Pricing Optimisation in Insurance: A Malaysian Guide
Pricing optimisation in insurance is the practice of adjusting premiums, after the actuarial technical price has been set, so that they meet a business objective such as profit, retention or growth. For a Malaysian insurer or takaful operator (ITO), the question is not whether the models can be built. It is whether they can be built so that every price movement is explainable to Bank Negara Malaysia (BNM), to the board and to the policyholder.
This guide explains how optimisation differs from risk-based pricing, which data and constraints matter, what the Malaysian rules require, and how to govern a model in production.
TL;DR
Risk-based pricing estimates the expected cost of loss. Price optimisation adds a demand layer on top, so that prices serve an objective such as retention or margin.
In Malaysia, BNM’s fair treatment policy prohibits unfair discriminatory practices, including charging different premiums to people of the same class and essentially the same hazard.
Constraints are not an afterthought. Smoothing, risk pooling, Shariah requirements and caps on increases belong inside the model from the first design session.
Governance decides whether a model survives audit: board-approved policies, monitoring metrics and explicit triggers for remedial action.
The strongest programmes start with data readiness and a measured pilot, not with an algorithm.
Table of Contents
What price optimisation means in insurance
Traditional insurance pricing answers one question: what is the expected cost of the risk being transferred? Actuaries estimate claim frequency and severity, add expenses and a profit margin, and arrive at a technical price.
Price optimisation asks a second question: given that technical price, what should the insurer actually charge each customer segment to reach a strategic goal? According to the US Casualty Actuarial and Statistical Task Force white paper, it supplements traditional actuarial models with quantitative customer demand modelling to pursue objectives such as profit or retention, whereas risk-based pricing looks only at the expected cost of future loss and expenses.
That difference is the whole debate. A risk-based premium can be justified by claims data. An optimised premium also reflects how likely a customer is to accept, renew or shop around. The first is easy to defend. The second needs careful design, because it can drift away from cost and risk.
How insurance pricing works, and where optimisation fits
Pricing is a sequence. The Casualty Actuarial Society overview of price optimization describes actuarial models determining the cost of risk transfer first, with optimisation acting as a later, often data-driven overlay that adjusts those premiums to meet organisational business metrics.
In practice, the sequence looks like this:
Experience analysis: claims, exposure and expense data are cleaned and reconciled.
Technical pricing: frequency and severity models produce the expected cost per risk.
Competitive and demand analysis: the insurer estimates how conversion and renewal respond to price.
Optimisation: a model searches for the price set that best meets the objective inside defined limits.
Filing, implementation and monitoring: prices go live, and actual behaviour is compared with the model’s expectation.
Most of the risk sits between steps 3 and 4. If demand sensitivity is allowed to override cost without limits, the result is the pattern regulators worry about. If it is bounded by cost, fairness and regulatory constraints, the same technique becomes a way to allocate adjustments sensibly.
Data and methods insurers use
The technical price is built from the insurer’s own history. In Malaysia’s medical and health insurance and takaful (MHIT) market, ITOs use past claim experience, including frequency and severity, together with expenses and expected profit margins, to set expected claim costs.
Industry data is becoming part of the picture. From 1 January 2025, ITOs are required to submit MHIT claims data to a central industry platform, enabling industry-wide analysis of medical inflation, claims ratios and related metrics. For a modelling team, that means a shared benchmark against which an individual insurer’s assumptions can be tested.
On top of that foundation, optimisation typically needs four further inputs:
Policy-level history, including renewals, lapses, endorsements and channel.
Price change history, so that elasticity can be estimated from what customers actually did and not from assumption.
Competitive price information, where it can be sourced lawfully and reliably.
Business constraints, expressed as numbers the model can respect.
On methods, gradient boosting and generalised linear models remain common for the technical price, while demand models are often logistic or machine-learning classifiers. Recent academic work on pricing optimisation with off-policy evaluation is still research, so treat it as direction of travel and not as a production standard. We link to what we have verified and leave out the rest.
If you are building this capability, our article on insurance predictive analytics covers the underlying modelling foundations in more detail.
Objectives and constraints a model must carry
An optimisation model is only as good as the limits written into it. In Malaysia, those limits are partly set by the regulator.
BNM’s MHIT policy document states that repricing must balance equity among policyholder groups and the benefits of risk pooling, consistent with sound actuarial principles and regulatory and Shariah requirements. That is a design brief. It tells a modeller that individual-level maximisation is not the goal, and that the pool matters.
Three further constraints appear in BNM’s supervisory commentary on repricing. First, smoothing is generally applied as an overlay to moderate significant impacts from technical adjustments, especially for higher-risk groups, and to ensure price relativities make sense. Second, ITOs must demonstrate that adjustments are limited to what is necessary to maintain the target ratio of price relative to claims. Third, there are interim limits on the size of increases.
For repricing performed up to the end of 2026, ITOs must spread premium increases over at least three years, so that at least 80% of policyholders are expected to see yearly adjustments from medical claims inflation of less than 10%. Because this measure is tied to a period, confirm the position that applies to your next repricing cycle on BNM’s website before you set model parameters.
A worked illustration
Consider a hypothetical MHIT portfolio with three age bands. The technical model says the oldest band needs a sharp increase. A purely demand-led model might say that band is the least price-sensitive and can bear the largest rise, and that a younger, more sensitive band should be held flat to protect volume.
That output would breach the spirit of the Malaysian approach twice: it would load the oldest group beyond what cost supports, and it would ignore the instruction to moderate impact on higher-risk groups. A compliant model reverses the logic. Cost sets the floor and the relativities, smoothing sets the pace, and demand analysis is used only inside those limits, for example to choose the timing and communication of the change.
The illustration is invented to show the logic, not drawn from any insurer’s data.
Benefits and risks of price optimisation
When it is bounded properly, optimisation helps an insurer hold margin without lurching between large corrections, retain good customers by identifying who is most likely to leave and why, and make the effect of a rate change visible before it is filed.
The risk is a loyalty penalty. The Washington State insurance commissioner’s Technical Assistance Advisory 2015-01 notes that price optimisation may result in different premium increases for policyholders with the same loss history and risk profile. The same advisory treats premiums, rates or rating factors unrelated to cost and risk as unfairly discriminatory under that state’s law.
That is a United States position, and it is not Malaysian law. But it shows where scrutiny falls, and a Malaysian model built to survive review should be able to answer the same question: if two customers carry the same risk, can you explain why they pay different amounts?
What Malaysian rules require
Malaysia does not publish a rule headed “price optimisation”. The relevant duties come from fair treatment and product-specific policy documents.
BNM’s Fair Treatment of Financial Consumers policy requires financial service providers to deliver fair treatment and to ensure consumers are not subject to unfair discriminatory practices. It lists among those practices charging different premiums to individuals of the same class and essentially the same hazard. Some provisions took effect on 1 April 2025.
For MHIT, the requirement is that pricing and management comply with generally accepted actuarial principles and fair treatment considerations. There are also consumer-facing duties:
Enhanced product disclosure sheets for MHIT products must include three-year industry average medical inflation rates and claims ratios, capped at two and a half A4 pages.
ITOs must obtain minimum information to assess a consumer’s needs before advising or recommending, and direct channels must provide self-assessment tools such as a financial budget calculator.
The practical reading for model teams is straightforward. Any variable that moves price should be tied to cost, risk or a documented, defensible pool-level reason. A variable that exists only because it predicts inertia is the one to challenge first. Your compliance and legal advisers, and your appointed actuary, should confirm how these policies apply to your specific lines of business.
Governing, testing and monitoring the model
A model that cannot be audited will not stay in production. BNM’s MHIT policy requires board-approved, documented repricing policies and procedures, experience-monitoring metrics and explicit triggers for remedial action. Under the fair treatment policy, the board is also responsible for oversight of fair treatment requirements, including in repricing.
Translated into engineering terms, a governable pricing model needs:
A versioned record of every input, assumption and parameter behind each price set.
Pre-release tests: sensitivity runs, impact distributions by segment, and a check that no two like-for-like risks diverge without a documented reason.
Post-release monitoring of lapse, claims ratio and complaint patterns against the model’s expectation.
Thresholds that trigger review or rollback, agreed in advance and owned by named people.
Clear accountability across actuarial, data science, compliance and the board.
Pricing governance also belongs in the wider risk framework. Teams that treat it as part of enterprise risk management and compliance controls, alongside other model risks, tend to find that review cycles are shorter and that accountability is clearer.
Our perspective
Most pricing projects fail at the handoff. A consultancy designs the strategy, a separate integrator builds the model, and a third party or an overstretched internal team runs it. By the time a price change reaches a customer, nobody holds the full chain from objective to outcome, and when a regulator asks why a segment moved, the answer is scattered across three organisations.
We approach insurance pricing as a production system from day one. That means engineering for continuity, governance and auditability alongside the model itself. The constraints discussed above, smoothing, pooling, fair treatment, board-approved procedures, are not obstacles to optimisation. They are the specification. A model that treats them as inputs from the first sprint is cheaper to approve, easier to defend and faster to adjust.
We also believe the sensible order is data before algorithms. Many insurers discover that policy, claims and renewal data live in systems that were never designed to be joined. Establishing readiness, then prioritising the use cases with measurable value, then building and operating the system under one accountable team, is how we think a pricing capability should be delivered. Where an insurer wants proof before scale, a short, measured test against agreed unit economics is the right gate, and our guide on proving AI ROI in a 90-day test sets out how.
— Thomas Samuel
A practical path to a pricing capability
A sequence that keeps risk low and evidence high looks like this.
Start with readiness. Audit the data you hold: policy-level history, price change records, claims and expense allocation. Identify gaps in lineage and quality before any modelling starts. This step often changes the scope of the project more than any later one.
Next, prioritise use cases by value and feasibility. Renewal pricing, new business conversion, and cross-sell offers carry different regulatory sensitivity. A well-run programme begins where the evidence is strongest and the fairness case is clearest.
Then build with constraints embedded. Cost-based relativities, smoothing rules, caps and documentation requirements are coded into the optimisation, not applied afterwards by hand.
Finally, operate. Monitoring, evaluation and ongoing optimisation keep the model aligned with claims experience, industry data and regulatory change. For the budget conversation, our AI project cost benchmarks give a way to frame what a build and run engagement involves.
Insurers exploring adjacent use cases can also read how agentic AI is changing underwriting and how claims data can be put to work in claims triage analytics, since pricing quality depends on both.
Where Sentient Concepts can help
Sentient Concepts is an end-to-end AI firm. One senior team owns strategy, build and operation, so the people who design a pricing model are the people accountable for running it. For insurers, that starts with readiness and data diligence, moves through engineering and deployment, and continues with managed AI operations that include monitoring and evaluation.
If you are scoping a pricing optimisation programme, or need a defensible view of what your data can support, see our insurance and reinsurance work or talk to us through the services overview.
Frequently asked questions
What is price optimisation in insurance?
It is the use of quantitative customer demand modelling, on top of actuarial pricing, to set premiums that meet objectives such as profit or retention. Risk-based pricing looks only at the expected cost of loss and expenses, as the US regulators’ white paper explains.
How does insurance pricing work?
Actuaries estimate expected claim frequency and severity, add expenses and a margin, and produce a technical price. Optimisation, where used, is a later overlay that adjusts those premiums to meet business metrics, within the limits of regulation and fair treatment.
Is price optimisation allowed in Malaysia?
No Malaysian policy named in our sources addresses it by that label. BNM’s fair treatment policy does prohibit unfair discriminatory practices, including different premiums for the same class and essentially the same hazard. Any model should be reviewed against that standard with your compliance team.
What should a pricing optimisation model account for?
It should account for cost-based relativities, risk pooling, smoothing of increases, Shariah requirements where relevant, and any caps on annual adjustments. For MHIT repricing up to the end of 2026, that includes spreading increases over at least three years.
How should insurers govern the model?
With board-approved, documented procedures, experience-monitoring metrics and defined triggers for remedial action. The board remains responsible for oversight of fair treatment, including repricing.
Sources
Casualty Actuarial and Statistical Task Force: price optimization white paper
Bank Negara on its ongoing supervision of ITOs and repricing of MHIT products
Milliman: Malaysia new policy document on medical and health insurance/takaful
Milliman: Malaysia interim measures on medical and health insurance/takaful
Recommended