From Evidence to Trigger:

How Satellite Data Became the Infrastructure of Insurance

In the late 2000s, I was part of a team running the first pilots of satellite-based crop insurance verification in an emerging agricultural market where formal insurance had only recently been introduced. The client was an insurance company. The question was straightforward: when farmers file a claim for crop damage, did they actually plant? Did they cultivate? What satellite monitoring can prove? Did the satellite observe the reported losses?

We used mostly SPOT and IRS imagery, pulling from archives where we could, tasking new acquisitions where we had to. The analysis was manual, methodical, and slow from today’s perspective.

But the harder problem was never the data. It was the courtroom.

First of all, a judge had to accept a satellite image as legitimate evidence for or against a claim. That required something specific: not just the image, but a formal document certifying the metadata: acquisition date, geographic coordinates, sensor specifications, and the signature of the data provider.

That document transformed a satellite image from a picture into a verified fact – and the court accepted it. It was a small moment that opened the door.

Remote sensing data had existed for decades before insurance absorbed it. Landsat launched in 1972, SPOT in 1986. Even then, the technology didn’t constrain adoption; a bottleneck was shaped from non-technical questions: Could the data be independently verified? Could it be explained to a non-specialist – a judge, an underwriter, a regulator – without losing its meaning? Could it survive a dispute?

Hurricane Andrew in 1992 was a borderline – the storm cost insurers $15.5 billion, three times what the industry had modelled. Eight insurance companies went bankrupt. Painfully and clearly, the industry understood that historical claims data could not predict catastrophes. It required physical models, which in turn required physical data.

Catastrophe modelling firms started integrating satellite-derived inputs – terrain elevation, land cover, flood extents – into their hazard modules. Remote sensing entered the insurance market not through the front door of product innovation, but through the back door of risk quantification.

The first place satellite data became a direct insurance instrument, instead of a modelling input, was agriculture, specifically in emerging markets where physical inspection was impractical at scale.

NDVI, the most adopted index derived from satellite imagery, provided a measurable proxy for crop health. This made index-based insurance possible in regions where individual farm assessments would have been prohibitively expensive. By the mid-2000s, pilot programmes were running across Sub-Saharan Africa, South Asia, and emerging markets in Eastern Europe and Asia.

Index-based products carried basis risk – the gap between what the index measures and what the farmer actually experienced. Reducing that basis risk required higher-resolution data, faster delivery, and crucially, better verification frameworks. Next, the legal question arose: which data? How verified? Who accepted?

The metadata document from our pilot in 2009 was one small answer to that question. Across dozens of similar pilots in different markets, the answer was being constructed piece by piece.

The move from satellite data as evidence to satellite data as trigger is the most consequential transition in this story. And it is largely invisible to anyone outside the industry.

In the evidence model, data supports a human decision: an adjuster reviews imagery, a court reviews documents, and an underwriter uses analysis to inform a judgment. The data is an input.

In the trigger model, data is the decision. When a predefined satellite measurement crosses a predefined threshold – wind speed, flood extent, soil moisture, ground deformation – a payment is released automatically. No human review. No claims process. No dispute.

This required three things to align simultaneously: data that is fast enough and accurate enough to be trusted as a trigger; legal and contractual frameworks that accept satellite measurements as binding; and an insurance market willing to underwrite products structured around objective indices rather than assessed losses.

By the mid-2010s, all three were converging. Sentinel-1 and the commercial smallsat constellations transformed data latency from weeks to hours. Parametric products gained regulatory acceptance in key markets. And a generation of insurtech companies – Descartes among them – built business models entirely around the trigger architecture.

The Descartes data centre product matters not because data centres are a large insurance market – they are not, yet – but because of what it clearly signals about where satellite data now sits in the commercial workflow.

A hyperscale data centre campus can carry $10 billion in insured value. The primary risk is not physical destruction, but operational interruption. A flood that shuts down cooling systems causes much less structural damage than financial loss. Traditional property insurance was not designed for this. Parametric insurance triggered by satellite-observed flood extent is.

The AI infrastructure boom has created a class of assets with a fundamentally new risk profile: low physical vulnerability, extreme operational sensitivity, and concentration in geographies exposed to natural disasters. Parametric insurance built on satellite data is structurally better suited to such risk than anything that was here before.

The investment from Battery Ventures that Descartes attracted in June 2025 should be considered not as an insurance bet but as a technology infrastructure bet. The firm that backed Guidewire, which transformed P&C insurance into a software industry, is now backing the company that is transforming risk transfer into a data-triggered system.

The distance between those two points is not primarily technical, although satellites have got better, data have got cheaper, and algorithms have significantly improved. However, the real shift was institutional: building the frameworks, standards, and legal precedents that allowed objective data to be trusted as the basis for financial decisions.

The progress was made slowly, mostly invisibly, in pilots and courtrooms and regulatory consultations across dozens of markets. And it is still happening.

What Descartes and the data centre product represent is not the final destination – it is evidence that the process has gone far enough for a technology investor to make a large bet on the outcome.

And insurance is only one corner of this story. The same satellite data, the same verification frameworks, and the same trust infrastructure are now being built into climate finance at a global scale. That is a larger conversation to be continued in the next papers.