Complaint root-cause analysis should connect a customer's account to the product version, transaction, communication and operational event they experienced; distinguish symptom, contributing factor and controllable cause; search for similarly affected non-complainants; assign a product or process change; and verify through later outcomes that the intervention worked.
- Key takeawaysDo not code the customer's wording as the root cause.
- Decision frameworkComplaints become useful product evidence when a fintech connects the customer's experience to the process, product version and decision that caused it.
- Separate the symptom from the causeRecord what the customer experienced in their own terms, then record the investigated cause separately.
- Connect the case to a real eventLink the complaint to transaction identifiers, policy versions, communications and relevant system incidents.
- Look for affected customers who never complainedA confirmed systemic issue should prompt a search for the broader population.
Answer first: Complaint root-cause analysis should connect a customer's account to the product version, transaction, communication and operational event they experienced; distinguish symptom, contributing factor and controllable cause; search for similarly affected non-complainants; assign a product or process change; and verify through later outcomes that the intervention worked.
Key takeaways
- Do not code the customer's wording as the root cause.
- Preserve event, version and decision evidence for each case.
- Look for affected customers who never complained.
- Measure the fix against recurrence and cohort outcomes, not deployment completion.
Decision framework
| Decision area | Evidence to require | Risk it controls |
|---|---|---|
| Symptom | What the customer reported or experienced | Useful intake, not cause |
| Event | Transaction, message, queue or product version | Reproducible evidence |
| Cause | Controllable design/process/control condition | Action target |
| Extent | Other exposed customers and cohorts | Potential systemic harm |
| Verification | Post-change recurrence and outcome test | Evidence of effectiveness |
Complaints become useful product evidence when a fintech connects the customer's experience to the process, product version and decision that caused it. Counting cases and meeting response targets are necessary operational tasks. They do not establish that the underlying problem has been fixed.
A customer saying a payment disappeared may be describing a delayed settlement, an unclear status, a ledger error or an unanswered support request. One broad payment complaint label can hide four different interventions.
Separate the symptom from the cause
Record what the customer experienced in their own terms, then record the investigated cause separately. Preserve uncertainty. Not every case will support a confident technical explanation.
The FCA's review found examples where reporting supported complaint-handling operations without sufficiently identifying themes, causes or preventive actions. 1 That is the difference between managing the queue and improving the service.
Use a taxonomy with enough detail to guide action, but avoid hundreds of labels that staff cannot apply consistently. Test a sample with two reviewers and compare classifications.
Disagreement can reveal ambiguous categories or missing evidence.
Connect the case to a real event
Link the complaint to transaction identifiers, policy versions, communications and relevant system incidents. Record the customer journey stage and the time between the event and the complaint.
For a multi-firm product, identify which party controls the suspected cause. A platform may own the wording while the regulated provider owns the decision. An unresolved handover should remain visible rather than disappearing into an other category.
Protect access to sensitive case information. Product teams often need a pattern and a representative example, not unrestricted access to every customer's personal circumstances.
Look for affected customers who never complained
A confirmed systemic issue should prompt a search for the broader population. Determine which records can identify similar events, what the search misses and whether remediation is needed.
For example, a hypothetical release shows an incorrect fee explanation to 4,000 customers, but only 30 complain. Fixing those 30 cases does not establish that the remaining 3,970 understood the charge. The next step is to identify exposure and assess the consequences.
Do not assume silence means satisfaction. Customers differ in confidence, time, language and willingness to pursue a complaint. A low complaint count can coexist with a confusing or difficult service.
Learn from external decisions carefully
The Financial Ombudsman Service publishes anonymised final decisions. 2 They can help teams identify recurring reasoning and evidence questions, but an individual outcome depends on its facts. A startup should not turn one decision into a universal rule.
Use external cases to challenge internal assumptions. What evidence would establish what the customer saw? Can the company explain its decision contemporaneously? Is a support promise consistent with the underlying process?
Document the relevance and limits of each example. The purpose is to improve investigation and design, not to search for an isolated case that supports a predetermined view.
Verify the fix
Give each recurring cause an owner, proposed change, implementation date and outcome measure. The change may concern wording, product logic, staff training or a provider arrangement.
Compare complaint rates against the affected activity, with appropriate delay for reporting. If transaction volume doubles, an unchanged complaint count may mean something different from a stable-volume period. Show absolute numbers as well as rates.
Review a sample of post-change journeys. A falling complaint rate can reflect a harder contact route rather than better outcomes. Confirm that customers can still raise problems and that the underlying event is genuinely less frequent.
Link complaints to the version customers encountered
Capture the interface, wording, policy and relevant release at the time of the event. A product team inspecting today's screen can otherwise conclude that a reported misunderstanding is impossible.
Suppose a hypothetical cancellation flow changed twice during a quarter. A monthly total may conceal that complaints cluster around the first version and arrive after the second has already launched. The version record helps determine whether the latest change solved the cause or merely changed its appearance.
Keep the investigation proportionate, but preserve enough history to avoid repeatedly reopening the same question.
Cost remediation without letting cost decide the finding
Estimate affected cases, correction effort, refunds where appropriate and communication.
Show uncertainty and the work needed to refine it.
The potential bill should inform planning, not distort the factual conclusion. A large population is a reason to improve the evidence and governance, rather than narrow the definition until the problem looks smaller.
After remediation, test a sample of corrected records and customer communications. A bulk processing job can complete while some accounts remain wrong. Preserve exceptions and assign ownership until they are resolved.
This closes the loop between listening and action. Complaint analysis becomes credible when management can show not only what it learned, but which customers were considered, what was changed and what evidence supports the claim that the issue is now better controlled.
What should management receive?
Provide the leading causes, affected populations, overdue actions and evidence that completed changes worked. Include examples where an initial explanation proved wrong.
Choose one recurring complaint and follow it through to the verified fix. Record the affected population, change and outcome. That is the evidence that customer feedback has reached the product rather than merely moved through the complaints queue.
A practical 90-day implementation plan
- Link complaint records to transaction IDs, communication versions, decision logs and relevant operational events.
- Apply a simple Five Whys or causal tree and separate evidence from inference.
- Query the exposed population, including similar customers who did not complain, and assess redress or remediation.
- Define a success measure and review date before closing the product or process action.
- Present one completed case to the product forum with its evidence trail, affected population, decision, remediation and post-change outcome measure intact.
Frequently asked questions
Is a complaint category the same as a root cause?
Usually not. 'Delay' or 'poor communication' describes a symptom; the cause may be a queue rule, dependency, staffing assumption or product-state error.
What if a fintech has very few complaints?
Use proportionate case review and supplement it with call listening, support contacts, journey drop-off, staff insight, Ombudsman decisions and FCA material.
When is a complaint-led fix complete?
When the responsible forum has evidence that the change reduced the identified harm across the exposed population, not simply when code or training was delivered.
Reporting by Shoreditch Talk




