IFRS 17 and IFRS 9 were built under deadline. The bill for opacity is now coming due.

The implementation programmes are largely complete.

Across the financial services sector, institutions have invested hundreds of millions in new actuarial engines, expected credit loss frameworks, data platforms and finance architectures to comply with IFRS 17 and IFRS 9. The immediate regulatory objective has been achieved: financial statements are being produced, audits are being completed, and supervisory expectations are, in most cases, being met.

Yet a more fundamental question is beginning to emerge.

Do organisations understand the numbers their models produce sufficiently to use them as a basis for strategic decision-making?

For many institutions, the answer is considerably less certain than the audit opinion may suggest.

This distinction matters because regulatory compliance and decision usefulness are not equivalent objectives. One establishes whether reported financial information faithfully represents transactions in accordance with accounting standards. The other determines whether management understands the economic mechanisms driving those outcomes well enough to allocate capital, price risk and formulate strategy.

Increasingly, those two objectives are diverging.

 From Measurement to Understanding

IFRS 17 fundamentally changed the measurement of insurance contract liabilities. Estimates of future cash flows, explicit risk adjustment, contractual service margins and discount rates now interact within a highly dynamic valuation framework. Likewise, IFRS 9 replaced an incurred-loss impairment methodology with forward-looking expected credit loss estimation, requiring institutions to incorporate macroeconomic expectations into financial reporting.

These standards were designed to improve economic representation rather than simplify it.

The resulting models are therefore necessarily sophisticated. They integrate actuarial judgement, statistical estimation, economic forecasting and extensive assumptions regarding future uncertainty.

However, sophistication alone does not create understanding.

Many implementation programmes were undertaken under significant regulatory deadlines, constrained resources and evolving interpretation of the standards. Unsurprisingly, success was primarily measured by whether institutions could produce compliant financial statements within prescribed reporting timelines.

The consequence is that many organisations now possess highly capable measurement systems whose explanatory capability has received comparatively less attention.

A model may reconcile perfectly while providing limited insight into the underlying economic drivers of its outputs.

 The Difference Between Accurate Measurement and Useful Explanation

Financial reporting is intended to produce information that is relevant to economic decision-making. This objective extends beyond numerical accuracy.

Boards and executive committees do not merely require reported values; they require defensible explanations for why those values changed and whether those changes are expected to persist.

This distinction is often underestimated.

A model may estimate liabilities or expected losses with considerable precision while remaining unable to attribute movements uniquely to their underlying causes. In complex financial systems, multiple drivers frequently evolve simultaneously:

• claims frequency;
• claims severity;
• inflation;
• portfolio mix;
• management actions;
• policyholder behaviour;
• macroeconomic conditions; and
• changes in model assumptions.

Observed financial outcomes represent the combined effect of these interacting processes.

The statistical challenge is not simply estimating the outcome. It is determining whether available information contains sufficient independent evidence to isolate the contribution of each driver with acceptable confidence.

Frequently, it does not.

This is not necessarily a weakness of the model. It is a property of the data-generating process itself.

 Structural Limits to Explainability

Modern analytics can produce highly detailed movement analyses, variance decompositions and attribution reports. These outputs often reconcile exactly to reported balances and present coherent narratives regarding changes in profitability or reserves.

Reconciliation, however, should not be confused with identification.

Several different combinations of assumptions may explain the same observed outcome equally well. When explanatory variables exhibit strong interdependence—a common characteristic in insurance and banking—multiple plausible explanations can coexist without sufficient statistical evidence to distinguish between them.

Consequently, explanation becomes inherently uncertain even when measurement remains accurate.

This distinction has significant governance implications.

Where uncertainty surrounding explanation is understated, management may develop excessive confidence in strategic conclusions that are only weakly supported by available evidence.

The greatest model risk is therefore not necessarily inaccurate prediction.

It is unwarranted certainty regarding causation.

 A Practical Illustration

The experience of KB Insurance illustrates this distinction.

The insurer reported strong statutory profitability under applicable accounting standards. Nevertheless, movements in surrender value reserves materially reduced distributable earnings available to its parent company, rendering planned shareholder distributions unsustainable despite apparently healthy accounting performance.

Nothing within the accounting model was incorrect.

The financial statements faithfully represented the required accounting treatment.

However, the information required for strategic capital allocation differed from the information required for statutory reporting.

The accounting outcome and the economic decision were governed by different questions.

Institutions that fail to distinguish between these perspectives risk making capital allocation decisions that are technically compliant yet economically inefficient.

 The Parallel Under IFRS 9

An analogous challenge exists within expected credit loss modelling.

IFRS 9 requires institutions to estimate future credit losses using reasonable and supportable forward-looking information. This necessarily introduces macroeconomic scenarios, management overlays, expert judgement and model uncertainty into impairment estimation.

The resulting provisions satisfy the objectives of financial reporting.

Whether they provide sufficiently robust insight for lending strategy, pricing decisions or portfolio optimisation is a separate question.

Academic research has demonstrated that implementation choices under IFRS 9 can materially influence internal credit assessments and provisioning behaviour, even among institutions evaluating economically similar borrowers. Such variation does not necessarily imply non-compliance; rather, it illustrates that considerable judgement remains embedded within apparently objective models.

As economic uncertainty increases, the explanatory uncertainty surrounding those judgements increases accordingly.

Complexity Does Not Eliminate Uncertainty

A common organisational response to weak explanatory power is to increase model sophistication.

Additional variables are introduced.

Further segmentation is applied.

Machine learning techniques are incorporated.

While these developments frequently improve predictive performance, they do not necessarily improve explainability.

Indeed, increasing complexity may reduce the ability of management and governing bodies to evaluate whether model outputs remain economically plausible.

This creates a subtle governance paradox.

The more technically sophisticated a model becomes, the greater the need for disciplined model risk governance, independent validation and explicit communication of uncertainty.

Greater complexity should increase humility, not confidence.

Explainability as a Governance Requirement

Financial regulators increasingly emphasise governance over models rather than models themselves.

A robust modelling framework should therefore be capable of answering questions such as:

• Which assumptions contribute most materially to reported outcomes?
• How sensitive are conclusions to reasonable alternative assumptions?
• Which explanatory relationships are supported by evidence, and which rely primarily on expert judgement?
• What degree of uncertainty surrounds management’s interpretation of observed movements?
• Are strategic decisions being supported by causal evidence or merely by accounting outputs?

These questions extend beyond regulatory compliance.

They concern the quality of organisational decision-making.

Beyond Compliance

The implementation phase of IFRS 17 and IFRS 9 is largely complete.

The next phase is considerably more challenging.

Institutions must evolve from producing compliant financial information to developing analytical capability that supports explanation, governance and strategic decision-making.

This requires more than better models.

It requires explicit recognition of the limits of inference, disciplined treatment of uncertainty, rigorous attribution methodologies and governance frameworks capable of distinguishing measurement from explanation.

Compliance establishes that reported numbers can be relied upon.

Explainability establishes whether those numbers can be acted upon.

The distinction is becoming one of the defining governance challenges facing financial institutions.

The organisations that recognise it earliest are likely to make better strategic decisions—not because they possess more sophisticated models, but because they possess a more sophisticated understanding of what their models can, and cannot, legitimately explain.