Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Insurance Sales Complaint Root Cause and Redress Measurement

Abstract. Insurance sales complaints arise when the process of acquiring, replacing, renewing, or changing insurance creates an alleged deficiency in consent, advice, disclosure, product compatibility, or the management of commercial incentives. Their measurement requires a distinction between the underlying distribution failure, its recognition by the customer, its expression as a complaint, its capture by the institution, adjudication, and the delivery of an effective remedy. This monograph develops a comparative analytical synthesis of these distinctions in insurance distribution and consumer protection. It examines international conduct standards, European Union legislation and supervisory evidence, United Kingdom complaint rules and payment protection insurance redress, Australian internal dispute resolution and remediation guidance, United States complaint information, and the institutional distinction between conduct investigation and compensation. It specifies reproducible methods for complaint coding, exposure adjustment, reporting-bias sensitivity, counterfactual loss assessment, remediation verification, and causal evaluation. Worked examples are explicitly hypothetical. The central finding is that complaint volume, established misconduct, affected population, and restored customer outcomes have distinct denominators and evidentiary requirements. A defensible assessment combines individual investigation with independent surveillance of sales practices, population tracing, and measurement of the implemented remedies and subsequent customer outcomes. The study is a comparative doctrinal and methodological synthesis rather than an original estimation of market-wide incidence; its conclusions specify the conditions under which quantitative evaluation can later be conducted.

Keywords: insurance distribution; sales complaints; market conduct; mis-selling; root cause analysis; complaint selection; exposure normalization; consumer redress; remediation; causal inference; insurance intermediaries

Introduction

The sale of insurance establishes a relationship between a promise of future risk transfer and the circumstances of the person purchasing that promise. A sales complaint provides evidence of a perceived disruption in that relationship. A customer may discover, at the moment of purchase or years later, that the insured event is different from the event they expected to cover, that a premium was charged without meaningful consent, that a replacement policy introduced an exclusion, or that the advice given did not correspond to the financial circumstances known at the time. The complaint is then both an individual request for a response and a potential observation of a failure in the distribution process.

Insurance intensifies the problem of delayed discovery. At purchase, a policyholder cannot fully observe the quality of performance under all future states of the world. The customer may receive a clear explanation of the premium while misunderstandings about exclusions, eligibility, waiting periods, surrender costs, or the rights available after a change of employment remain latent. Recognition is often triggered by a claim, cancellation request, renewal notice, or a review of financial records. The event that reveals the injury and the event that created it must therefore be recorded separately.

The research has three substantive questions. First, how can a complaint taxonomy isolate the distribution-stage contribution to harm while retaining links to claims and servicing? Second, how can rates be made interpretable under changing product, channel, cohort, and complaint-capture conditions? Third, how can institutions evaluate whether corrective measures restore customers' positions and reduce future harmful sales? These questions join legal analysis, measurement theory, operational governance, and causal inference. They also connect the scientific disciplines of insurance distribution and insurance consumer protection.

The method is a structured comparative reading of selected primary instruments, institutional supervisory studies, data documentation, and relevant published methods. The research cutoff is November 19, 2026. Legal examples are tied to the indicated instrument or version, including original EU legislation and dated United Kingdom provisions; the research does not certify every national implementation, all subsequent amendments, or jurisdiction-specific remedies. The empirical evidence is used within its documented sample scope. No confidential firm-level complaint panel has been supplied, so the numerical illustrations demonstrate calculation and identification problems rather than estimate existing market outcomes. A complete population study would require transactional microdata, consistent definitions, follow-up and independent access to records representing customers who did not complain.

Distribution Complaints as Supervisory Evidence

The complaint as an observation of a sales process

A complaint record should be interpreted as an observation generated by an institutional process. It records a customer expression or allegation, associated events, and the response of an organization within a defined capture system. It may contain credible information about a sales failure before the allegation has been verified. It may equally record a disagreement with a correctly explained contractual limitation. Both observations are valuable for understanding customer experience, while they have different significance for a conclusion of misconduct.

Singh's (1988) study of complaint behavior distinguishes the empirical problem of classifying customer responses to dissatisfaction from the simple presence of a formal complaint. The accessible abstract supports the proposition that these responses have multiple dimensions. In insurance the practical consequence is that complaint databases observe a selected manifestation of dissatisfaction, shaped by the customer and the organization. A person may leave a distributor, cancel coverage, warn acquaintances, or do nothing. The observed complaint requires a mechanism that converts recognition into recorded data.

Complaint evidence is strongest when paired with the contemporaneous record of the sale. A complaint about unexpected unemployment exclusions may be associated with a sales script, recording, product version, customer occupation, and the original eligibility decision. The investigation should identify whether the failure was an unsuitable recommendation, misleading description, invalid consent, a different defect, or no established distribution violation. The same contractual exclusion may produce many complaints for different reasons. Conversely, a defective eligibility rule may produce injuries that appear in claims, policy cancellation, and accounting adjustments as well as complaints.

Scope of conduct supervision

The European Union's Insurance Distribution Directive (European Parliament and Council, 2016) establishes a basis for relating sales processes to consumer protection. Articles 14 and 15 concern complaint and out-of-court redress arrangements; Articles 17 and 20 establish substantive conduct and demands-and-needs requirements. Article 25 establishes product oversight and governance requirements in its defined scope. The obligations are implemented through the applicable national framework. The source is the original directive; an actual assessment of a transaction requires the operative national rules and date.

The IAIS (2024) Insurance Core Principles place conduct responsibilities along the distribution chain, with the insurer retaining its responsibility as the risk carrier and responsibilities in outsourced functions. A complaint can therefore reveal a problem extending across the entities performing design, marketing, sale, or service functions. The supervisory standards orient national implementation; they do not independently determine private liability or remedies in all jurisdictions.

The significance of an individual complaint depends on both the severity of the possible injury and the transferability of its cause. A serious allegation tied to one script used across a national sales network calls for a cohort investigation even before many complaints are recorded. A high volume of low-severity, separately explained dissatisfaction may call for a different intervention. Prioritization by frequency alone leaves open the possibility that rare but severe defects escape review.

Sales versus claim-stage defects

Consider two customers whose claims are denied because a policy excludes a pre-existing condition. The first customer received a sufficiently clear explanation and purchased the policy for other risks. The second was explicitly told that the excluded condition was covered. Their complaints share a claim outcome, but the second presents a distribution-stage hypothesis that requires examination of the representation at inception. A third customer may have received inaccurate claims advice after inception without any defect in the sale. These cases show why a code should preserve the discovery stage while separately identifying the hypothesized causal stage.

The 2026 EIOPA report provides a real example of the boundary. Its country-specific supervisory findings on page 23 describe German complaint growth associated largely with delays in claims processing and expressly leave open whether deeper sales or advice failures explain those complaints. This is evidence that a sales-specific interpretation requires further investigation. Neither a denial nor a delay automatically establishes a flawed distribution process.

Evidence triangulation

The evidentiary sequence begins with the complaint narrative, then integrates sales records and independently sampled transactions. A regulator or conduct function should compare the rate of reported problems with the frequency of the same observable defect in a sample of sales. Differing results can suggest a recognition or reporting gap, or that the audit test is not measuring the same underlying event. Investigators should determine which of these explanations remains plausible before constructing a prevalence estimate.

For a systemic issue, the investigation should document a common mechanism, its period of operation, the products and channels exposed, the characteristics of affected customers, and the conditions under which the mechanism caused detriment. The complaint initiates the inquiry; transactional and control evidence establish its reach.

Complaint Taxonomy and Root Cause Attribution

Separating allegation, defect, and mechanism

A useful coding system distinguishes the customer's allegation, the observable failure, the mechanism that produced it, the responsible process, and the injury. These fields may have different evidence states. A customer alleging an unwanted policy, for example, may have an unverified consent allegation, an established missing opt-in record, a hypothesized preselected default, and a premium charge that is independently confirmed. Treating these as one definitive misconduct label obscures the uncertainty and the action required.

The following taxonomy is a research coding aid for existing conduct concepts rather than a new legal classification. Each row identifies a possible distribution cause and a record that could support or weaken it.

Sales issue Candidate defect Evidence
Consent and authorization Unrequested insurance, coerced or substituted assent Opt-in events, recordings, signatures, billing chronology
Advice and needs Recommendation inconsistent with materially known needs Needs assessment, recommendation rationale, product version
Disclosure and representation Misleading description of exclusions, cost, duration, or benefits Script, marketing copy, policy summary, timing of disclosure
Eligibility and targeting Sale outside eligibility or designated customer segment Underwriting rule, eligibility data, target-market assessment
Remuneration and steering Incentive distorts the offered option Commission schedules, target thresholds, alternatives, matched files
Replacement and switching Misrepresentation of retained or lost benefits Old/new policies, surrender costs, replacement analysis
Bundling and access Optional product presented as required or insufficiently separable Checkout sequence, lender or retailer instructions, opt-out logs

The need to keep a direct sales code distinct from the form of the complaint is especially acute with automated distribution. A customer may describe an unauthorized charge without knowing the interface default responsible for it. A manual reviewer can link the narrative to the applicable version of the interface and the consent-event sequence. Versioning is indispensable: the causal mechanism may cease operating while new complaints arrive about the historic version.

The FCA's (2018) ICOBS demands-and-needs rules illustrate the substantive distinction between matching the proposed contract with the customer's requirements and the existence of personalized advice. A non-advised distribution process may still have a relevant demands-and-needs obligation. Coding all non-advised complaints as disclosure only would systematically miss product compatibility failures.

Evidence hierarchy and attribution

An investigation should resolve the applicable obligations and preserve contemporaneous original evidence before interviewing staff or modifying the relevant records. Contemporaneous policy versions, call logs, signed records, timestamps, suitability documents, audit trails, and sales-target schedules should be evaluated for provenance and completeness. Consumer testimony provides evidence of communication, expectations, and reliance. A scripted form marked complete is evidence that an event was recorded; it is not in itself proof of a meaningful exchange. A missing recording is likewise distinct from affirmative proof of what occurred.

The FCA (2017) PPI complaint guidance explicitly addresses evaluation of sales evidence, systemic practices, rejected claims, and purchase counterfactuals. Its provisions pertain to payment protection insurance and cannot be imported unchanged into other legal contexts. They illustrate the broader research logic of connecting a plausible common practice to the facts of a particular transaction.

A root cause should identify an actionable process or condition that explains why the observable failure occurred. The statement that an adviser sold an unsuitable policy identifies a failure event. A compensation threshold, deficient decision rule, or training material that repeated a misleading explanation provides a testable causal hypothesis. Evidence may establish more than one contributing cause. A flawed product eligibility table can coexist with steering incentives and incomplete checking by the seller.

A practical record should therefore include a primary cause, secondary causes, the level of evidence for each, a versioned reference to relevant records, and the status of human review. Counts of primary causes can sum to the number of coded cases. Counts of all implicated causes may exceed it; reports must specify which convention applies. Unknown and disputed causes should be retained in the denominator for the quality of classification rather than silently redistributed.

Coding reliability and classification error

A stratified sample should be independently coded by two trained analysts blinded where feasible to the previous case outcome. Raw agreement can be reported with the confusion matrix and Cohen's kappa, defined as kappa = (p_o - p_e)/(1 - p_e), where p_o is observed agreement and p_e is chance agreement computed from the two sets of marginal category proportions. Rare codes can yield unstable or counterintuitive kappa, so per-category precision, recall against an adjudicated sample, and disagreement narratives are useful companions. The adjudication panel should resolve factual uncertainty separately from disagreements about the codebook.

A measured false-negative rate in the presence of an audit reference is directly relevant to estimates of systemic reach. If a classifier trained on adjudicated complaints misses unfamiliar reasons expressed in other languages, the missing records reflect limitations of surveillance as well as underlying behavior. A statistical language model can assist in identifying similar narratives or versions; its output should remain a coding suggestion until validated. Confidential information must be minimized, access controlled, and decision trails available. IAIS (2024) customer information safeguards make the privacy issue part of conduct supervision, not a separate afterthought.

Reporting Bases, Exposure Normalization and Reporting Bias

Defining the estimand

An observed rate has a numerator, denominator, reporting rule, observation window, and unit. For new sale complaints, a relevant cohort rate can be defined as the number of distinct sales with at least one qualifying reported complaint within a fixed follow-up horizon divided by the number of eligible sales in the same inception cohort, multiplied by 1,000. The count of distinct sales differs from the number of complaint contacts and the number of unique customers. Renewal or switching grievances may require a different exposure base.

A calendar-period flow rate, by contrast, may use newly registered complaints in the period divided by policies in force or policy-years. It is useful for operational workload and customer contact, but it combines sales from different vintages. If 500 complaints about policies sold in 2018 surface during a 2026 campaign, dividing them by 2026 new sales would confound the historical exposure and current operation. A sales cohort table should preserve inception, discovery, intake, decision, and implementation dates.

The FCA's (2026) opened complaint data describes contextual rates normalized per 1,000 relevant policies or accounts and a revision to historical 2025 data following a reporting anomaly. That revision demonstrates the necessity of preserving data vintage and avoiding the interpretation of a corrected series as fresh change in actual conduct. The accessible page was a May 2026 snapshot; the latest numerical tables were not re-estimated here.

Composition adjustment

Suppose a distributor offers two channels, A and B. In a hypothetical baseline period, A registers 80 complaints from 10,000 sales, and B registers 20 from 10,000. Rates per thousand are 8 and 2, and the combined rate is 5. In the following period A registers 36 complaints from 4,000 sales and B registers 48 from 16,000, giving rates 9 and 3. The aggregate rate becomes 4.2 even though each channel's rate has increased by 1.

Fixing the baseline exposure shares at 0.5 and 0.5 gives standardized rates of 5 and 6. The decline in the crude rate is explained by movement toward the lower-rate channel; the standardized comparison reveals deterioration within channels. A similar adjustment can be made for product, geography, vintage, advised status, distribution partner, or customer segment when adequate cell support exists.

For stratum g and period t, define r_gt = 1000 C_gt / E_gt and a standardized rate R_t = sum_g w_g r_gt, with nonnegative fixed weights summing to one. The reference distribution must be disclosed, cells with zero exposure cannot contribute an estimate, and sparse cells should be displayed with uncertainty rather than imputed silently. If the choice of channel itself is influenced by harmful conduct, composition adjustment describes within-channel differences but may remove part of the total policy effect; both crude and standardized metrics should be retained.

From reported complaints to underlying harm

A simplified identification argument makes the selection problem explicit. Let H be the number of genuinely harmful sales in a cohort, q the probability that an injured purchaser recognizes harm and expresses a qualifying complaint within the horizon, and s the conditional probability that the organization captures it. When only one qualifying recorded complaint per harmful sale is counted, expected verified recorded harmful sales equal H q s, assuming classification accuracy and the defined conditional probabilities. The complete recorded complaint count can also contain unsubstantiated allegations and non-misconduct dissatisfaction; it is not identical to this verified count.

If 60 verified injured sales are observed and q = 0.30 and s = 0.80 are assumed, the implied count is 250. If q = 0.15 with capture unchanged, it is 500. These are scenario calculations, not empirical estimates. In general q and s vary with the magnitude of harm, available documentation, language, legal awareness, channel, and the ability to challenge the institution. Without independent information, the components cannot be separated using the complaint count alone. For comparisons across time, a reporting campaign may change q, and a better intake system may change s.

ASIC's (2024) general insurance complaint review provides evidence of capture imperfections: the review found that approximately one-sixth of complaints in its assessment were missed by insurers. The reviewed population included general insurance complaints rather than solely sales cases. Its result motivates a capture audit, not a universal adjustment factor for insurance sales.

EIOPA (2026) reports that French insurance mediation volumes increased substantially following improved access and simplification alongside other factors. The French mediation example is particularly relevant: growth in complaints can reflect reduced barriers to redress as well as deteriorating customer experiences. A performance target requiring a mechanical reduction in volume can therefore penalize accessibility.

Count models and lag structure

When counts are small, sampling variability can dominate apparent performance. Under an independent Poisson approximation, an observed count C has a rough standard error of sqrt(C), and the rate per thousand has standard error 1000 sqrt(C)/E. For a zero count an exact one-sided 95% Poisson upper bound is approximately 3/E in rate units before scaling. Dependence among branches or sales staff, repeated customers, and common scripts can invalidate the independent approximation. A negative binomial model or clustered resampling may better reflect overdispersion.

A regression of count on product, channel, time, and relevant covariates can use log exposure as an offset. If exposure is policy-years, the coefficient describes a rate per policy-time; if it is new sales, it describes complaints per sale within the design window. An estimate is meaningful only when the numerator and time at risk correspond. For time to first complaint, right-censoring must be addressed; cohorts observed for only three months are not directly comparable with cohorts followed for two years. A survival or cumulative incidence analysis can preserve follow-up and competing events such as lapse, provided the interpretation of censoring is made explicit.

Data integrity and surveillance fairness

A minimum analytic extract should contain pseudonymous sale and customer identifiers, product and contract versions, channel and distributor, exposure and cohort, relevant complaint-event times, allegation and root-cause codes, evidence status, decision, and remedy reconciliation. It should also carry the definitions and changes in reporting policy. Records from different organizations must be linked with permissible identifiers and deduplicated; absent links should be counted as a data-quality limitation.

Accessible complaint processes may elicit otherwise unobserved harms. IAIS (2025), in its fair treatment application paper announcement, places diverse consumer needs across the product lifecycle including complaints. An operational implication is to check accessibility through multiple channels and evaluate rates by segment with sufficient privacy protection. A low complaint rate among consumers facing greater barriers cannot establish a low rate of misconduct.

Broker and Insurer Responsibility for Redress

Responsibility along the chain

A broker's recommendation, an insurer's product design, a bank's bundling choices, and a service provider's interface may jointly shape the same sale. Investigations should reconstruct the functions performed rather than infer responsibility from the order in which customers encountered entities. For each hypothesized defect establish the entity controlling the decision, the duties applicable at the transaction date, relevant contracts and delegations, and the entity able to implement the relevant remedy. Intercompany allocation of remediation costs and customer-facing responsibility are distinct questions.

The IAIS (2024) conduct responsibility provisions are useful for analyzing the chain, while the EU directive's product governance provisions identify obligations relevant to manufacturing and distribution within their scope. A distributor may hold the only sales recording and the insurer the only premium and claims data. Neither can reliably estimate injury without a mechanism to exchange the necessary information lawfully. The remedy must follow the applicable legal framework; a shared operational investigation does not itself create joint legal liability.

A customer-facing arrangement should prevent repeated handoffs. A complaint received by one participant can be logged with the date, acknowledged, transferred through an agreed secure process where appropriate, and tracked until the designated party confirms resolution. If two entities separately register the case, internal operational counts can retain both while the joint remediation ledger reconciles the underlying sale. This protects against double reimbursement and omissions without erasing legitimate separate complaints.

Counterfactual redress

Redress restores an affected customer to the position justified by the facts, the relevant law or scheme, and the counterfactual course of events. A sound analysis distinguishes cancellation of an unwanted product, correction of an erroneous charge, restoration of coverage, replacement-policy reinstatement where available, and compensation for quantifiable financial detriment. Other remedies, including nonfinancial relief, depend on applicable rules and the facts.

ASIC (2022) RG 277 consumer remediation uses counterfactual restoration as the guiding approach and addresses interest or foregone returns and the use of consumer-beneficial assumptions. Its guidance is jurisdiction-specific. The transferable analytic task is to determine what would probably have happened without the relevant failure, then value the difference on a consistent basis while applying the applicable compensation law.

Consider an illustrative unwanted policy with premiums of 300 monetary units, financing interest attributable to those premiums of 18, a permissible offset of 40 for benefits already received, and an additional authorized time-value adjustment of 12. In a scheme in which those components are appropriate and nonoverlapping, redress is 300 + 18 - 40 + 12 = 290. A legal rule may determine that an offset is inappropriate, that interest compounds, or that the customer would have acquired another product; the arithmetic is conditional on those decisions. For a flawed replacement sale the major harm may be lost coverage or surrender value, which cannot be captured by merely refunding incremental premiums.

A useful transaction ledger records the counterfactual, each cash flow and date, method of interest, applicable tax treatment, prior payments, nonfinancial actions, and evidence of execution. Calculation quality can be checked through independently reproduced samples including boundary dates, partial policy periods, financed premiums, and cases with prior refunds. Report totals with a reconciliation to transactions rather than an unexplained sum of offers.

Reaching silent customers

The systemic inquiry should define the entire exposed population from the original decision process. A complaint can indicate a defect in a template applied to many sales. The number of affected customers should be identified through the template's version history, policy portfolio, and eligibility conditions. Records should distinguish confirmed injury, probable injury under documented permissible assumptions, no injury, and unresolved cases. Sampling is appropriate for estimating scope where complete review is impracticable; determining redress eligibility still requires a legally and operationally justified approach.

The FCA's (2019) systemic complaint guidance includes consideration of potentially disadvantaged customers who have not complained, under the relevant framework. ASIC's RG 277 likewise discusses assumptions that minimize exclusion of affected consumers and under-compensation. This supports a separation between the complainant register and the affected-customer register. A firm that compensates all complainants may still leave much of a systemic injury unremedied.

If records are missing because of a firm-side failure, a remediation procedure should document the significance of unavailable evidence and the applicable presumptions, rather than mechanically require evidence that the consumer could not reasonably have retained. Conservative identification of potential exposure can be paired with transparent, evidence-based cohort assumptions. Governance review should assess both erroneous exclusion and unjustified inclusion.

Supervisory Analytics, Reviews and Remedial Programs

Turning signals into investigations

Conduct supervision allocates attention under scarce resources. A review queue can combine complaint count relative to exposure, recent deviation from a stable baseline, severity, unusual product or staff concentrations, evidence of a shared process, and underrepresentation of groups unable to complain easily. The queue should distinguish an investigation priority from a conclusion of fault. A statistical alert may result from a temporary migration problem, a change in complaint definition, or other operational events. Investigation must determine which mechanism produced it.

ASIC's (2021) systemic issue requirements in RG 271 identify clear accountabilities, analysis, escalation, investigation and internal reporting. Paragraphs 118–120 are designated enforceable in the reviewed guide; paragraph 121 expresses ASIC's expectation for prompt identification and remediation of affected consumers when a systemic issue is confirmed. The legal force of different statements in one guide must be distinguished. The research implication is to establish a documented escalation and decision trail.

The initial review should freeze evidence, specify the risk hypothesis, identify transaction versions and exposed cohorts, select a sampling design, and estimate likely severity. The sample must contain both complaints and transactions without recorded complaints. A fraud, coercion, or other high-severity signal can justify urgent investigation even without adequate statistical power. Any temporary interruption of sales should be governed by the relevant authority, contract and legal obligations.

Product and channel interventions

Sales controls must be aligned with causes. A change in the commission schedule may address pressure to steer customers but will not correct inaccurate eligibility logic. Rewriting disclosure will not cure unauthorized enrollment if the consent interface continues to enroll by default. Training may improve recommendation quality, but its effects can decay when incentives or workload favor contrary behavior. Changes should target the mechanism, establish an owner, and specify an independently observable outcome.

EIOPA's (2022) credit protection insurance warning identifies problems involving choice, switching and remuneration in bank distribution. The supervisory evidence supports reviewing product and distribution design together, while its credit protection context limits generalization to all insurance. A high commission may be a warning indicator; a transaction-level analysis needs to establish whether commercial incentives and available alternatives affected the customer outcome.

A sales journey test complements complaint analysis by exposing the process directly. EIOPA's (2025) coordinated mystery shopping results report that 84% of discussed products met at least one predefined shopper need or objective, with limited correlation between process thoroughness and alignment. That particular criterion does not establish full suitability, market-wide compliance, or all objectives being met. Its methodological importance is that documentation length and consumer outcome are different endpoints. Joint analysis should examine actual product match, evidence quality, price and limitations, and the consumer's understanding.

Measuring remediation delivery

The redress pipeline contains successive states: exposed, assessed, eligible, offered, accepted where relevant, implemented, reconciled, and closed. Denominators should follow the relevant cohort at each stage. An expedited workflow may improve the time to decision while increasing the delay between decision and payment; performance evaluated only at the decision stage misses this result.

Indicator Numerator or time event Denominator or origin
Assessment coverage Exposed sales assessed Identified exposed sales
Eligibility coverage Eligible customers reached or validly actioned All identified eligible customers
Payment completion Reconciled paid customers Eligible customers due monetary remedy
Value completion Verified paid amount Assessed amount due under the scheme
Decision latency Decision timestamp Complaint receipt or remediation identification
Implementation latency Completed remedy timestamp Decision or eligibility timestamp
Residual injury Independently estimated unresolved loss Relevant affected cohort

Each indicator needs a reporting cutoff, definition for partial and disputed cases, and an explicit treatment of unreachable customers. The average payment completion time among paid customers excludes those still waiting. Time-to-completion analysis should retain open cases and display their outstanding age; median and tail statistics should be presented with cohort coverage. A portfolio can report 95% completion by customer count while a small number of unpaid large injuries account for most outstanding value.

In ASIC's (2024) review of implementation recording, most reviewed insurers lacked adequate recording of when complaint outcomes were actually delivered. This is a concrete reason to reconcile payments with bank or accounting settlement rather than rely on case closure flags. It is not a measurement of the efficacy of a particular sales-remediation program.

Governance and incentives for the complaint function

A complaint unit should be measured against data completeness, integrity and customer outcomes rather than a target that mechanically penalizes receipt of complaints. Root-cause analysts need authority to test the systems operated by the sales division. Where commercial performance measures reward sales volume, independent QA sampling should not be allocated entirely by the same commercial managers. Reporting to senior decision makers should separate unresolved material risks, blocked remediation cases, completed process changes and verified customer restoration.

A corrective program closes when there is evidence that the specified mechanism has been addressed, the exposed population has been investigated, required remedies have been delivered or handled under the applicable rules, and independent testing has found the new process operating as designed. A fall in complaints is supporting evidence only when exposure, capture and latency are sufficiently stable or accounted for.

Comparative Redress and Conduct Outcomes

European Union distribution evidence

The IDD provides EU minimum standards, while national supervision and redress operate within distinct national frameworks. The 2026 application report's French mobile-phone insurance campaign describes 197 mystery-shopping visits covering 11 distributors in 2024, with demands and needs not requested in 80% of visits. These findings concern that campaign and product context. They cannot be interpreted as the rate of mis-selling in all French insurance or used to infer what share of purchases caused monetary loss.

Alongside reported complaints, direct sales observation reveals failures that a customer may be unable to identify. The policy question is then whether changing the sales process reduces those failures without undermining meaningful coverage access. A market-specific comparison can use common scripted profiles and audit criteria; legal and linguistic variations must remain visible. A pooled estimate needs a documented sampling design and weights rather than an unweighted average of countries.

United Kingdom complaints and historic PPI

The UK FCA combines conduct rules such as ICOBS with requirements and guidance for complaint handling in DISP. Payment protection insurance illustrates the scale of remediation possible when a distribution issue persists across institutions and historical cohorts. The FCA's historical PPI redress summary reports that more than GBP 38.3 billion had been paid in refunds by April 2021. The figure is a cumulative monetary measure at an indicated date. It does not establish the number of injured sales, the amount still owed, the loss per policy, or the causal effect of a specific supervisory intervention.

The PPI-specific complaint framework draws attention to contemporaneous evidence and the purchase decision absent a failing. Its value for general insurance research lies in identifying the difference between correcting contract performance and evaluating the sale itself. A legally defined time limit changes the number and timing of complaints; the pre-deadline flow should not be compared naively with later complaints as if customer harm had disappeared.

The FCA's 2014 general insurance add-ons study combined quantitative and qualitative consumer research with a simulated behavioral experiment and compared add-on with standalone sales across five product markets. Its March 2014 report supports the methodological illustration without establishing a numerical experimental effect for the present analysis. This provides a useful example of examining the sales environment independently of later complaints, complementing historical redress counts with process-level evidence.

Australian internal dispute resolution and remediation

Australian guidance links complaints to systemic issue identification and sets expectations for investigation and remediation. RG 271 defines the relevant complaint-handling framework, while RG 277 provides separate remediation guidance. A comparison of the two instruments illustrates the distinction between resolving a complaint and restoring a population of affected customers.

In the 2024 ASIC review, 11 general insurers supplied complaints data; internal and external dispute resolution pathways had markedly different volumes and identified systemic issues. These channels involve different selection and aggregation processes. It would be misleading to divide systemic issue counts by their complaint totals and interpret the ratio as a causal measure that one organization is inherently more effective. A single systemic issue may encompass many customer cases, and external disputes are a selected subset. The appropriate comparison would track matched issue definitions, case escalations, time periods, and the additional information available at each stage.

The Australian remediation guidance highlights the treatment of missing records and counterfactual compensation. The measurement consequence is to store customer inclusion decisions and compensation assumptions as part of the reproducible research record, with tests for the effect of alternative justified assumptions.

United States complaint information

American insurance regulation is predominantly state-based. The NAIC (2025) Market Regulation Handbook describes confirmed complaints and variation in state confirmation criteria. An index using confirmed complaints differs from a measure of all reported dissatisfaction; confirmation depends on the regulatory adjudication convention as well as the underlying facts.

The NAIC (2022) complaint index explanation emphasizes that company and market complaints and premiums can affect the index. A conventional relative index compares the company's share of the relevant complaint count with its share of the premium base. If a firm accounts for 6% of confirmed complaints and 3% of relevant premium, the hypothetical relative index is 2. An index of 2 does not mean that 2% of purchasers were harmed. Premium exposure varies with price and risk mix, and comparison requires aligned lines of business, periods and geography.

A sales-specific index would require the portion of complaints independently identified as distribution related and an appropriate sales exposure base; the publicly described general index is not that measure. The United States example therefore illustrates how regulatory reporting conventions affect comparison.

Asia and institutional boundaries

The Hong Kong Insurance Authority's official complaint form (2025) expressly includes sales practices, misleading representations, and unsuitable advice within its conduct remit for insurers and licensed intermediaries. Its complaint handling guide (2021) explains that its regulatory assessment and follow-up do not extend to adjudicating claims or ordering compensation. These institutional documents distinguish supervisory investigation from compensation delivery; they do not establish an exhaustive account of Hong Kong law or a basis for comparative compensation calculations.

This distinction also applies more generally to interpreting foreign statistics: a supervisor receiving complaints for regulatory investigation and a dispute-resolution body determining individual remedies may count different populations. Institutions must be compared by function, admissibility and outcome definition.

Cross-jurisdictional comparability

Jurisdiction or standard Evidence unit studied Main interpretive limit
EU / EIOPA Supervisory reports and test purchases Different national conditions and selected product campaigns
UK / FCA Firm complaints, rules and historic refunds Complaint cohorts, legal eligibility and cumulative period
Australia / ASIC Internal dispute resolution and remediation reviews Broad insurance scope; not a sales-only population
US / NAIC State-based closed confirmed complaint information Confirmation and premium-denominator differences
Hong Kong / IA Conduct function, official complaint documents Institutional remit; no exhaustive legal or rate comparison
IAIS International supervisory standards and guidance Implementation depends on jurisdiction

Comparison is strongest for shared functional questions: whether consumers have an effective way to raise sales problems, whether common mechanisms are investigated, whether exposures and remedies can be traced, and whether changes are evaluated. A cross-national league table constructed from raw complaint totals would conflate these dimensions and should be avoided.

Causal Evaluation of Corrective Measures

Outcome specification and timing

The question whether a remediation program works contains two distinguishable tasks. Retrospective restoration asks whether historic affected customers received the appropriate remedy. Prospective prevention asks whether a corrective change reduced new harmful transactions relative to a credible counterfactual. The first task can be assessed through coverage and payment reconciliation against the defined eligibility and loss rules. The second requires an evaluation design that accounts for changes in clientele, product, distribution, and data capture.

Define the prospective outcome before intervention: for example, audited materially incompatible recommendations per 1,000 eligible sales within a fixed product family. Define independent secondary outcomes such as sales complaints observed after a common follow-up period, rates of unauthorized enrollment, accessibility of complaints, charge reversals, and customer understanding. Track coverage take-up and cancellations as potential unintended consequences. A reduced audit error rate with a simultaneous reduction in access to suitable insurance is a different outcome from improved matching with unchanged access.

Interrupted series and comparative designs

An interrupted time series can estimate a shift in level or slope after a known implementation date. A credible application requires sufficient observations before and after, consideration of seasonality, serial correlation, and concurrent events. In insurance, a new claims event, premium revision, regulatory reporting change, or media campaign may coincide with intervention. Alternative dates, negative-control outcomes, and stable exposure definitions can test whether the result is sensitive to these disruptions.

Where rollout occurs at different times across comparably situated channels, a difference-in-differences design may use unexposed or not-yet-exposed groups to estimate the change that would otherwise have occurred. Callaway and Sant'Anna (2021), in difference-in-differences with multiple periods, provide methods for group-time effects in staggered adoption designs under stated identification assumptions. A conduct application should specify the treatment group, intervention date, comparison group, no-anticipation and parallel-trends assumptions, overlap, and aggregation weights. Treatment effects may differ substantially across branches and time.

For illustration, suppose audited defective sales fall from 8 to 3 per 1,000 in treated branches and from 6 to 5 in comparable untreated branches. A simple two-period difference is (3 - 8) - (5 - 6) = -4 per 1,000. The computation subtracts the comparison-group change from the treated-group change. The interpretation as an avoided defect rate requires the counterfactual trend assumptions and unchanged measurement; if the treated branches simultaneously improve their detection of defects or change the product population, the arithmetic alone establishes no causal effect.

A common-control training program may create spillovers across nominally untreated sites. A branch-level design then needs to identify actual exposure rather than the administrative rollout label. Where participants are randomly assigned to an evaluation of permissible alternative training formats or process improvements, cluster randomization can address staff interactions, provided no group receives a knowingly deficient or unlawful process. Regulatory obligations remain applicable to all groups.

Detection changes and multiple outcomes

A particularly difficult intervention combines better complaint access with improved sales. The observed complaint rate might rise as newly accessible channels reveal prior injuries, while independent audits show fewer defective new sales. Estimating solely from complaint volume in this setting confounds prevention and detection. Separate sale-cohort outcomes from report-calendar outcomes, hold audit criteria stable, and document changes in access and capture.

Mechanism-specific results matter. Suppose sales advice errors decline after a decision-aid intervention, while unauthorized enrollment is unchanged. Aggregating both into a single complaint measure dilutes the effect on advice and risks misattributing unrelated variation. The evaluation should preregister primary outcomes, report secondary outcomes, provide uncertainty intervals, and distinguish confirmatory analysis from exploration. Multiple subgroup tests can produce false discoveries; small or privacy-sensitive cells require appropriate reporting safeguards.

Sensitivity, auditability, and reproducibility

A robust analysis includes missing-data sensitivity, alternative exposure definitions, alternative cohort windows, checks for classification drift, and potential changes in the customer mix. Test pre-intervention trends and assess overlap, recognizing that failure to reject a pretrend test does not prove parallel trends. If policy eligibility rules changed at the same time as the intervention, report whether restricting to stable eligibility populations changes the estimate and explain the resulting estimand.

For a population remediation, reproduce a random sample of calculations, oversample high-severity and poorly documented cases, verify that money reached the intended customers, and reconcile assessed liabilities with payments, unresolved cases, and permitted adjustments. Estimate uncertainty in the number of exposed customers and in counterfactual assumptions where justified. Distinguish a financial provisioning estimate from redress actually delivered; the accounting status need not coincide with consumer outcomes.

The analysis should preserve data dictionaries, code versions, cohort definitions, model and test specifications, and the source and date of each regulatory assumption. Reproducibility does not require releasing identifiable customer data; secure access arrangements or deidentified aggregates can be sufficient for appropriate independent review. Before reporting an effect, a reviewer should be able to reconstruct the numerator, denominator, period, comparison, and observed or modeled counterfactual.

Research agenda and inferential boundary

The cited institutional studies demonstrate selected process weaknesses, capture failures, supervisory approaches, and cumulative remedies. They do not jointly furnish a harmonized international panel of sale-level complaints and independent harms. A next empirical study would need matched sales, recordings or audit labels, independent customer surveys, external complaint linkage, operative policy and regulatory versions, and remedy outcomes. It could estimate capture probabilities by segment, distinguish time from purchase to recognition from time to resolution, compare corrective measures using pre-specified designs, and test the stability of classification across markets.

The scientific contribution of the present synthesis is the coordinated use of established measurement and causal methods across these stages. It identifies why reports of customer dissatisfaction, misconduct findings, and restored financial positions cannot be substituted for one another in statistical or legal conclusions. No new named theory, legal doctrine, or previously unverified empirical finding is advanced.

Conclusion

Insurance sales complaints provide a distinctive source of conduct evidence because they can reveal defects in the acquisition of a contingent financial promise long after that acquisition. Their usefulness depends on reconstructing the distribution mechanism that generated the alleged injury. Individual allegations should remain linked to independently assessed facts, the stage of causal contribution, the applicable legal duties, and the remedy justified by the circumstances.

The comparative evidence establishes a coherent set of research priorities. Conduct standards connect the sale with customer needs and fair treatment; complaint frameworks require investigation of recurring problems within their scope; independent supervisory observations expose weaknesses in capture and sales processes; remediation guidance addresses the identification and restoration of affected customers. The particular obligations and remedies vary across jurisdictions, products and historical periods.

Three conclusions follow. First, causal attribution requires evidence of a shared process as well as careful review of individual transactions. A claim denial may reveal a sales defect, but the denial is not the sales defect itself. Second, credible frequency measurement requires a matched exposure denominator, cohort follow-up, transparent classification, and recognition that reporting and capture may change. Third, successful redress is an implemented outcome for the affected population, not merely an offer, closed case, or cumulative payment sum. A preventive program should be evaluated against independently measured subsequent sales outcomes using a design that identifies an appropriate counterfactual.

These conclusions specify a research and audit method grounded in existing legal, supervisory and statistical work. The monograph's international perspective rests on functional comparison, rather than assuming that the statistical indicators or remedies of different jurisdictions are interchangeable. Subject to the applicable law, a scientifically defensible conduct assessment remains traceable from the transaction and complaint through the identified cause, population scope, remedy calculation, actual delivery, and tested subsequent performance.

References

  • European Parliament and Council of the European Union (2016). Directive (EU) 2016/97 of 20 January 2016 on insurance distribution (recast). Official Journal of the European Union L 26, 19–59. Insurance Distribution Directive

  • International Association of Insurance Supervisors (2024). Insurance Core Principles and Common Framework for the Supervision of Internationally Active Insurance Groups. December 2024, ICP 19, paragraphs 19.0.8–19.0.10 and 19.12. Insurance Core Principles

  • International Association of Insurance Supervisors (2025). IAIS publishes Application Paper on fair treatment of a wide range of consumers. July 7. fair treatment application paper announcement

  • European Insurance and Occupational Pensions Authority (2026). 3rd Report on the Application of the Insurance Distribution Directive (IDD). EIOPA-BoS-26-077, March 30, p. 23. country-specific supervisory findings

  • European Insurance and Occupational Pensions Authority (2025). EIOPA publishes the results of the first coordinated mystery shopping exercise in the EU’s insurance sector. June 17. coordinated mystery shopping results

  • European Insurance and Occupational Pensions Authority (2022). EIOPA calls for better value for money in bancassurance in warning to banks and insurers. October 4. credit protection insurance warning

  • Financial Conduct Authority (2018). FCA Handbook, Insurance: Conduct of Business Sourcebook, ICOBS 5: Identifying client needs and advising, particularly ICOBS 5.2.2–5.2.2D (effective October 1, 2018). ICOBS demands-and-needs rules

  • Financial Conduct Authority (2019). FCA Handbook, Dispute Resolution: Complaints, DISP 1.3: Complaints handling rules, especially DISP 1.3.3 R, 1.3.5 G and 1.3.6 G (latter effective April 1, 2019). systemic complaint guidance

  • Financial Conduct Authority (2017). FCA Handbook, DISP Appendix 3: Handling Payment Protection Insurance complaints, sections 3.2, 3.4–3.6 (relevant versions including August 29, 2017). PPI complaint guidance

  • Financial Conduct Authority (2026). Opened complaints. Data publication, May 15, 2026 revision snapshot. opened complaint data

  • Financial Conduct Authority (2023, updated 2025). PPI complaints. Consumer information and historical redress summary. historical PPI redress summary

  • Australian Securities and Investments Commission (2021). Regulatory Guide 271: Internal dispute resolution. September 2021, RG 271.117–271.121, pp. 34–35. systemic issue requirements

  • Australian Securities and Investments Commission (2024). REP 802: Cause for complaint: Complaints handling in general insurance. December 2024, pp. 4, 7. general insurance complaint review

  • Australian Securities and Investments Commission (2022). Regulatory Guide 277: Consumer remediation. September 2022, RG 277.69–277.85, 277.114–277.128. RG 277 consumer remediation

  • National Association of Insurance Commissioners (2025). Market Regulation Handbook. Chapter on publishing complaint information and confirmed complaints, pp. 125–126. Market Regulation Handbook

  • National Association of Insurance Commissioners (2022). How to File a Complaint and Research Complaints Against Insurance Carriers. Consumer article. complaint index explanation

  • Singh, J. (1988). Consumer Complaint Intentions and Behavior: Definitional and Taxonomical Issues. Journal of Marketing, 52(1), 93–107. DOI: 10.1177/002224298805200108. study of complaint behavior

  • Callaway, B., and Sant’Anna, P. H. C. (2021). Difference-in-Differences with Multiple Time Periods. Journal of Econometrics, 225(2), 200–230. DOI: 10.1016/j.jeconom.2020.12.001. Author preprint, arXiv:1803.09015, revised December 2020. difference-in-differences with multiple periods

  • Financial Conduct Authority (2014). General Insurance Add-Ons Market Study: Provisional Findings and Proposed Remedies. MS14/1, March 2014, p. 6. general insurance add-ons study

  • Insurance Authority of Hong Kong (2025). Complaint Form. July 2025, p. 1. official complaint form

  • Insurance Authority of Hong Kong (2021). How the Insurance Authority Handles Your Complaint. August 2021, pp. 3–4. complaint handling guide