AI Underwriting: Speed vs Accuracy in Risk Assessment

When war-risk insurance for vessels transiting the Strait of Hormuz collapsed in March 2026 — stranding more than 150 ships and sending single-voyage premiums from tens of thousands to several hundred thousand dollars overnight — it exposed something the insurance industry had quietly known for years. Traditional underwriting, built on committees, spreadsheets, and institutional memory, cannot price risk at the speed reality now demands. The question facing every insurer from Muscat to Kuala Lumpur is no longer whether artificial intelligence belongs in the underwriting room. It is whether the models are genuinely accurate enough to be trusted when the stakes are this high.

The Speed Imperative Is No Longer Theoretical

The Gulf maritime crisis crystallised a problem that had been building for years across emerging market insurance sectors. When the International Group of P&I Clubs withdrew war-risk coverage effective March 5, 2026, the commercial response needed to be near-instantaneous. The U.S. International Development Finance Corporation's subsequent $20 billion Maritime Reinsurance Plan — with Chubb confirmed as lead partner — proved that state-backed capital can move at speed when political will exists. But the pricing, structuring, and risk stratification of policies within such a programme is precisely where AI underwriting tools are now being stress-tested at scale.

Insurers deploying machine learning models re-priced voyage-by-voyage war-risk exposure within minutes of geopolitical triggers, drawing on satellite vessel-tracking data, real-time conflict mapping, and historical loss databases. Their human counterparts were working through the night. The throughput gap was not marginal — it was several orders of magnitude. Yet speed without accuracy is not underwriting. It is speculation with a premium attached.

Where AI Underwriting Actually Breaks Down

The accuracy problem in AI underwriting is more nuanced than critics acknowledge, and more serious than proponents typically admit. Current-generation models perform exceptionally well in high-frequency, data-rich lines: motor, personal health, and standard commercial property. In these categories, AI underwriting platforms have delivered loss ratio improvements of 8 to 15 percentage points compared to traditional methods — figures cited by several Lloyd's of London syndicates active in the GCC market.

The failure modes emerge at the margins. That is precisely where the most consequential risks sit. Takaful products, which operate under Shariah-compliant risk-sharing structures rather than conventional indemnity frameworks, present a structural challenge for models trained predominantly on Western actuarial datasets. Al Madina Takaful's recognition as Best General Takaful Provider in the Middle East Region 2026 — awarded at the London Stock Exchange in August and accepted by CEO Usama Al Barwani — reflects the sophistication the sector has built over decades. But that sophistication is largely human-built, rooted in community trust, Shariah board oversight, and relationship-driven distribution. Training an AI model to replicate or enhance those dynamics requires data that, in many cases, has never been systematically collected.

Climate-linked underwriting presents a parallel problem. Flood modelling across the Nile Delta, cyclone exposure in the Arabian Sea, heat-stress mortality risk in West Africa — all require localised hazard data at resolutions that global models struggle to deliver. Insurers underwriting infrastructure or agriculture in Nigeria, Kenya, or Oman are working with proxy data and assumption-heavy extrapolations. When an AI model misprices that risk, the error compounds across an entire portfolio before a single claims event surfaces the problem. By then, the damage is done.

ESG Integration Is Reshaping the Underwriting Question

Tawuniya's achievement of an MSCI ESG 'A' rating — detailed in its 2025 Sustainability Report and outlined publicly by CEO Othman Alkassabi in April 2026 — signals something worth paying close attention to. For leading GCC insurers, the underwriting question is now inseparable from sustainability strategy. Aligning with the Principles for Responsible Investment is not a reputational exercise. It carries direct implications for how underwriting models are trained, what data inputs are weighted, and which risks a company chooses to decline entirely.

AI underwriting tools incorporating ESG scoring are beginning to discriminate between counterparties at the point of risk acceptance — not just at pricing. A family office-backed real estate development in Riyadh that meets green building standards may receive materially different terms than an otherwise comparable project that does not. That is a significant shift. For private investors and family offices active across the GCC and Central Asia, it creates a concrete incentive structure: the cost of insurance capital is increasingly tied to environmental and governance posture, not just asset quality.

The firms moving fastest on this integration are not always the largest. Boutique managing general agents operating in UAE free zones — and in markets like Georgia and Azerbaijan — are deploying AI underwriting platforms with ESG filters built in from inception, unburdened by legacy systems that would require expensive retrofitting. Few outside those markets have noticed. They should.

The Reinsurance Layer Is Where AI Accuracy Gets Tested

Reinsurance is the stress test for any underwriting system. When primary insurers lay risk off onto reinsurers — or when state actors like the DFC structure $20 billion facilities to backstop commercial markets — the quality of the underlying risk assessment cascades upward. A primary AI underwriting model that misprices frequency risk by three percentage points creates a compounding distortion by the time that book of business reaches retrocessional markets. The numbers tell a complicated story, and not everyone at the top of the tower is reading it clearly.

This is why leading reinsurers operating in the Gulf and Southeast Asian markets now require primary cedants to disclose their underwriting methodology as part of treaty negotiations. Whether a risk was priced by an experienced human underwriter or a third-party AI platform with a 90-day training dataset is becoming a material consideration. Munich Re and Swiss Re have both signalled — through market guidance issued in late 2025 — that model transparency will be a condition of capacity access in certain specialty lines by 2027. That deadline is closer than it sounds.

What This Means for Investors and Family Offices

For private investors with material exposure to the insurance sector — whether through direct equity stakes, participation in Lloyd's syndicates, or investments in insurtech platforms across Southeast Asia and Africa — the AI underwriting debate resolves into a single due diligence question: where in the accuracy-speed tradeoff does this business actually sit?

The most durable positions belong to insurers that have combined AI-driven efficiency with domain-specific model governance. Proprietary regional datasets. Shariah-compliant product architecture that is genuinely embedded rather than bolted on. Leadership that treats ESG integration as a pricing discipline rather than a disclosure obligation. The Gulf's leading incumbents — and a growing cohort of well-capitalised challengers in markets from Casablanca to Ho Chi Minh City — are beginning to demonstrate that speed and accuracy are not a binary choice. They are a function of data quality, model governance, and institutional expertise. The firms that understand this distinction will define the next generation of insurance capital allocation across emerging markets. The firms that do not will find out the hard way.