Democrance's Data Chief: Some Estimates Put UAE Insurance Fraud at 30% of Policyholders

A claims fraud statistic surfaced by Tanja Magas, Chief Data and Analytics Officer at UAE insurtech Democrance, is stark enough to be worth leading with: some estimates put the share of UAE policyholders engaging in some form of fraud at 30%. "This obviously accounts for massive losses to insurers," she said -- and the more useful part of the conversation isn't the alarming number itself, but the specific, unglamorous mechanism Democrance built to actually do something about it.

A fraud tool built on correlation, not black-box prediction

What Magas calls "audit alerts" is deliberately simple in structure: run a regression analysis across an insurer's existing book of business overlaid with its claims activity, and look for which data points correlate with high, repeated, or abnormal claims. Once those correlators are identified, the system builds a workflow that flags any new claim matching that pattern to a human claims-servicing representative for review -- and the model retrains on that reviewer's ongoing feedback to improve accuracy over time.

The design choice worth noting is what the tool doesn't do: it doesn't auto-deny claims or make a fraud determination on its own. It narrows a flood of claims down to the specific ones that warrant a closer look, leaving the actual judgment call with a person. "Instead of having to look through a flood of claims," Magas said, "our technology is able to pinpoint the exact ones that require further review." For an insurer drowning in claims volume with limited investigator headcount, that's a materially different (and more immediately deployable) proposition than a fully automated fraud-scoring black box.

The workflow is also explicitly product-agnostic -- it depends only on having claims data to train against, not on any specific line of business, which means the same mechanism works regardless of an insurer's book composition. And the underlying logic transfers directly to a different point in the value chain: the same trigger-and-alert pattern, pointed at the submissions stage instead of claims, can flag a broker- or customer-declared data point that doesn't match a third-party data set at the moment of quoting, before a policy is even bound.

The micro-insurance opportunity, and why the distribution model has to be rebuilt from scratch

Democrance's name is itself a compressed mission statement -- "democratizing insurance" -- and its original focus was micro-insurance for under- and uninsured populations. Magas cited figures that make the category's scale hard to dismiss as niche: the global micro-insurance market reached $78 billion in 2021, projected to reach $112 billion by 2027, a roughly 6% CAGR. Set against that is a much starker statistic: emerging markets, where most of the underinsured population actually lives, account for only about 19% of total global insurance premium volume. The gap between those two numbers is, in her framing, the real size of the opportunity.

Capturing it, though, requires abandoning conventional insurance distribution almost entirely. Traditional channels -- an agent call, a bank debit -- don't fit customers who may not have a formal bank relationship at all. Magas's example: deducting premium directly from a prepaid mobile phone plan, in partnership with a telco, can substantially outperform conventional distribution for exactly this segment, because it meets the customer inside a payment relationship they already have rather than asking them to establish a new one. That's the "out-of-the-box thinking" she argues micro-insurance specifically demands -- not a smaller version of a traditional product, but a genuinely different distribution architecture built around how underinsured populations actually transact.

Benchmarking against Apple and Amazon, not just other insurers

On Democrance's own positioning as a B2B, white-label SaaS platform for insurers, brokers, and distribution partners, Magas made a specific methodological point about how the company measures its own product experience: rather than benchmarking against other insurance platforms, Democrance benchmarks against category-agnostic consumer technology leaders like Google, Apple, and Amazon. Her justification is a customer-behaviour statistic she cited directly: 62% of customers say their experience with one industry shapes their expectations of every other industry they interact with. If a customer's baseline for a smooth digital experience is set by Amazon, comparing an insurance quoting flow only against other insurance quoting flows sets the bar too low.

Five layers, and the one that quietly blocks everything above it

Asked to describe her actual operating model as a data chief, Magas laid out a five-layer capability stack that doubles as a useful maturity checklist for any insurer assessing its own data readiness. The first layer -- a single source of truth for all platform data -- is, in her account, where most insurers actually get stuck, even though it's the least glamorous layer to invest in: "A lot of insurers today struggle to do anything upstream in data science or artificial intelligence because they just don't have the underlying data available... or it is not accurate." No amount of investment in the more exciting layers above it compensates for a broken foundation.

The second layer is regulatory data extraction and reporting -- a genuine operational burden given Democrance operates across 20 different markets, each with its own compliance requirements. The third is business intelligence dashboards, including user-behaviour analytics that can surface commercially actionable signals directly: a sharp drop-off on a quoting page, for instance, might indicate a product's pricing isn't competitive rather than a UX problem. The fourth is integrations -- both regulatory data feeds specific to a market (DHA for medical insurance, RTA for motor in the UAE) and third-party data used to prefill customer information automatically once, say, an Emirates ID number is entered, improving both conversion and underwriting accuracy simultaneously.

The fifth and most advanced layer is where Democrance does bespoke, high-ROI client projects -- and Magas's flagship example is a coverage recommendation engine, modelled explicitly on Amazon's product recommendations, that suggests appropriate coverage to a business or individual applicant based on the profile of similar peers (a restaurant of a given size, location, and menu type gets matched against what comparable businesses carry), currently running above 90% accuracy across covers and improving as the model retrains.

Where the industry is heading, and what happens to the agent

On broader trends, Magas pointed to two converging forces: an explosion of connected-device data (an estimated one trillion connected devices globally within three years, from fitness trackers to telematics boxes), and advances in "hybrid" modelling techniques that borrow causal-inference approaches from more data-rich fields like physics to compensate for insurance-specific data scarcity while an insurer builds up its own dataset over time.

Her prediction for how this reshapes day-to-day insurance work is specific rather than vague futurism: fewer traditional sales agents, with the role shifting toward portfolio advisory and customer education rather than single-product pitching; underwriting largely automated except for genuinely unusual, high-complexity risks; claims and customer service turnaround measured in seconds rather than days, handled substantially by bots; and, most concretely for motor insurance specifically, a shift from static annual premiums toward pricing that updates on driving behaviour monthly or even daily. On the question of whether automation simply eliminates jobs, her answer was direct: "It will just change the nature of work and shift people to higher-impact, higher-ROI positions, versus spending their days doing manual, mundane tasks that could be automated away."

Not a payments company, and not trying to be one

One structural detail about Democrance's own build-versus-partner decisions is worth noting for insurers evaluating similar platforms: Democrance draws a deliberate line around what it builds in-house versus where it partners. Sales, policy administration, servicing, and first-notice-of-loss are built directly on the platform. Payments -- a genuinely essential piece of any end-to-end insurance workflow -- is explicitly not something Democrance tries to own; it integrates with best-in-class providers (Magas cited Amazon Pay Services as one example) rather than building payment processing capability itself. That's a useful model for any insurtech deciding where to concentrate engineering effort: build the pieces that are genuinely differentiated, and treat commodity infrastructure like payments as something to integrate rather than reinvent.

What this means for the region

Magas's own read on where the region currently sits relative to these trends is candid rather than promotional: much of the UAE market is "struggling with some of the basics," meaning the more advanced layers of her data-capability stack -- the AI-driven recommendation engines, the real-time dynamic pricing -- remain aspirational for most local insurers until the foundational data layer is actually solid. For GCC insurers evaluating where to invest first, that's a useful corrective to the temptation to chase headline AI use cases: the fraud-detection tool, the recommendation engine, and the dynamic pricing model all depend on the unglamorous first layer being done properly, and skipping ahead to layer five without it is, on this account, exactly the mistake to avoid.

This post draws on the FS Brew episode 16: Power of Data in Micro-insurance and beyond- Tanja Magas, Democrance.