An illustrative diagnostic for actuaries and claims leaders on insurance claims reserving data quality South Africa, inconsistent case estimates, IBNR and claims movements.
A claims loss development triangle can look actuarially troubling when the real issue is inconsistent claims data. For a South African insurer, that distinction matters. If case estimates are not captured, revised and reported consistently, the actuary may strengthen IBNR for the wrong reason, while the head of claims may defend operational performance using numbers that do not reconcile cleanly.
This is not a minor reporting irritation. It affects reserve adequacy, board confidence, audit discussions, SAM assurance and management’s view of whether claims inflation is real. In a market already dealing with repair delays, load-shedding effects, supplier disruption, legal backlogs and pressure on household and commercial policyholders, executives need to know whether a movement in the triangle reflects risk experience or data behaviour.
This article sets out an illustrative diagnostic scenario. It is not a client case. It reflects a pattern that can arise where claims operations, actuarial reserving and finance reporting rely on different interpretations of case estimates and claims movements. For broader context, see Zorinthia’s insurance data strategy and related insurance examples.
Consider a South African short-term insurer with a sizeable commercial motor book. Claims are notified through brokers, call centres and digital channels. Assessments are performed by external providers. Some larger claims are reviewed by senior claims handlers. Finance receives monthly summaries. The actuarial team receives an extract used to build incurred and paid development triangles.
At the latest quarterly reserving review, the incurred triangle shows unexpected deterioration between 12 and 24 months of development. Several accident quarters appear to be developing worse than expected. Paid claims have increased, but not enough to explain the full movement. The gap sits mainly in outstanding case estimates.
The actuary asks whether severity is worsening. The head of claims says the book has not changed materially, although imported parts are more expensive and some repairs are taking longer because workshops are affected by power interruptions. The finance team notes that month-end journals were adjusted to align outstanding claims to the general ledger. None of these explanations is impossible. None is sufficient on its own.
A review of the underlying claim records shows inconsistent case estimate handling. In some claims, the assessor’s expected repair cost overwrites the original case estimate. In others, a claims handler adds a separate reserve for storage, towing or legal costs. A few claims remain at a nominal estimate until the invoice is received. Large claims are adjusted manually in the reporting pack before the actuarial extract is finalised.
The triangle is not “wrong” in a simple sense. It is using data produced by real operational activity. The problem is that the same financial concept is not being treated the same way across claims teams, systems and reporting steps.
A loss development triangle organises claims experience by origin period and development age. Depending on the reserving approach, it may use paid claims, incurred claims or claim counts. For incurred triangles, the usual measure is paid amounts plus outstanding case estimates, sometimes adjusted for recoveries or other accounting treatments.
The triangle is useful because it shows how claims mature over time. It can reveal late reporting, changes in settlement speed, deterioration in severity, or unusual development in specific accident periods. But it cannot, by itself, distinguish between a real claims trend and a change in how claims are administered.
That is the core insurance claims reserving data quality South Africa issue. A triangle built from inconsistent case estimates may still produce a mathematically coherent result. The actuarial model can run. The chain ladder factors can be calculated. IBNR can be estimated. The weakness is not in the arithmetic. It is in the meaning of the input data.
For example, if one claims team updates case estimates when new information is received, while another waits until settlement authority is approved, the same claim type will appear to develop differently. If recoveries are netted against outstanding estimates in one extract but shown separately in another, incurred development changes again. If reopened claims are treated as new notifications in some cases, claim counts and average incurred amounts become distorted.
The executive question is therefore not only “What is the reserve?” It is “Can we explain why the reserve moved?”
The most useful diagnostic is often a movement-level review. Instead of debating the final triangle first, the actuary and head of claims should trace how selected claims moved from one valuation month to the next.
For each sampled claim, the review should reconcile:
This is not an exercise in checking every claim manually. A focused sample can be enough to reveal whether the problem is systematic. For instance, select claims from the accident quarters showing unusual development, split by branch, claim handler team, broker channel, large-loss indicator and repair supplier. Include both ordinary claims and claims with late legal or assessment activity.
The aim is to identify whether claims movements are being captured as economic changes or reporting artefacts. A genuine reserve increase because new damage is discovered is different from a system overwriting an old estimate without retaining the movement history. A payment that reduces outstanding reserve is different from a payment that is recorded while the original case estimate remains unchanged.
If movement history is weak, the triangle becomes a summary of current balances rather than a reliable record of development.
Before concluding that actuarial assumptions are too weak or claims handlers are reserving poorly, the insurer should test whether key terms mean the same thing across the business.
“Case estimate” is a common example. In one team, it may mean the best current estimate of gross claim cost before recoveries. In another, it may exclude VAT, towing or legal fees. In a specialist claims unit, it may include expected defence costs. Finance may report outstanding claims net of certain recoveries, while actuarial reserving may require a gross view.
The same applies to dates. An incident date, report date, captured date, assessment date and reserve change date are not interchangeable. If a case estimate is approved on Friday but captured after month-end because a branch was offline during load-shedding, the reporting treatment must be clear. Otherwise, development patterns can shift between months for administrative reasons.
A practical definition test should ask:
These are business control questions, not technical modelling questions. The actuary needs them answered because reserving depends on stable measurement. The head of claims needs them answered because operational performance cannot be defended if the numbers are assembled differently each month.
For a fuller governance view, see Zorinthia’s page on claims data governance.
South African insurers should also allow for genuine operational and economic drivers. Not every unusual triangle movement is a data defect.
Commercial motor claims may develop differently when imported parts prices rise, repair networks are overloaded, or vehicles spend longer off the road. Property claims may be affected by severe weather, municipal service delays, or supplier capacity constraints. Liability claims may move late when legal advice changes the expected settlement range. Load-shedding can affect call centre throughput, assessment scheduling, document submission and repair cycle times.
The diagnostic should therefore map data movements against operational events. If a new supplier panel was introduced, did estimate accuracy improve or worsen? If settlement authority thresholds changed, did reserve increases cluster around approval points? If a claims system migration occurred, were opening estimates converted cleanly? If a new large-loss review process was implemented, did it create a once-off strengthening that should be treated separately in reserving analysis?
This is where the actuary and head of claims need to work from the same evidence. The actuary can quantify development patterns. Claims leadership can explain process changes. Finance can confirm how balances were posted. Without that joint view, the insurer risks turning a data inconsistency into a reserve assumption, or dismissing a real deterioration as a system issue.
Claims data contains personal and sensitive information: identity numbers, vehicle details, medical reports, bank details, legal correspondence, assessor photographs and call recordings. A reserving diagnostic does not require unrestricted access to all of it.
POPIA should shape how the review is performed. The team should use only the fields needed for the reserving and movement analysis, restrict access to identifiable information, and avoid copying full claim files into uncontrolled spreadsheets. Where detailed file inspection is necessary, it should be limited, logged and justified.
This is particularly important where actuarial teams, finance teams, external advisers and claims operations all need to collaborate. Good governance does not mean blocking analysis. It means making sure the analysis uses the minimum necessary personal information, under clear accountability, with proper retention and access controls.
A focused diagnostic can usually be framed around one material portfolio and a limited number of valuation periods. It should produce a short, decision-ready view rather than a large technical report.
The output should show:
The conclusion may not be a single answer. The insurer may find that part of the reserve movement is genuine severity deterioration, part is delayed repair settlement, and part is inconsistent case estimate treatment. That is still progress. It allows management to avoid a broad, unsupported adjustment and instead take targeted action.
For the CFO, the important outcome is defensibility. For the actuary, it is a cleaner basis for IBNR and reserve judgement. For the head of claims, it is a clearer view of operational performance and leakage. For the board, it is confidence that reserving discussions are not being driven by hidden data defects.
If the incurred triangle moves materially, management should not begin by asking whether the model is sophisticated enough. The better first question is: “Can we trace the movement from claim-level events to the reserving triangle, using agreed definitions of case estimates, payments, recoveries and status changes?”
If the answer is no, the next action is not to buy another reporting tool. It is to run a controlled diagnostic on the claims movements behind the triangle, agree the business definitions, and fix the governance points that affect reserving decisions. That is the practical starting point for stronger claims reserving assurance in a South African insurer.