
- The business case for moving from sample-based QA to full-call population monitoring.
- The value of AI in identifying thematic failures, training priorities and adviser-level trends.
- The need for governance, benchmarking and clarity over what firms want AI to achieve.
- The likely growth of agentic AI and AI assistants across financial services workflows.
- The cultural challenge of helping teams use AI to do better work, rather than simply automate existing processes.
#Regulaitor
RO-AR.com contact list
