How Can AI Support Complaints Root Cause Analysis Across Delegated Partners?

Complaint counts can show that customers are dissatisfied, but they do not explain what needs to change.

Across delegated arrangements, complaints may be recorded by coverholders, MGAs, TPAs or delegated claims administrators using different categories and narrative styles. Similar problems can appear under different labels, while one category may contain several underlying causes.

The FCA's current good-practice work emphasises granular management information, clear governance, action and measurement of impact. AI can help teams analyse more narrative evidence consistently, but root-cause conclusions and remediation require investigation by accountable people.

Complaint counts can hide systemic patterns

Volume is useful context, but it can mislead when used alone. A partner with more customers will usually record more complaints. A partner with stronger recognition and recording may appear worse than one that under-records. Product, distribution channel and customer cohort also affect the comparison.

Complaint categories often describe symptoms such as delay, communication or service. The underlying cause might be a system rule, unclear wording, inadequate staffing, a hand-off failure or an incentive within the customer journey.

Root-cause analysis needs to connect the complaint with wider evidence. That may include upheld and non-upheld outcomes, Financial Ombudsman referrals, quality assurance, claims handling, journey times, vulnerability information and process changes.

Established root-cause analysis starts with governance

A governed taxonomy helps teams record products, journeys, issues, outcomes and potential harm consistently. Retain the customer's original narrative because a standard code cannot capture every detail.

Traditional analysis combines management information, file sampling, interviews and process mapping. The person who owns the relevant process should help investigate the cause and agree the response. Conduct or complaints specialists provide challenge, while senior forums oversee material themes and actions.

These methods remain necessary. Their limitation is scale: people may read only a sample of narratives, and partners may classify similar issues differently. Manual analysis can also focus on the most visible recent issue rather than patterns developing across the portfolio.

AI can broaden thematic analysis

AI can classify complaint narratives against an approved taxonomy, cluster similar descriptions and highlight emerging language that does not fit existing categories. It can link themes with claims outcomes, customer journey times or quality findings where permitted.

The system should show representative source records and explain which features contributed to a theme. A cluster is a hypothesis for investigation, not proof of cause. Reviewers need to test whether the pattern reflects actual customer experience, recording practice or a change in portfolio mix.

Use AI to prioritise evidence rather than decide individual complaints, uphold outcomes or calculate redress. Those decisions involve case-specific facts, policy terms, regulation and judgement.

Fair comparison and action remain human-led

Normalise rates for relevant volumes and compare similar products, channels and cohorts. Show data completeness and taxonomy differences. A partner should not be penalised because it records complaints more accurately or serves customers with different needs.

Complaint data can include health, vulnerability and other sensitive personal information. Apply purpose limitation, data minimisation, access controls and appropriate retention. Test classifications for uneven performance across language, product and customer groups.

When an investigation confirms a cause, assign an owner, action, target date and outcome measure. Follow-up should ask whether the intervention reduced harm or improved the journey, not simply whether the action was marked complete. Record governance challenge and retain the evidence behind the decision.

Example

A hypothetical insurer receives complaints from several delegated claims partners. Categories differ, but AI identifies a recurring narrative theme concerning long periods without customer updates.

The conduct team samples the source records and compares matched claim types and journey stages. One partner's volume is explained by a more complex claims mix. Across two others, the investigation identifies an unclear hand-off between loss adjusters and customer-service teams.

The process owners revise the hand-off control and communication prompts. The oversight forum tracks customer-update intervals, repeat complaints and quality-assurance findings over subsequent periods. AI has helped find the shared pattern; people have confirmed the cause and owned the response.

FAQs

• Can AI determine the root cause of a complaint?

  AI can identify themes, relationships and candidate causes. Process owners and conduct specialists must investigate the evidence before confirming a root cause or deciding remediation.

• Should non-upheld complaints be analysed?

  They may still reveal friction, misunderstanding or communication weaknesses even where the individual outcome was fair. Analysis should retain the complaint outcome and wider context.

• How can delegated partners be compared fairly?

  Adjust for customer and transaction volumes, product mix, channels, cohorts and recording quality. Combine rates with qualitative review and make data limitations visible.