2026 Presentation Summaries - Actuarial Society of South Africa

Presentation Summaries

THE 2-POT SYSTEM: REVIEW OF OBSERVED EXPERIENCE AND OUTCOMES

John Anderson

Presentation with Paper | 60-Minute
Relevant Practice Area: Retirement Matters
Suggested Audience Knowledge Level: Foundational

The 2-pot retirement system was introduced on 1 September 2024. It aims to balance short term liquidity needs with long term savings.

The paper aims to analyse industry experience, using data from large administrators from the private and public sector, representing the majority of retirement funds in South Africa.

Actual experience and trends are analysed, including engagement.

A propensity model is utilised in the analysis, to explain the main factors to predict the likelihood of accessing funds early or preserving. This model is beneficial to estimate liquidity requirements, as well as enabling segmentation of members in order to provide proactive interventions and communication.

The insights are useful to advisors, trustees, employers, administrators and regulators.

ECONOMIC COMPLEXITY AND INDUSTRY RETURNS IN SOUTH AFRICA

Adam Balusik, Dr Eden Gross

Presentation | 30-Minute
Relevant Practice Area: Investments
Suggested Audience Knowledge Level: Intermediate

This study investigates the evolution of South Africa’s economic complexity from 2008 to 2023 and its relationship with industry-level financial performance. The study draws from historic trade and economic complexity data from the Observatory of Economic Complexity and maps this to sectoral return data from the Johannesburg Stock Exchange. We examine how shifts in the country’s global competitiveness – observed through changes in its Economic Complexity Index – correlate with local investment performance across sectors. For academics, the findings provide empirical evidence linking macro-level complexity dynamics to sectoral performance, enriching the literature on structural transformation in emerging markets.

Given the theme of this year’s Convention, we believe that this presentation provides novel insight into extending the toolkit used by actuaries and will expose the audience to new methods of insight into understanding international economic flows.

WHAT IS FAIRNESS ANYWAY? CASE STUDIES ON THE EVOLUTION OF FAIRNESS IN FINANCIAL SERVICES

Siebert Benade, Sarika Besesar, Lufuno Netshitenzhe, Nakita Zackery

Presentation | 60-Minute
Relevant Practice Area: Professional Matters
Suggested Audience Knowledge Level: Foundational

Fairness sits at the core of actuarial work in financial services, yet it is rarely defined explicitly and is increasingly contested. Traditionally, fairness has been viewed through a technical lens: risk-based pricing, contractual correctness, and financial sustainability. However, evolving regulation, heightened consumer expectations, and reputational dynamics are expanding this definition beyond purely actuarial constructs.

This interactive session, presented by members of the Market Conduct Committee, explores how fairness is evolving across financial services. Using a series of realistic, South African case studies spanning life insurance, healthcare and short term insurance, the session will highlight key tensions actuaries face in practice. These include claims decisions, funding of high-cost medical treatments, and products that are technically sound but deliver poor outcomes for customers.

A central theme is the trade-off between individual and collective fairness. For example, in healthcare, funding a high-cost drug for one patient may materially impact affordability for the broader risk pool. Similarly, in insurance, decisions that are technically correct may still lead to outcomes perceived as unfair by customers or society.

The session will be interactive, using live polling to test audience views at key decision points. This will demonstrate how perceptions of fairness differ across stakeholders and evolve as more information becomes available. The aim is not to provide definitive answers, but to explore fairness in complex, real-world contexts that actuaries face.

This aligns with the Convention theme of Broadening Boundaries by challenging actuaries to move beyond traditional technical definitions and engage with broader societal, behavioural, and ethical considerations. It also reflects the increasing role of actuaries in areas such as market conduct, public interest, and policy development.

Practical outcomes for delegates include:
• Insight into key market conduct risks across the product lifecycle, including healthcare funding decisions, product design, and claims decisions.
• Greater awareness of the tension between individual and collective fairness, and how this impacts decision-making.
• Practical tools to support professional judgement where technical correctness and perceived fairness diverge.
• Real world examples of the actuary’s role in balancing stakeholder interests and acting in the public interest.

BROADENING RESERVING BOUNDARIES: DEEP NEURAL NETWORK IN LIFE INSURANCE

Dr Jan Blomerus

Presentation with Paper | 60-Minute
Relevant Practice Area: AI, Machine Learning & Data Science
Suggested Audience Knowledge Level: Intermediate/ Advanced

This presentation explores the use of deep neural networks in life insurance reserving, with a particular focus on how modern machine learning methods can complement and extend traditional actuarial approaches. The topic is based on my doctoral research in life insurance reserving using deep feed-forward neural networks and is motivated by the growing need for actuaries to engage with richer datasets and more complex relationships, with a new analytical toolkit.

The presentation aligns strongly with the Convention theme of “Broadening Boundaries”. It demonstrates how actuarial thinking can be expanded beyond established reserving techniques by incorporating artificial intelligence in a disciplined, transparent, and practically relevant way. Rather than positioning machine learning as a replacement for actuarial judgement, the session will show how these methods can form part of a broader actuarial toolkit for dealing with increasingly complex business environments.

The presentation will cover the context of the reserving problem, the rationale for considering deep neural networks, the structure of the modelling framework, and key findings from the research. It will compare the strengths and limitations of deep learning with those of more traditional methods, considering predictive

performance, model flexibility, interpretability, governance, and implementation challenges. Attention will also be given to the conditions under which these methods may be useful in practice, and where caution is required.

The originality of the work lies in applying deep neural networks to a core actuarial function in life insurance reserving, while maintaining a clear focus on practical actuarial relevance. The session is intended not only to present research results, but also to translate them into insights that can support practitioners, educators and decision-makers.

Attendees can expect several practical outcomes. First, they will gain a clearer understanding of where deep learning may add value in reserving applications. Second, they will be equipped to assess the opportunities and risks of using such models in actuarial work. Third, they will gain a framework for thinking about how artificial intelligence can be adopted responsibly within actuarial practice, including issues of validation, professional judgement and communication. More broadly, the session aims to encourage actuaries to engage confidently with emerging analytical methods and to consider how the profession can contribute in wider and more data-driven contexts.

SA3 SALARY SURVEY INSIGHTS OVER THE LAST DECADE

Wilhelm De Wet, Jurie Gouws

Presentation | 60-Minute
Relevant Practice Area: Wider Fields
Suggested Audience Knowledge Level: Foundational

SA3 has successfully conducted an annual Salary Survey among actuarial professionals in South Africa for the past 10 years. The results are shared with participants, and with over 1,500 participants in 2025, representing a significant portion of the Actuarial Community in South Africa, it is widely used as a benchmark for salary discussions across the country.

At the 2020 and 2023 ASSA Conventions, we presented insights from our SA3 Salary Surveys, with these sessions proving to be among the most popular at each event.

This year we aim to revisit and expand on these insights, now drawing on data from 2016 – 2025. We will explore remuneration and broader actuarial market dynamics in greater depth, with the objective of shedding light on key trends and developments in the South African remuneration landscape.

For this year’s presentation, we intend to:
• Share key findings from our 2025 SA3 Salary Survey.
• Highlight key remuneration trends over the past decade.
• Share demographic and market related insights and trends.
• Provide a sneak peak into the 2026 Salary Survey’s results.
• In line with this year’s theme, “Broadening Boundaries”, compare South African remuneration and cost-of-living indexes with a selection of international markets.

For attendees managing teams or representing HR functions, we believe this presentation will support you in remaining competitive, attracting top talent, and making informed remuneration decisions.

For those building their careers, we hope this presentation will empower you with objective, data-driven insights into market remuneration, helping to ensure fair compensation and strengthening your position during hiring or remuneration reviews.

We look forward to sharing our findings and supporting our attendees in navigating the evolving compensation landscape with confidence.

CROP SUITABILITY, YIELD VOLATILITY AND LOSS DISTRIBUTIONS SHIFTS UNDER CLIMATE CHANGE: AN ACTUARIAL MODELLING FRAMEWORK

Dylan Durieux, Jerome Mahadeo, Kivaan Naidu

Presentation with Paper | 60-Minute
Relevant Practice Area: Climate change
Suggested Audience Knowledge Level: Foundational

In climate-sensitive agricultural systems, shifts in crop suitability act as a transmission channel through which climate impacts manifest as financial losses and macroeconomic volatility. This paper presents a modelling framework that links climate variability, remotely observed crop conditions, and actuarial risk outcomes in.

South Africa. As climate change alters rainfall patterns and growing conditions, climate dependent changes in crop suitability and vegetative health may cause volatility in agricultural output and trade.

Using satellite-derived indicators – including the Normalised Difference Vegetation Index (NDVI), soil moisture proxies, and climate data – we model how acute hazards and chronic climatic trends affect crop growth dynamics and yield variability for nutritionally important crops such as maize and wheat. Climate-linked yield distributions are generated and translated into potential farm-level revenue impacts, which are aggregated across regions to capture production risk.

Given the structure of South Africa’s agricultural system, the framework places particular emphasis on a subset of staple field crops that are considered critical for the food security of the country. These crops are also central to regional trade balances and exhibit strong sensitivity to climate change. By focussing on these systematically important crops, the framework is calibrated to capture the channels through which climate risk most materially propagates into food availability, price dynamics and financial loss distributions, while retaining general applicability to other crop types.

The framework captures both acute risks – such as droughts, floods, heatwaves, hail, and wildfire – and chronic risks, including shifts in temperature distributions, rainfall seasonality, evapotranspiration, and water availability. Multi-source data are integrated across satellite observations, station and climate data, farm and district production statistics, and hydrological indicators.

Yield variability is ultimately translated into loss metrics, including exceedance probabilities, value-at-risk, and tail risk measures, with explicit treatment of regional and cross-crop dependence. The results demonstrate how climate-driven changes in crop suitability can materially alter loss distributions, correlation structures, and capital requirements over time. Embedding Earth observation indicators into workflows provides a practical pathway for climate-informed pricing, reserving, capital modelling, stress testing, and climate risk governance in banking and insurance. Keywords: Climate risk, banking, insurance, agriculture

DISABILITY TRENDS FORM DOWN UNDER: INSIGHTS FOR THE SOUTH AFRICAM MARKET

Izahn Findlay, Ingrid van den Goorbergh, Lize-Mari Wiggill

Presentation | 60-Minute
Relevant Practice Area: Life Insurance
Suggested Audience Knowledge Level: Foundational

At previous conventions, we discussed the deteriorating income protection claims experience in Australia and highlighted existing challenges in the South African group market. We have since observed similar trends for Australian Total and Permanent Disability (TPD) products. This presentation builds on those discussions by providing an overview of recent Australian TPD claims experience and examining its relevance for South African disability products. In line with the Convention theme of Broadening Boundaries, the session draws on international experience to explore how lessons from one market can inform sustainable product development and risk management in another.

Drawing on experience analysis and insights from market experts, the discussion illustrates how societal factors, product design, underwriting practices, claims decision-making, and regulatory settings shape TPD outcomes in Australia. Perspectives are broadened through engagement with South African and global product specialists, as well as underwriting and claims professionals. Within this context, the role of mental health claims in emerging experience is examined, alongside the importance of insurable interest and financial underwriting in interpreting observed trends.

The session highlights areas of relative resilience in South Africa, alongside risk factors that warrant continued focus from pricing, valuation and product design perspectives. Attendees will gain practical insights that support informed decision-making and sustainable disability product development within the South African market.

Key Outcomes:
• Gain an understanding of similarities and differences between the Australian and South African TPD markets.
• Consider how various stakeholders influence the success and sustainability of a TPD product.
• Understand potential influences of mental health-related claims in Australia and South Africa.
• Discuss the importance of financial underwriting at inception and over the lifetime of the policy.
• Identify potential risk factors relevant to pricing, valuation and product design.

THE HUMAN SIDE OF WORK: ITS OKAY NOT TO BE OKAY

Baveshan Harbhajan

Panel Discussion | 60-Minute
Relevant Practice Area: Life Insurance
Suggested Audience Knowledge Level: Foundational

Mental health plays a vital role in employee performance and engagement, particularly within modern, fast-paced work environments.

Actuaries operate under significant pressure, given the considerable impact of errors in their work and the demands imposed by lengthy qualification examinations, which can contribute to burnout and anxiety. The profession requires sustained attention, leading to increased mental fatigue, while prevalent personality traits such as conscientiousness often foster perfectionism and internalised stress. Continuous accountability and the behind-the-scenes nature of actuarial work may result in feelings of isolation and difficulty detaching from professional responsibilities.

By prioritising mental health, organisations and employees alike build resilience, encourage collaboration, and ensure sustainable success. Our panel seeks to address the specific challenges actuaries encounter and discuss ways for individuals to maintain their mental wellbeing.

READ, READ AGAINST THE DYING OF THE LIGHT: THE CASE FOR ACTUARIES AS READERS IN A POST-LITERATE SOCIETY

Pamela Hellig

Presentation with Paper | 60-Minute
Relevant Practice Area: Education and Professional Development
Suggested Audience Knowledge Level: Foundational

“What Orwell feared were those who would ban books. What Huxley feared was that there would be no reason to ban a book because there would be no one who wanted to read one.” Neil Postman, Amusing Ourselves to Death

The ability and appetite to read effectively are beneficial, if not essential, to thriving as an actuary and to many productive pursuits of human society in general. Reading, i.a
• supports a wide range of cognitive functions, including stronger memory, better concentration, and a slower rate of mental decline.
• deepens our understanding of human experience by allowing us to inhabit different perspectives, strengthening empathy and interpersonal insight.
• cultivates forms of analytical and structured thinking that are otherwise impossible to develop.
• fuels creativity and intellectual innovation by connecting readers to the accumulated knowledge contained in written works and sharpening their ability to evaluate, challenge, and extend that knowledge.

Yet, technological and social shifts-including the rise of smartphones, short‑form content, and AI-have contributed to what many describe as a post‑literate society. In such a context, the ability to read deeply or write cogently is no longer assumed, nor widely practiced. South Africa’s well‑documented literacy crisiS reflects additional historical and structural factors: as of 2021, 81% of Grade 4 learners could not read for meaning in any language, highlighting the scale and urgency of the problem.

This paper examines the implications of a post‑literate society for the actuarial profession. It explores how declining literacy – both in the form of illiteracy (inability to read) and aliteracy (ability but unwillingness to read) – threatens critical competencies across the actuarial lifecycle. It also proposes practical measures that educators, employers, institutions, and individuals can take to reinforce reading and writing as essential professional skills.

The topic aligns strongly with the convention’s theme of “Broadening Boundaries”. By engaging with an issue that falls outside conventional actuarial discourse but has profound consequences for the profession’s long‑term sustainability, the paper expands our understanding of the forces shaping the profession’s future. Attendees of this presentation will:
– learn about the historical trajectory from oral cultures to literate societies, and understand the emerging shift toward post-literacy;
– gain insight into the broad and sometimes under‑recognised benefits of literacy, including how these benefits underpin the analytical, ethical, and professional demands placed on actuaries;
– consider practical recommendations on how the actuarial community can address both illiteracy and aliteracy at key stages of education and professional practice.

LEAN AND MEAN ORSA: GETTING THE BOARD ONBOARD

Hannaa Hoosen, Zante Killian, Jandrei Strauss

Presentation | 30-Minute
Relevant Practice Area: Enterprise Risk Management
Suggested Audience Knowledge Level: Intermediate

This year’s ASSA Convention challenges actuaries to broaden the boundaries of our work and deepen the value we bring to our organisations. In that spirit, this presentation explores how the Own Risk and Solvency Assessment (ORSA) can evolve from a compliance-driven exercise into a decision-useful tool that meaningfully supports boards in steering the business.

Due to lengthy compliance requirements, many ORSAs have become highly detailed and technical, which can make it harder to translate outputs into business decisions. This provides an opportunity to consider how the ORSA can add more value – by refining the narrative and providing clear, forward-looking insights to support decisions.

Recently, we reviewed an ORSA as the Head of Actuarial Function that challenged conventional thinking. It adopted a deliberately “lean and mean” approach – moving away from multiple scenarios and extensive outputs, and instead focusing on a single, deeply explored scenario developed through cross-functional engagement. It was both comprehensive and practical, bringing together the full insurance value chain to assess what could realistically threaten sustainability, and how the organisation would respond in both the short and longer term. A key theme is making the ORSA tangible and actionable. When done well, the ORSA becomes a tool that informs real choices, rather than a document that simply records them.

We shall also reflect on the importance of iteration. Each ORSA cycle provides an opportunity to revisit assumptions, unpack emerging risks, and refine insights year by year. Embedding this process into the rhythm of the business ensures that it remains relevant, responsive, and aligned with evolving priorities and risks.

Ultimately, this session does not propose a single “right” way to approach the ORSA. Instead, it offers a perspective on how a more focused, business-engaged approach can enhance its impact. By prioritising clarity, integration, and practical application, actuaries can help position the ORSA as a tool that boards actively use – rather than one they simply review. Join us as we challenge the audience to consider whether less can mean more to truly get the board on board.

THE MISSING C-SUITE: GOVERNING ALGORITHMIC DECISION-MAKING IN FINANCIAL SYSTEMS THROUGH THE AI SPINE FRAMEWORK

Ziyanda Kitantou

Presentation with Paper | 30-Minute
Relevant Practice Area: Wider Fields
Suggested Audience Knowledge Level: Intermediate

Across South African financial services, credit limits are set, claims are settled, payments are routed, and premiums are priced by automated systems performing functions historically held by named executives, yet without the named executive’s accountability scaffolding. This paper terms the resulting condition the missing C-Suite: a structural gap in which the work of senior decision-making is being done at scale and in real time, while the governance machinery the institution was designed to require has not followed.

The paper’s central argument is that financial risk enters at the decision, not at the model. Existing Model Risk Management frameworks therefore govern the wrong layer, leaving the consequential step (where an algorithm determines whether a claim is paid, credit is extended, or a premium is set) structurally unowned.

Aligned with the ASSA 2026 theme of Broadening Boundaries, it locates this gap within the actuarial profession’s natural domain of risk quantification and capital adequacy.

To close it, the paper introduces the AI Spine, a governance architecture that restores named accountability across algorithmic decision systems. It is operationalised by two quantitative instruments: the Decision Materiality Index (DMI), a seven-dimension methodology that triages decision classes by materiality, and the Governance Capital Add-on (GCA), which quantifies the capital implication of governance deficit within the ORSA framework under SAM Pillar II. The architecture is situated against emerging international regulation (including the EU AI Act and the Basel Committee’s 2024 report on the digitalisation of finance) and the November 2025 PA/FSCA joint report on AI in the South African financial sector.

The paper concludes that algorithmic governance is an emerging actuarial domain in its own right, distinct from the model risk management discipline that produced it: the governance of autonomous financial decision systems. For the actuarial profession, this represents both a professional obligation and a strategic opportunity to define the next frontier of fiduciary risk management in South African financial services.

KEYWORDS
Algorithmic governance; AI Spine; Decision Materiality Index; ORSA; SAM Pillar II; Governance Capital Add-on; November 2025 PA/FSCA joint report; algorithmic accountability

THE DATA WE DON’T SEE: WHAT THE PINKDRIVE TEACHES ACTUARIES ABOUT THE PROTECTION GAP

Sandra Lehmann, Noelene Kotschan

Presentation | 30-Minute
Relevant Practice Area: Public Interest

Actuaries model risk for insured populations. PinkDrive works with the people our models never see — the medically uninsured, in communities where formal healthcare doesn’t reach. If the profession is serious about broadening its boundaries, it needs to understand what the protection gap actually looks like on the ground, and what that means for product design, public policy, and the public interest mandate actuaries claim to hold.

This presentation brings PinkDrive founder and CEO Noelene Kotschan into the room to challenge actuaries with the reality of healthcare delivery to uninsured South Africans — and to ask what role actuarial thinking should play in closing the gap.

The invisible population
PinkDrive deploys mobile mammography units staffed by qualified medical practitioners to rural and township communities across South Africa, delivering free cancer screening to thousands of women with no access to formal healthcare. What does the cancer burden look like when you go to the people instead of waiting for them to come to you?

What actuaries assume vs. what’s actually happening
Actuarial models of morbidity, incidence, and claims experience are built on insured-population data — a fundamentally incomplete picture. PinkDrive’s screening data reveals what actuaries don’t see: late-stage detection rates, geographic disparities, and the real cost of the protection gap in uninsured communities.

The NHI that already exists
While the profession debates NHI in theory, PinkDrive has been running a working version since 2009. It pools third-party funding from corporate donors and partnerships — not government, not user premiums. It delivers free-at-point-of-care services with no means-testing. This is the NHI funding-and-delivery architecture in miniature, sustained for over 15 years without government funding. What could actuaries learn from how this model works — and what could PinkDrive gain if actuarial tools like funding adequacy modelling and cost-per-life-year analysis were applied to its operations?

Design from the ground up
PinkDrive didn’t build a hospital and hope people would come. They put a mammography machine on a truck and drove it to Soweto. This is human-centred design in healthcare delivery — the kind of thinking actuaries need when designing for financial inclusion.

The public interest challenge
Actuaries hold a professional obligation to act in the public interest. PinkDrive lives one. What would it look like if actuarial expertise were deployed — practically, on the ground — to help organisations like PinkDrive optimise their reach, model their sustainability, and measure their impact??

MEASURING WELFARE GAINS FROM INCOME-TARGETED BOND PORTFOLIOS IN RETIREMENT DECUMULATION

Dr Shaun Levitan

Presentation with Paper | 60-Minute
Relevant Practice Area: Retirement & Investments
Suggested Audience Knowledge Level: Intermediate

This paper examines a critical weakness in current defined contribution retirement decumulation practice: the widespread use of conventional bond indices as the “safe” asset, even though they do not hedge the income objective retirees actually face. Retirees need a stable and sustainable stream of post-retirement income, yet standard fixed-income allocations are generally designed around capital preservation rather than liability hedging. Building on the work of Merton, Muralidhar and Martellini, the paper evaluates whether a retirement bond — a deferred, fully amortising, income-targeted bond portfolio designed to support stable retirement withdrawals — can materially improve retirement outcomes relative to traditional bond-index-based strategies.

The paper’s originality lies in extending prior research beyond deterministic back-tests into a fully stochastic decumulation framework. It models interest-rate dynamics, equity returns, bond-index performance and retirement-bond pricing within an integrated simulation framework, and evaluates outcomes using Monte Carlo simulations. Performance is assessed using measures that are directly relevant to retirees: mean withdrawals per dollar of retirement wealth and downside income risk, measured through the semivolatility of annual withdrawal changes. This allows the Programme Committee to assess the work not only as a conceptual contribution, but also as a rigorous and practically relevant framework grounded in real retirement objectives.

The topic aligns strongly with a convention theme focused on innovation, actuarial relevance and improving member outcomes, because it addresses one of the most important unresolved questions in retirement design: what should count as the true risk-free asset in decumulation? The analysis shows that replacing a conventional bond index with a retirement bond improves the decumulation frontier by reducing downside income risk and increasing expected retirement income. It further shows that the retirement bond can serve as a more appropriate goal-hedging instrument, including in balanced-fund and target-date-fund structures, and that the benefits remain robust across different interest-rate environments and cost-of-living adjustment assumptions.

Attendees can expect three practical benefits: first, a clearer framework for assessing decumulation risk in income terms rather than capital terms; second, insight into how goals-based fixed-income design can improve retirement-product efficiency; and third, implementable ideas for actuaries, consultants, asset managers and policymakers seeking to design or replicate retirement-bond solutions using existing liquid securities. The session therefore offers both theoretical originality and direct practical relevance for modern retirement product design.

RETHINKING THE GLIDE PATH: A CRITICAL EXAMINATION OF TARGET DATE FUNDS AND THE CASE FOR INCOME-ORIENTED ALTERNATIVES IN SOUTH AFRICAN RETIRMENT FUNDS

Dr Shaun Levitan, Teesta Bhattacharjee, Kholodelo Mohale, Dr David Rodda, Simthande Soke

Presentation with Paper | 60-Minute
Relevant Practice Area: Investments
Suggested Audience Knowledge Level: Foundational

Target date funds have become the dominant default investment strategy in DC systems globally. Their intuitive appeal — a single fund that automatically de-risks as the member approaches retirement — has driven widespread adoption. Yet the academic foundations underpinning this approach are fragile.

This paper presents a rigorous and structured critique of TDFs through three lenses: the state of the academic literature, current South African practice, and a set of practically implementable alternatives.

The paper begins with a comprehensive literature review, tracing the intellectual origins of the glide path concept. It shows that justification for TDFs rests primarily on behavioural finance arguments — specifically, the heuristic appeal of age-based de-risking and the reduction of sequence-of-returns risk near retirement.

Optimality of TDFs holds principally under an asset-accumulation objective, over specific time horizons, and in the absence of human capital and liability considerations. Once the full member lifecycle is modelled and the ultimate objective of sustainable retirement income rather than terminal wealth — the theoretical case for conventional glide paths weakens substantially.

The paper then turns to practice. It provides a detailed empirical description of the glide path structures, asset allocation ranges, and de-risking schedules employed by the largest umbrella fund TDF offerings in South Africa.

The paper concludes by demonstrating that relatively modest modifications to the standard TDF framework can yield materially better outcomes for members. These alternatives are evaluated against the conventional TDF on metrics relevant to retirement adequacy, and the analysis is calibrated to reflect realistic South African member profiles.

The paper speaks directly to the profession’s ongoing engagement with retirement fund adequacy, member outcomes, and the governance responsibilities of trustees and consultants who select default investment strategies. It bridges academic rigour with the practical realities of the South African retirement landscape, and invites actuaries to reconsider assumptions that have too often been imported wholesale from other markets.

Attendees will leave with a clear-eyed view of what the academic literature does and does not support regarding target date investing; a structured framework for evaluating the TDF offerings in umbrella funds against an income-adequacy standard rather than an asset-return benchmark; and a set of concrete design modifications they can carry back into trustee conversations, investment committee deliberations, and member communication strategies. The paper is intended to be genuinely provocative — not to dismiss TDFs, but to raise the bar for what “good enough” should mean in default fund design.

THE OVERLOOKED TAIL: ADVANCED AGE MORTALITY AND ITS IMPACT ON RESERVING

Tinotenda Madzingira

Presentation | 30-Minute
Relevant Practice Area: Life Insurance
Suggested Audience Knowledge Level: Foundational

In line with this year’s theme, “Broadening Boundaries,” this presentation examines an emerging and underexplored risk frontier within traditional life insurance: mortality experience at advanced ages in underwritten portfolios.

Credible experience is now beginning to emerge for policyholders aged 80 and above within risk-only products written in South Africa starting in the early 2000s. Early indications suggest that mortality at these advanced ages is materially heavier than previously assumed. This cohort is particularly significant: policies often carry substantial sums assured due to decades of benefit escalation, while lapse rates are negligible at these durations.

Despite the potential financial implications, assumption setting at advanced ages has historically received limited attention. Industry focus has tended to prioritise earlier durations and central age bands, where data is richer and pricing sensitivity is highest. As portfolios mature, however, this imbalance may expose insurers to unanticipated reserving strain.

This presentation will:

Analyse emerging mortality experience at advanced ages within underwritten portfolios
Benchmark observed outcomes against South African industry tables and past CSI publications
Compare findings with international mortality experience in comparable markets
Explore practical approaches to setting and validating mortality assumptions for older ages

The session aims to highlight a developing risk area, challenge conventional prioritisation in experience investigations, and provide practical guidance for actuaries managing maturing life portfolios.

HIERARCHICAL COMPARTMENTAL LOSS RESERVING MODEL EVALUATION

Hannes Van Rensburg, James Grove, Stephan Marais

Presentation with Paper | 60-Minute
Relevant Practice Area: Short-Term Insurance
Suggested Audience Knowledge Level: Intermediate

Loss reserving models are commonly evaluated and trained using isolated metrics that emphasize triangle‑level goodness of fit, with limited regard for the wider underlying claims development process. In practice this may include an A vs E metric (practical) or an error function (statistical / theoretical) evaluated within a single cohort projection method (say incurred development). In addition, processes and methods used in setting loss reserves are evaluated and reviewed mostly independent from one another, with limited reference between inter-related measures across paid and incurred amounts, reported and settled numbers or premium representations of the same portfolio.

We propose that claims process‑based evaluation metrics inspired by the relationships used by hierarchical compartmental reserving models may provide substantial insights into various assumptions underlying ultimate projections of claims and their appropriateness relative to alternative selection bases. Representing claims development and final settlement as a set of interdependent stochastic processes or compartments enables actuaries to evaluate and allow for their interdependent messages.

We also propose using these evaluation metrics to guide the selection of age-to-age factors within the chain ladder framework, with the objective of optimising overall model performance. By jointly optimising these interconnected metrics, the approach enables data-driven adjustments to paid and incurred development factors, incorporating information beyond the observed claim amounts alone.e practical guidance for actuaries managing maturing life portfolios.

FUNDING FOR RETIREMENT BENEFITS – COMPARING DEFINED BENEFIT, DEFINED CONTRIBUTION AND HYBIRD OPTIONS

Warren Matthysen, Frederik Van der Vyver, Stephen Walker

Presentation with Paper | 60-Minute
Relevant Practice Area: Retirement Matters
Suggested Audience Knowledge Level: Intermediate

This paper examines the evolution of retirement funding structures and explores how future scheme design can improve retirement outcomes in South Africa. It is motivated by the growing inadequacy and uncertainty of retirement outcomes in a system dominated by individual defined contribution (DC) arrangements, where members bear the full burden of investment, longevity, and annuitisation risk.

Historically, defined benefit (DB) schemes provided greater certainty about retirement income but exposed employers to uncertain and often escalating costs. In contrast, DC schemes offer predictability of contributions but shift the uncertainty of outcomes onto individuals. This trade-off has contributed to suboptimal retirement outcomes, highlighting the need for alternative models that better balance cost certainty with benefit adequacy.

The paper positions Collective Defined Contribution (CDC) schemes as a compelling hybrid solution to consider. CDC combines fixed contributions with collective investment and risk-sharing mechanisms, enabling members to target a more stable lifelong income without requiring employer guarantees. By pooling longevity and investment risks across members, CDC introduces an element of cross-subsidy that can smooth outcomes and improve overall retirement security.

Drawing on international experience from countries such as the Netherlands, Denmark, and the United Kingdom, the paper highlights key design features and governance principles that underpin successful collective systems. These include transparent benefit adjustment mechanisms, strong governance frameworks, and clear communication to manage expectations around variable outcomes.

The South African context is then evaluated, identifying structural challenges including fragmented funds,

limited risk pooling, inadequate annuitisation, and inconsistent retirement incomes. Against this backdrop, CDC is considered as a potential mechanism to improve both adequacy and stability of retirement outcomes, while remaining aligned with the regulatory and economic environment.

We adopt a combined quantitative and qualitative approach, testing hypothetical CDC designs against current DC outcomes to assess their impact on replacement ratios, income stability, and downside risk. This analysis is complemented by a conceptual design framework tailored to South Africa, incorporating principles of sustainability, intergenerational fairness, and transparency.

Ultimately, the paper considers whether incorporating collective risk-sharing mechanisms into retirement funding—particularly through CDC—offers a credible pathway to achieving more certain and sustainable retirement outcomes.

AUTONOMOUS AI-DRIVEN DISCOVERY OF ACTUARIAL RESERVING MODELS: LESSONS FROM SELF-IMPROVING CODE

Sikhulile Mdaka, Schoonraad Liebenberg

Presentation with Paper | 60-Minute
Relevant Practice Area: Wider Fields
Suggested Audience Knowledge Level: Intermediate

Loss reserving underpins every general insurer’s balance sheet, regulatory capital calculation, and IFRS 17 reporting, and the reserving opinion is a personally signed professional act. Yet the default approach on most books remains a volume-weighted chain ladder with relatively narrow method variation testing. The reserving literature offers a rich menu of alternatives – Bornhuetter-Ferguson, Cape Cod, Benktander, Mack chain ladder, alternative tail curves and credibility-weighted blends – and practicing actuaries have long recognised that thorough exploration of this method space produces more accurate and better-justified reserve estimates. The binding constraint has never been theoretical; it has been time and practicality.

We present an application of the “autoresearch” paradigm recently introduced by Karpathy (2026) to this problem. An AI coding agent is given a fixed evaluation metric, a compatible reserving library, and a written instruction file describing the experiment protocol. The agent iterates autonomously: it modifies the reserving code, runs the evaluation, keeps or discards the change on the metric, and commits each result to version control together with a plain- English journal entry that records the hypothesis, the specific change, the outcome, and the actuarial interpretation. Every decision is traceable and reviewable.

Initial runs produce material reductions (c.40%) in held-out prediction error relative to a volume-weighted chain ladder baseline, with improvements concentrated in per-line-of-business method specialisation, tail-curve selection, and targeted outlier treatment amongst others. The final paper will present the full empirical trajectory alongside a critical assessment of overfitting risk, the limits of single-diagonal evaluation, the governance and audit implications for appointed-actuary work. The work broadens a few boundaries. It broadens the scale of experimentation accessible to a single practitioner from a handful of manual variations to thousands of automatically evaluated configurations. And it redraws the boundary between actuary and technology: the actuary shifts from method implementer to research director, specifying the metric, the constraints, and the interpretation, while the agent performs the mechanical search.

Attendees will leave with a practical open‑source repository they can apply to their own triangles, and a clearl documented approach for embedding AI‑assisted experimentation into a governance‑approved reserving process as generative AI enters the workflow.

BAYESIAN COMBINED ACTUARIAL NEURAL NETWORKS FOR MOTOR INSURANCE CLAIM FREQUENCY MODELLING

Dr Wilson Mongwe, Katlego Thaba

Presentation | 60-Minute
Relevant Practice Area: AI, Machine Learning & Data Science
Suggested Audience Knowledge Level: Intermediate

Actuaries are well known for their models. However, it often proves a challenge for the actuary to neatly, succinctly and in a clearly articulable manner, both ascertain the drivers of uncertainty/variability between observed outcomes versus modelled results, and to communicate that uncertainty in a manner that is easy to follow, especially to non-actuarial stakeholders. Therefore, a framework that empowers the actuary with the ability to decompose uncertainty sources, especially between the underlying data and model selection as an example, becomes invaluable not only to the actuary, but also to key stakeholders to whom actuaries communicate model outcomes for decision making.

To bridge this gap and across the variety of models and model applications actuaries make use of, we propose a Bayesian extension of the Combined Actuarial Neural Network (CANN) for Poisson claim frequency modelling, by applying a post-hoc Laplace approximation with automatic relevance determination (ARD) priors. The Bayesian CANN retains the CANN’s GLM skip connection and adds three capabilities absent from standard actuarial neural networks:

1. principled architecture selection (which is achieved via the log marginal likelihood),
2. native feature importance (achieved through ARD precisions), and
3. policyholder-level uncertainty decomposition into aleatoric and epistemic components.

To illustrate the benefits that this approach offers, we benchmark against five models (Poisson GLM, FFNN, CANN, LocalGLMnet, XGBoost) on four motor datasets. From a results standpoint, Diebold–Mariano tests confirm the Bayesian CANN matches standard neural networks on Poisson deviance, while the post-hoc Laplace step adds an insignificant computational overhead. The log marginal likelihood achieves a strong Spearman negative correlation with out-of-sample deviance on the two high-signal datasets, validating it as a principled alternative to validation-based architecture search. Epistemic uncertainty concentrates in the highest-risk deciles across all datasets, identifying approximately 10% of policies as candidates for re-underwriting. We also document substantial disagreement between interpretability methods on low-signal portfolios, cautioning practitioners against relying on any single feature importance measure.

In conclusion, the Bayesian CANN offers actuaries uncertainty quantification, automatic model selection, and interpretability at negligible cost, which is a practical Bayesian upgrade to an architecture the profession already uses.

Keywords: Bayesian neural networks, Laplace approximation, ARD, claim frequency, CANN, uncertainty quantification

THE RELEVANCE OF THE GREEN TRANSITION ACTUARIES

Lance Moroney, James Williams

Presentation | 60-Minute
Relevant Practice Area: Enterprise and Financial Risk Management
Suggested Audience Knowledge Level: Intermediate

A discussion of the relevance and importance to actuarial work of the transition to an environmentally sustainable (“green”) economy.

The topic will leverage off a volume of material relevant to the topic, including but not limited to:
• Financial projection frameworks for insurance companies, banks and pensions scheme under alternate climate futures;
• IFoA series of papers on climate risk;
• Papers presented at previous ASSA conventions; and
• PA guidance on climate-related risk practices for insurers.

The presentation will be structured around the following discussion points:
• What is the green transition, and the importance of this for the sustainability of our planet;
• Political challenges to the transition, and why the green transition remains relevant;
• A discussion of the concept of net zero, and the context of South Africa;
• The benefits arising from the green transition to:
– Individuals;
– businesses; and
– the economy
• The role that the insurance industry can play to support the green transition;

Key discussion areas will be the role that actuaries can play to support the insurance companies that they work for, and the economy, on the journey towards an environmentally friendly and, resilient & sustainable economy:
• The need for further understanding of the context of the green transition by actuaries;
• Modelling skills needed by actuaries, and relevant technical frameworks; and
• The relevance of this topic to key insurance company processes such as product development, the ORSA and disclosure.

Questions will be posed to the audience to encourage engagement and highlight aspects of the topic that are important for actuarial work.

THE FUTURE OF MORTALITY: MEDICAL INNOVATION, LIFESTYLE AND TECHNOLOGY

Adam Musnitzky, Carla McConney, Moritz Drefs

Presentation | 30-Minute
Relevant Practice Area: Life Insurance
Suggested Audience Knowledge Level: Foundational/ Intermediate

Historically, life expectancy has continued to rise over time, typically driven by waves of innovations in medicine and healthcare, alongside lifestyles changes. These have swept through countries and societies at different times and rates, as these new developments take hold.

The past two decades have seen a slowing down in this trend with population-level life expectancy plateauing in many advanced countries with cardiovascular gains reducing, smoking gains running off and poor lifestyle factors – blurred with post-pandemic excess mortality. However, the recent rise of AI, the growth in new medical innovations, such as GLP1 drugs shows cause for optimism and may reverse the slowing life expectancy trend.

Despite such slowdown in cardiovascular gains, there is untapped potential for improvements across other causes of death such as the increasing use of personalized cancer treatments, better control of ageing diseases and AI driven medicine. These future impacts are key for mortality providers and actuarial assumption setting.

Forward-looking trends in medical science, lifestyle trends and technological developments, coupled with recent mortality data enables insurers to forecast the potential rate of mortality improvement for several future decades. Having a strong understanding of these potential future impacts, developing a range of future plausible scenarios and have future fit products is essential for mortality business.

Our presentation will explore:
• The key drivers of future long-term mortality across the globe, their historical trends and potential for future disease prevention
• Insights into where the next future impacts are expected to be generated
• Different potential scenarios of future mortality impact and what are the key underlying drivers of each scenario
• An understanding of the potential insurance portfolio and pricing impact
• How insurers should be thinking about product development and which causes of death should be key

LIVING LONGER, LIVING HEALTHIER – ACTUARIAL PERSPECTIVES IN AN INTER-CONNECTED WORLD

Claudia Pastellides, Katherine Tregoning, Motlodiwa Modise, Sayuri Naidoo, Lara Wayburne

Presentation with Paper | 60-Minute
Relevant Practice Area: Healthcare
Suggested Audience Knowledge Level: Intermediate

Improvements in longevity across insured populations are reshaping actuarial risk, healthcare costs and system sustainability. While life expectancy continues to increase, emerging evidence suggests that these additional years are often with prolonged morbidity and higher healthcare utilisation. This creates opportunity to reframe longevity as a multidimensional actuarial challenge spanning health, life insurance and retirement systems, positioning the transition from lifespan to healthspan at the centre of actuarial relevance in an interconnected world.

Drawing on recent demographic, mortality and healthcare utilisation experience, this paper highlights a growing divergence between gains in life expectancy and years lived in good health. Age-related increases in inflation-adjusted healthcare costs, driven primarily by cardiovascular disease, other chronic conditions and cancer, are intensifying intergenerational cross-subsidisation within healthcare funding models. At the same time, patterns of escalating healthcare utilisation in the period preceding death indicate increasing resource intensity, creating both financial pressures and opportunities for intervention and better care management.

The rising burden of mental health conditions among younger and working-age populations is shifting morbidity patterns to younger ages. This trend is associated with increased healthcare utilisation often leading to sustained or accelerated morbidity, and productivity loss. The downstream implications of this shift highlight the need to understand emerging morbidity dynamics and designing targeted interventions, preventative interventions to support long term system sustainability.

These trends highlight the tension between annual pricing cycles and sector-specific responses, calling for longer term actuarial approaches. In alignment with the Convention theme of Beyond Boundaries, this paper explores lifecycle-based health financing approaches that integrate earlier risk identification with long-term liability management. It considers the role of enhanced data integration, predictive analytics and AI-enabled risk stratification in identifying morbidity trajectories earlier, personalising risk assessment and enabling more proactive and preventative interventions.

By shifting longevity measurement from years lived to years lived in good health, this paper argues for an expanded actuarial role, one that extends beyond managing financial risk to actively shaping health outcomes, ensuring financial sustainability and supporting societal wellbeing across traditional practice boundaries.

EXPANDING THE ACTUARIAL FRONTIER IN A 2.5°C WORLD

Fagmeedah Petersen

Presentation | 30-Minute
Relevant Practice Area: Professional Matters
Suggested Audience Knowledge Level: Foundational

I have positioned this presentation on Climate Change as a Professional matters presentation because it will cover 5 practice areas
• Short Term insurance
• Life Insurance
• Healthcare
• Investments, including transition
• Pension Funds

I have non-executive director experience across all areas described above, and I will give practical advice on how to operationalise climate change risk awareness and talk about what needs to change now. I will also give an updated view on what the unfolding of climate change is likely to be now that the IPCC have demonstrated that 1.5’C increase will have been breached over a decade by 2028. I will also introduce the Southern African model developed by the CSIR.

I will provide a digestible reading list on the subject catering to the various areas.

A CEDANT, A BROKER AND A REINSURER WALK INTO A BAR

Maximilian Popescu, Francois Berry, Neethal Natha

Panel Discussion | 60-Minute
Relevant Practice Area: Short Term Insurance
Suggested Audience Knowledge Level: Foundational/ Intermediate/ Advanced

Who’s buying the round? The state of reinsurance in South Africa through three pairs of eyes.

Reinsurance is the largest single counterparty exposure on most South African short-term insurers’ balance sheets, yet placement decisions are often inherited, relationship-driven, and poorly interrogated.

The 2021 riots, the 2022 KZN floods and the global hard market that followed reset retentions, withdrew aggregate protection, and forced cedants to retain volatility they had priced away for a decade.

This panel puts a cedant, a broker and a reinsurer on stage to work through real placement tensions, capital substitution vs earnings protection, data quality vs capacity, loyalty vs cycle timing, and asks whether the South African market is structuring reinsurance for the risks it actually faces, particularly on the rise of secondary perils.

RISK MANAGEMENT AS DECISION SUPPORT IN INSURANCE

Darshan Purmessur, Nicolai Baron von Rummell

Presentation with Paper | 30-Minute
Relevant Practice Area: Enterprise and Financial Risk Management
Suggested Audience Knowledge Level: Intermediate

This proposal examines the risk management function in insurance when understood not only as a governance and compliance mechanism, but as a source of structured decision support. Using South Africa’s Prudential Standard GOI 3 as its regulatory context, the paper argues that a compliance-only conception of risk management is incomplete in an industry whose core activity is the pricing and aggregation of uncertainty. It develops a four-role taxonomy of the function — Architect, Controller, Integrator, and Enabler — in order to distinguish responsibilities that are primarily governance- and assurance-oriented from those that are more directly decision-facing and strategic.

Methodologically, the paper combines a structured reading of GOI 3 and related guidance with international risk-management frameworks, including COSO and ISO 31000, and with a decision-science perspective on how material decisions ought to be assessed under uncertainty. In addition, the project also draws on interviews with CROs from most major South African insurance groups and insurers, both to test some of the paper’s underlying assumptions and to bring in market evidence on how the function is currently implemented in practice. The analysis shows that South African regulation is relatively explicit on governance, risk appetite, monitoring, and reporting, but less specific on cross-silo aggregation, independent decision support, constructive challenge, and capital-efficient decision-making. It then argues that material decisions should be evaluated not only through expected outcomes or business-case projections, but also through explicit consideration of downside tails, dependence with existing exposures, and balance-sheet capacity.

The link to the Convention theme, Broadening Boundaries, is direct. The paper argues that the boundaries of the risk management function in South African insurance should not be drawn too narrowly around compliance and assurance alone. Broadening those boundaries means extending the function more deliberately into quantitative analysis, decision-relevant assessment, and strategic support, while retaining its

independence and without displacing management’s ownership of decisions. In that sense, the proposal is about broadening the practical and conceptual boundaries of risk management itself: from a function that mainly checks and reports, to one that also helps organisations make better decisions under uncertainty.

ONE-SHOT INDIVIDUAL CLAIMS RESERVING

Ron Richman

Presentation with Paper | 60-Minute
Relevant Practice Area: Short Term Insurance
Suggested Audience Knowledge Level: Intermediate

The chain-ladder (CL) method has dominated claims reserving for decades, yet individual claims reserving, which exploits claim-level features to improve reserve estimates, has not become established in practice. We attribute this to the absence of a methodology that is simultaneously practical, robust, and naturally connected to existing actuarial workflows

We present a new approach that restructures the CL algorithm to enable individual claims reserving. The key insight is reformulating CL factor estimation into a recursive one-shot projection-to-ultimate (PtU) framework: rather than rolling forward cumulative payments one period at a time, we estimate grossing-up factors that project each claim’s latest observation directly to ultimate. This representation is mathematically identical to standard CL on aggregate data, but creates a natural pathway for incorporating claim-level features through regression modelling.

The approach handles arbitrary dynamic stochastic covariates, including claim status, claims incurred and case reserves, without requiring their extrapolation, removing the principal barrier in existing micro-reserving methods. It separates RBNS and IBNR claims through consistent cohort construction, providing a novel decomposition of total CL reserves. We present results on two real datasets (accident and liability insurance) benchmarked against known lower triangles.

This work embodies Broadening Boundaries by demonstrating how the most familiar reserving technique can be extended to individual claims using modern machine learning, without abandoning its core principles. CL predictions serve as guardrails for ML predictions, making the approach accessible without deep learning expertise and satisfying regulatory expectations for explainability.

Practical outcomes: attendees will gain (a) a concrete, implementable algorithm for individual RBNS reserving applicable with any regression method from GLMs to transformers; (b) the insight that linear regression can be surprisingly effective, making micro-reserving accessible without ML infrastructure; (c) a method for decomposing CL reserves into RBNS and IBNR components; and (d) a bootstrap procedure for individual claim uncertainty quantification.

WHERE STP ENDS AND RISK BEGINS: LINKING AUTOMATION TO EMERGING EXPERIENCE

Shevani Rijhumal, Dr Matthew Parker

Presentation | 60-Minute
Relevant Practice Area: Life Insurance
Suggested Audience Knowledge Level: Intermediate/ Advanced

Straight through processing (STP) has become a powerful enabler of speed and competitiveness, allowing insurers to deliver fast decisions and attractive pricing in increasingly automated markets. From a business perspective, strong STP metrics are often interpreted as evidence of underwriting maturity and a sound basis for sustainable pricing.

In practice, however, the commercial trade offs are more nuanced. In many automated underwriting environments, the initial expectation is that outcomes will broadly align with fully underwritten pricing. Experience has shown that this assumption does not always hold. Even small and systematic simplifications in underwriting rules can allow a limited number of materially higher risk lives into the portfolio, with consequences that only become evident as experience develops over time.

This presentation explores why operational efficiency metrics alone provide an incomplete view of risk selection quality. A key challenge for the business is that the cost of these trade offs is not visible at point of sale, nor through early STP or acceptance rate metrics, but only later through emerging claims patterns and pricing experience studies.

Using practical examples from automated underwriting and rules based decision environments, the session illustrates how underwriting design choices, pricing assumptions, and governance interact to shape long term outcomes. The discussion focuses on how businesses can strike a more informed balance between speed, price competitiveness, and experience sustainability, supported by stronger feedback loops between underwriting, pricing, and experience analysis.

The presentation concludes by focusing on how businesses can strike a more informed balance between speed, price competitiveness, and experience sustainability. This involves treating underwriting design, pricing assumptions, and experience monitoring as connected decisions rather than separate functions, explicitly pricing for the level of underwriting applied, recognising the limits of operational efficiency metrics, and using emerging experience as an ongoing governance signal rather than a retrospective explanation. Stronger feedback loops between underwriting, pricing, and experience analysis allow businesses to remain competitive while maintaining a clearer line of sight on longer‑term sustainability.

BOUNDARYLOGIC – EXPLAINING WHY BLACK-BOX MODELS MAKE A PREDICTION AND HOW TO CHANGE THE OUTCOME

Dr Adriaan Rowan

Presentation| 60-Minute
Relevant Practice Area: Banking
Suggested Audience Knowledge Level: Intermediate

The increasing use of machine learning models in actuarial science and financial services raises critical questions about trust, transparency, and interpretability. While such models often achieve high predictive accuracy, they are frequently treated as “black boxes”, making it difficult for actuaries, regulators, and clients to understand why a specific decision was made. This lack of interpretability poses challenges in areas such as credit scoring and insurance underwriting where legislation require models to be explainable.

This study develops a novel framework for explainable artificial intelligence (XAI) by extending the use of biplots—traditionally a multivariate data visualisation tool—into the domain of counterfactual explanations. Counterfactuals answer practical “what-if” questions, such as “What would need to change for this loan application to be approved?” or “Which factors most influenced this mortality risk classification?”. By projecting high-dimensional data into an interpretable two-dimensional biplot, this approach identifies local decision boundaries and visualises the shortest actionable path from a data point to its counterfactual.

The method provides both local and global interpretations. Locally, it explains individual predictions by identifying which features must change—and by how much—to alter the outcome, while Shapley values are used to quantify the contribution of each variable to the prediction outcome. Globally, it creates feature importance rankings by aggregating distances to the decision boundary across the dataset. The approach also introduces a robustness metric, measuring the stability of predictions against small perturbations in input features, offering actuaries an alternative to predictive accuracy when comparing competing models.

Lastly the method generates surrogate models that are very interpretable approximations of the original model while retaining high fidelity The framework is model-agnostic, applying equally to black-box models such as gradient boosting machines and transparent models such as generalised additive models.

Attendees of this session will gain:
• An understanding of how biplots can be extended beyond data visualisation to identify local decision boundaries and generate counterfactual examples
• Practical tools for local interpretation, including how to use counterfactuals together with Shapley values to explain why an individual received a specific prediction and what actionable changes could alter the outcome.
• A framework for global interpretation, showing how aggregated counterfactuals values can identify the most influential features, evaluate model robustness, and construct interpretable and reliable surrogate models.
• Applications in actuarial practice, with case studies in credit risk and insurance prediction models, illustrating how these techniques help detect overfitting, validate model behaviour, and communicate results to stakeholders.

ANALYSING HOSPITAL-ACQUIRED INFECTIONS IN NEONATAL CARE FOR A SAMPLE OF MEDICAL SCHEME LIVES IN SOUTH AFRICA

Chanie Suttner, Stefanie Babiolakis, Michael Cohen, Jackie Duvenage, Melina Manferdini

Presentation with Paper | 60-Minute
Relevant Practice Area: Healthcare
Suggested Audience Knowledge Level: Foundational/ Intermediate

This presentation proposes a novel, data-driven approach to identifying and quantifying hospital-acquired infections (HAIs) in neonatal admissions within South Africa’s private healthcare sector, using Discovery Health administrative and clinical data. It aligns with the Convention theme of leveraging data and analytics to improve healthcare outcomes and directly supports the United Nation’s Sustainable Development Goal (SDG) 3: Good Health and Well-being, particularly target 3.2 on reducing neonatal mortality.

HAIs, defined as infections occurring 48 hours or more after admission, are a major contributor to avoidable neonatal morbidity, mortality, and cost. Infection, particularly sepsis, remains a leading cause of neonatal death globally and a key barrier to achieving SDG 3 targets. Despite this, there is no standardised surveillance framework in South Africa’s private sector, limiting transparency, accountability, and progress toward measurable quality improvement.

This study addresses this gap by developing a pragmatic, multi-source identification algorithm that integrates claims data and pathology results data proxy to detect likely HAIs. Focusing on neonatal intensive care unit (NICU) admissions, the analysis quantifies both incidence and financial impact, including prolonged length of stay and downstream healthcare utilisation.

Importantly, the study extends beyond measurement to assess variation across hospitals, adjusting for case mix to identify differences in infection rates that may reflect variation in infection prevention and control practices or broader system-level factors. This supports SDG-aligned health system strengthening by enabling benchmarking, improving accountability, and informing targeted interventions

The presentation will highlight key methodological considerations, including trade-offs between sensitivity and specificity when using administrative data, and the role of actuarial judgement in constructing robust, scalable surveillance metrics in data-constrained environments.

The presentation aims to provide:

a reproducible framework for identifying HAIs using routinely collected data,
quantified insights into the clinical and financial burden of neonatal HAIs,
approaches to benchmarking variation across providers.

The findings have direct implications for value-based healthcare and provider contracting contributing to South Africa’s progress toward SDG 3 by reducing preventable neonatal morbidity and mortality.

QUANTIFYING SEVERITY: A METHODOLOGICAL APPROACH FOR SHORT-TERM INSURANCE

Dreyer Swart, Kovlin Perumal

Presentation with Paper | 60-Minute
Relevant Practice Area: Short Term Insurance
Suggested Audience Knowledge Level: Advanced

Pricing actuaries in general insurance are typically concerned with two main objectives when developing rating models. First, by virtue of the pooling of risks, risk premiums should on average be sufficient to cover all claims-related outflow. Second, rating models should differentiate between policies representing different levels of risk. The latter objective is especially relevant in charging fair and competitive premiums and reducing the risk of anti-selection.

In typical insurance modelling frameworks, risks are decomposed into frequency and severity components, with the adequacy and risk differentiation of the overall risk premium depending on the performance of each component. Severity models can offer substantial scope for risk differentiation, since claims amounts may vary by several orders of magnitude; however, severity models often suffer from sparser data and a more complex pay-off structure compared to frequency models, which can lead to challenges in separating signal from noise. This can ultimately lead to models that fit the data poorly or fail to capture the sought-after risk differentiation available in the data.

This paper aims to address this issue by moving beyond the common assumption of a Gamma error distribution to test a wider set of severity assumptions and broader statistical tools, drawing evidence from both real-world and controlled synthetic portfolios, to develop a framework for effective severity modelling.

Using a case study approach, the analysis focuses on, but is not limited to, two core modelling philosophies: modelling severity through average cost per claim (ACPC) and damage ratio. For positive severity and ACPC, candidate structures include Gamma and Inverse Gaussian models, while Tweedie models are considered for aggregate loss cost and pure premium formulations. For damage ratio, the study assesses alternatives such as Beta, quasi-binomial, Gamma and Gaussian formulations. These methods are tested consistently across the datasets to determine whether any approach performs robustly across portfolios, or whether model choice is primarily driven by the underlying characteristics of the dataset.

The work also examines practical complications that materially affect real-world pricing: the treatment of large losses, the interaction between claim count and severity, and the role of alternative exposure and normalisation measures, including policy exposure, claim frequency and sum insured. Model performance is assessed using a reproducible validation framework focused on predictive accuracy, a calibration, stability and practical interpretability for pricing applications.

BUILDING YOUR FIRST AI AGENT: A HANDS-ON INTRODUCTION FOR ACTUARIES

Nicholas Thomadakis

Workshop
Relevant Practice Area: Wider Fields
Suggested Audience Knowledge Level: Foundational

Agentic AI represents a transformative shift for the actuarial profession and the financial sector, one that will redefine roles and challenge long-held ways of working. Unlike traditional AI tools, agentic systems can reason, plan, and autonomously execute complex, multi-step tasks. The profession stands at a pivotal inflection point as these systems move towards productionised deployment in South Africa’s financial sector.

This workshop translates that context into practice. Rather than discussing agentic AI in the abstract, attendees will build a working AI agent from scratch, gaining direct, hands-on experience with the technology that is already reshaping actuarial work.

Why Now?

South African insurers are already piloting and deploying agentic AI in production, while regulators are actively engaging with its governance and ethical implications.

Workshop Focus

Drawing on direct experience implementing agentic AI, this workshop will guide attendees through:
• What an AI agent is and how it works
• The key building blocks: reasoning, planning, tool use, and memory
• A live, guided build of a simple but functional AI agent, no prior coding experience required
• How to connect an agent to real data and tasks relevant to actuarial work
• Where human oversight fits in, and why it matters in regulated environments

Practical Outcomes
Attendees will leave having built and run their own AI agent, with a working mental model of how agentic systems are constructed and deployed.

Broadening Boundaries
This workshop embodies the Convention theme directly: it places new tools in actuaries’ hands and expands what they are capable of engaging with technically. By demystifying agentic AI through practice, it lowers the barrier to participation and grows the profession’s collective understanding of what an AI-enabled future could look like.

DESIGNING PARAMETRIC INSURANCE IN SOUTH AFRICA USING EXTREME VALUE THEORY

Chris Van Der Merwe

Presentation | 60-Minute
Relevant Practice Area: Short Term Insurance
Suggested Audience Knowledge Level: Intermediate

This session will consider global parametric insurance experience in recent years, highlighting both the barriers and opportunities for take-up in South Africa. The session will then go on to explore the potential for using Extreme Value Theory (EVT) to quantify extreme rainfall and flash-flood risk in South Africa, to design a local, decision-ready framework for insurers, reinsurers, and public-sector stakeholders.

South Africa has experienced repeated high-impact flood events in recent years, and stakeholders increasingly need defensible estimates of rare rainfall intensities (e.g. 1-in-50 to 1-in-200 year 24‑hour totals) to support pricing, capital, and resilience planning. The presentation will show how EVT isolates and models the tail of observed precipitation distributions to estimate return levels and uncertainty, and how these outputs can be operationalised.

Methodologically and what will be covered in my presentation: I will use daily rainfall observations from South African weather stations (e.g., SAWS and other vetted sources), apply data-quality filters (minimum record length, extent of missing data limits, and active stations), and perform homogenisation to remove non-climatic artefacts (station moves, instrumentation changes) using region-based reference series and multiple homogeneity tests. I will then fit either a Block Maxima Method or Peaks-Over-Threshold (POT) model, discussing how the selected method was considered to be the optimal choice. I will also demonstrate an event backtest: calibrating the model excluding recent flood years and assessing how “new” extremes map to implied return periods and model stability.

Practical outcomes for attendees:
• An understanding of global development in parametric insurance and application in the South African context.
• A replicable workflow to produce station-level return levels (e.g., 100‑year 24‑hour rainfall) with confidence intervals and fit diagnostics.
• A risk-ranking view of South African hotspots (e.g., coastal vs escarpment vs interior regimes) and where “beyond-historical” extremes are statistically plausible.
• Guidance on using EVT outputs to structure and price parametric flood/rainfall covers (trigger setting, payout curves, basis-risk considerations).
• Examples of how EVT can support SAM/Solvency capital validation and scenario design (including ORSA and climate pathway stress testing).
• A blueprint for public-sector applications (drainage design standards, cost–benefit of mitigation, and high-risk zoning).

BEYOND MEDICAL SCHEMES: WHY SOUTH AFRICAN HEALTHCARE COVER MAY NEED TO EVOLVE INTO INTERGRATED, LAYERED PROTECTION

Stephanie van Niekerk, Homaam Arend, Elizabeth Greyling, Allan Williams

Presentation with Paper | 60-Minute
Relevant Practice Area: Healthcare
Suggested Audience Knowledge Level: Intermediate

Private healthcare funding in South Africa has traditionally been anchored by medical schemes as the primary mechanism for accessing private care. In recent years, however, the environment has become increasingly characterised by rising healthcare costs, affordability pressures, benefit design constraints, and growing out of pocket exposure for members. These trends have prompted renewed consideration of the extent to which medical schemes – operating largely as standalone funding vehicles – are able to respond effectively to the evolving demands of a complex and cost intensive healthcare system.

While medical schemes remain central to private healthcare funding, their ability to address system wide cost and access pressures in isolation is inherently constrained. Efforts to maintain sustainability have led to increased reliance on managed care interventions and benefit restructuring, which, while supporting solvency, may also widen funding gaps and increase member cost participation. In parallel, Gap Cover and Primary Health Insurance (PHI) products have expanded to address specific unmet healthcare needs: with Gap Cover offsetting shortfalls in scheme reimbursement and PHI facilitating access to targeted areas of care, particularly primary healthcare services. However, when operating independently, these products tend to address discrete risks while leaving adjacent vulnerabilities unresolved.

Against this backdrop, there is growing interest in the potential role of more integrated or bundled healthcare solutions that align medical schemes with complementary products such as Gap cover and PHI. Such approaches may offer a pathway to enhanced financial protection, improved affordability and more coherent cover across the continuum of care. This paper adopts an exploratory perspective to examine the viability and implications of these bundled models.

At the same time, any movement toward greater integration introduces a range of complexities that warrant careful examination. Structural and regulatory distinctions between product types, differences in risk pooling and pricing mechanisms, potential misalignment of incentives, and increased administrative complexity may all present material barriers. In addition, questions relating to governance, distribution dynamics, and the potential for unintended consequences within an already fragmented healthcare system remain unresolved.

Accordingly, this research seeks to explore both the opportunities and constraints associated with bundled healthcare solutions. In doing so, it aims to contribute to a more nuanced understanding of whether, and under what conditions, such models could support a more holistic and sustainable approach to healthcare financing in South Africa.

THE NEXT GENERATION OF CUSTOMERS & IMPACT OF AI

Isa Van Zyl, Paul Calvey, Sandra Villars

Presentation | 60-Minute
Relevant Practice Area: Banking
Suggested Audience Knowledge Level: Foundational

Financial services is entering a structural shift driven by changing customer expectations, AI, and new digital ecosystems. Traditional definitions of “customer”, “distribution” and “risk” are being redefined, both by the way banks operate internally and by how customers engage with financial institutions externally. For actuarial and risk professionals, this creates a clear imperative to broaden boundaries and remain relevant in a rapidly evolving landscape.

Our insight-led session will explore three questions at the heart of this shift:
Who is the customer of the future?
How is AI changing the banking landscape?
How must the risk function evolve to stay effective and add value?

First, we will examine the customer of the future, drawing on insights from South African youth to test some common assumptions. Gen Z and Millennials are often described as digital-only and AI-first, but our research suggests a more nuanced picture of how younger customers discover, choose and engage with banks. We will explore what they actually value from financial institutions, and what this means for customer acquisition, loyalty, product design and distribution.

Second, we will consider how AI is reshaping banking. This is not only about customer-facing tools and recommendations, but also about how banks themselves are changing internally. AI is already influencing decision-making, productivity, service models and operating structures, while also opening the door to more personalised and timely customer experiences. At the same time, the rise of AI-enabled recommendations and AI-assisted decisions may reshape how customers choose where to bank, what products to buy, and how they transact.

Finally, we will explore the risk function of the future. As technology, customer behaviour and banking models evolve, risk functions will need to move from a traditional control role toward a more predictive, strategic and AI-enabled model. This raises important questions around controls, governance, talent and the role of risk in supporting growth.

The session will show why the convergence of customer evolution, AI and risk transformation is redefining the future of banking, and why actuarial professionals have an important role to play in shaping it.

WHO MARKS THE MARKERS? DEVELOPING AUTOMATED MARKING

Burt Verster, Matthew Smith

Presentation| 60-Minute
Relevant Practice Area: Education
Suggested Audience Knowledge Level: Foundational

This presentation tells the story of the journey from the first Mark prototype, presented at the 2017 Convention, to the current Auto Mark beta.

We will show how the idea evolved from a text analytics engine designed to grade F102 scripts and provide standardised feedback into a broader actuarial marking engine built to deliver faster, more consistent and more detailed feedback.

The journey was not a straight line: after building the first engine, our focus shifted to evaluating the accuracy and consistency of human markers, understanding where marking judgement differs, and building a human-marker dataset that could support effective evaluation.

The talk will then explain how those insights shaped the new Auto Mark engine.

We will cover what makes automated actuarial marking difficult, why we engineered a bespoke marking harness, and how the human-marker dataset became a critical benchmark for testing performance.

We will cover the key design decisions behind the engine, including quality-assurance data collection, calibration against Fellow-marked scripts, and the structure of the current system.

We will share where the engine performs strongly, where complex judgement remains challenging, and how we approach control, governance and trust when using automation in a professional setting.

Finally, we will share a sneak-peak into what we are building at ThinkActuary, and how we see the future of actuarial education developing.

This aligns with the Convention theme, broadening boundaries, by showing how actuarial expertise can be extended through purpose-built technology, assessment and quality assurance, while still recognising the limits of machine judgement.

Attendees will leave with practical lessons on building, evaluating and implementing human-in-the-loop actuarial tools, including the safeguards needed for specialist professional workflows. More broadly, the session will show how reliable AI systems can be designed to outperform generic ChatGPT use, and will give actuaries ideas for applying similar approaches in their own day-to-day work.

QUALIFIED ON PAPER: IMPOSTER PHENOMENON IN SOUTH AFRICA ACTUARIAL CAREERS

Nicolai Von Rummell, Caesar Balona, Margie Conradie

Presentation with Paper | 60-Minute
Relevant Practice Area: Professional Matters
Suggested Audience Knowledge Level: Foundational

‘The impostor phenomenon—persistent self-doubt about one’s competence and fear of being exposed as less capable than others believe, despite strong performance—is plausibly salient in the South African actuarial qualification pathway, where high-stakes examinations are completed alongside full-time employment and professional status can raise external expectations faster than workplace confidence develops. While impostor feelings have been studied in many student and professional populations, there is limited evidence on their prevalence, severity, and workplace correlates in actuarial early careers in South Africa.

This study investigates the impostor phenomenon among working actuarial students and recently qualified actuaries who are members of the Actuarial Society of South Africa (ASSA), excluding those currently enrolled at university. An online questionnaire distributed through ASSA’s membership channels measures impostor phenomenon using the Clance Impostor Phenomenon Scale and perceived team psychological safety using Edmondson’s seven-item scale, alongside background and work-context items aligned where possible with existing ASSA instruments.

The analysis addresses four questions: the distribution and severity of impostor scores; differences across qualification stage, demographic groups, and family educational background; levels of perceived psychological safety and their variation across the same groups; and the adjusted association between psychological safety and impostor severity, controlling for pre-specified background factors. Group comparisons report effect sizes and uncertainty intervals alongside statistical tests, and the associational analysis uses linear regression with pre-specified sensitivity checks for the progression measure and for plausible reverse-direction interpretations.

Because the design is cross-sectional, results are interpreted as adjusted associations rather than causal effects.

The study offers a baseline description of impostor feelings in South African actuarial early careers and tests whether variation in the team learning climate is associated with impostor severity, to support evidence-informed discussion about early-career development and workplace learning conditions.’

MY LIVED EXPERIENCE TO A THRIVING WORLD: REIMAGINING MENTAL WELLNESS

Nick Wells

Presentation | 60-Minute
Relevant Practice Area: Life Insurance
Suggested Audience Knowledge Level: Foundational

I’m Nick Wells, a qualified actuary, husband and father of three. I was and am a high-performing business development actuary at Swiss Re. I have a lot of experience navigating high-levels of expectations and stress and succeeding. I also have a lot experience of being able to achieve very little at all. For a long period I was severely anxious, hopeless, depressed, hospitalised, disabled, written off.

I became a claimant on the product I loved the most as an actuary – income protection. I always enjoyed the complexity of modelling it, designing it, selling it. And then I became a model point. End-to-end I was a claimant for one month short of three years, and had reached a point where my claim was in the pipeline to be approved for long-term payment: I was not expected to recover.

Miraculously I did recover, and am better than ever before. I have gained so much through what I have learnt. In true actuarial style, I tracked significant amounts of data on my journey and I want to use my experience to positively impact the industry.

In 2025 the Swiss Re team presented on ‘Innovating Mental Wellness Insurance: Balancing Risk, Support, and Product Design’. At a high-level we shared:

-Mental health is seen as the biggest health problem globally. There has been much more awareness and interest following the lockdowns and isolation associated with the Covid-19 pandemic.

-Consumers see their mental wellbeing as a priority and something insurers can help with. We presented some product ideas and shared an AI solution in this space.

In 2026, we will expand on this topic and present a personal mental health story coupled with further data on mental health trends, culminating in concrete proposals on how the insurance industry and actuarial profession could redesign products and processes to better serve the needs of customers – improving protection but also helping people to thrive.

TOWARDS A REASONABLE, AFFORDABLE, EQUITABLE, AND SUSTAINABLE ROAD ACCIDENT COMPENSATION SYSTEM

Dr Gregory Whittaker, Dr David Rodda, George Schwalb

Presentation with Paper | 60-Minute
Relevant Practice Area: Damages
Suggested Audience Knowledge Level: Intermediate

This paper examines the design of road accident compensation systems in South Africa, focusing on whether the current Road Accident Fund (RAF) meets the criteria of being reasonable, affordable, equitable, and sustainable, and how it compares to two alternatives: the proposed no-fault Road Accident Benefit Scheme (RABS) and compulsory third-party insurance (CTPI). The analysis is multidisciplinary, combining actuarial modelling with legal, economic, and institutional perspectives, and is grounded in both local evidence and an extensive international comparison of compensation systems.

The topic aligns strongly with the Convention theme of “Broadening Boundaries” by extending actuarial thinking beyond its traditional insurance applications to public policy and social protection. The paper demonstrates how actuarial methods can be used to evaluate entire systems rather than individual products, incorporating considerations such as constitutional rights, healthcare provision, governance failures, and fiscal sustainability. It also engages with global issues by benchmarking South Africa’s system against international models, highlighting common structural features and policy trade-offs across jurisdictions.

From a practical perspective, attendees will benefit in several ways. Firstly, the presentation provides a structured framework for evaluating compensation systems using the four criteria of reasonableness, affordability, equity, and sustainability, which can be applied to other social insurance contexts. Secondly, it offers insights into the strengths and weaknesses of different funding models (including fuel levies, insurance-based approaches, and hybrids), with implications for practitioners in both the public and private sectors.

Thirdly, the paper highlights real-world governance and operational challenges within large compensation schemes, offering lessons on the importance of institutional design and oversight.

Attendees will gain exposure to international best practices and comparative models, enabling them to contextualise local challenges within a global landscape. The session is designed to encourage critical discussion on policy trade-offs and the role of actuarial professionals in shaping sustainable, equitable systems, equipping participants with both conceptual tools and practical insights applicable across a wide range of actuarial and public policy domains