Who Decides If Your Crop Died
On the politics of algorithmic truth at the bottom of the income distribution
Lê Thị Phương stopped planting her third-season rice crop in 2020. She hasn’t risked it since. That season, the 2019-2020 dry season, saltwater pushed more than 90 kilometers inland through the Mekong Delta, poisoning wells and fields across ten of the Delta’s thirteen provinces. Her harvest failed. Her nine cows died drinking salt water. The brick house she’d built from years of three-crop income still stands in Bến Tre province. The field beside it lies fallow every dry season now.
She is not a metaphor. She is a specific farmer in a specific province in one of the most climate-threatened agricultural zones on earth. The Mekong Delta feeds a nation and is home to nearly 20 million people. Saline intrusion, once an episodic hazard tied to exceptional drought years, is now an annual occurrence. Since 2016, it has cost Delta farmers an estimated $2.8 billion. Scientists studying current sea-level trajectories warn that more than 45 percent of the Delta could be submerged within decades.
Help is coming. It looks like a satellite.
The traditional insurance adjuster is an economic impossibility for smallholder agriculture. Consider the arithmetic. A rice farmer in Bến Tre might work a plot of half a hectare. The crop is worth a few hundred dollars. Sending a human being to visit, inspect, document, and adjudicate a claim costs more than the policy is worth. This is why, across the developing world, smallholder farmers have historically had no meaningful crop insurance: not because insurers lacked interest, but because the unit economics of indemnity-based coverage simply don’t work at this scale.
Parametric insurance changes the arithmetic. Instead of an adjuster, a parametric policy pre-agrees on a trigger: a measurable index derived from satellite imagery, weather station data, or field sensors. When the index crosses a defined threshold, a payout fires automatically to a mobile money account. No visit. No claim form. No wait.
In the Mekong Delta, this is no longer theoretical. Igloo, a Singapore-based insurtech, launched a blockchain-based weather index insurance product for rice farmers across eight Delta provinces in 2022. Premiums start at $8 per hectare, with a minimum coverage area of 0.1 hectare. Rainfall data flows from Vietnam’s national meteorological service. Smart contracts read the data and release payments when the index trips. In parallel, Hillridge, an Australian insurtech, has partnered with MSIG Vietnam and Australia’s Department of Foreign Affairs and Trade to build satellite-based drought and typhoon products for the same region, using commercial satellite imagery calibrated to a five-kilometer grid. A 2024 market report from that consortium estimated that $6.8 billion in agricultural and aquaculture production across the Delta is potentially suitable for parametric protection.
The appeal, honestly stated, is real. Cash before the next planting decision, not after the harvest rots. Financial access for farmers who have never qualified for any insurance at all. A mechanism that reaches people the adjuster never could.
Here is where the question this piece wants to ask enters the frame.
The trigger is a design choice. It is developed by actuaries, satellite engineers, and product managers using historical data, statistical models, and actuarial tables. Someone, at Igloo’s Singapore office or Hillridge’s Australian headquarters, set the thresholds that determine when a payout fires in Bến Tre. The farmer in Bến Tre did not.
Rice plants cannot survive in soils with salinity above 4 grams per litre. This is established plant physiology. If a parametric salinity policy is calibrated to trigger at 4.5 g/L, and saltwater reaches 4.2 g/L in a specific field, that crop is dead. The satellite records nothing that crosses the contractual threshold. The mobile money payout does not arrive.
This is the system operating correctly. The threshold was crossed, from the farmer’s perspective. It was not crossed from the algorithm’s. And the farmer has no standing to challenge the difference.
The technical name for this gap is basis risk: the mismatch between what the index measures and what the policyholder actually experiences. It cuts in both directions. A farmer whose crop fails may receive nothing if the index never reaches its trigger. A farmer whose crop survives may receive a payout if it does. The parametric insurance literature has recognized basis risk as the field’s central structural problem since early pilots in the 2000s. AI and satellite technology have reduced it. They have not eliminated it. Policymakers, development banks, and insurtech startups tend to emphasize the reduction. The residual gap is where the farmer lives.
The most instructive case study of what basis risk looks like in practice occurred in Malawi in 2016, and the mechanism is worth understanding in detail.
Malawi had purchased parametric drought insurance from the African Risk Capacity, a sovereign risk-pooling facility backed by the African Union, the G7, and the World Bank, paying a premium of $4.7 million for the 2015-2016 agricultural season. In April 2016, President Peter Mutharika declared a state of national emergency. A severe drought had destroyed crops across the country. ActionAid estimated that 6.7 million Malawians were food insecure.
The African Risk Capacity’s satellite-based model returned a different figure. It is estimated that 20,594 people were affected. No payout was triggered.
When investigators from Lilongwe University eventually traced the discrepancy, it came down to one variable: maize. The model had been calibrated using a variety with a 120-to-140-day growing cycle. Malawian farmers, adapting to increasingly erratic rainfall over the preceding years, had shifted to a 90-day variety. The shorter-cycle maize is far more vulnerable to mid-season dry spells. The rainfall pattern of 2015-2016 was particularly brutal for the 90-day variety, and mild enough for the 120-day variety the algorithm assumed was in the ground.
The farmers had already adapted to a changing climate. The model hadn’t noticed.
Under pressure from civil society and the Malawian government, the African Risk Capacity recalibrated the model and issued a revised payout of $8.1 million in November 2016, seven months after the emergency declaration. ActionAid’s subsequent assessment noted that total drought response costs to the Malawian government reached $395 million. The insurance paid out roughly two percent of the actual need, months late, because the algorithm had been watching the wrong crop.
The algorithm is not corrupt. It does not take bribes or ignore the rain. It measures what it was told to measure, at the resolution it was given, against thresholds set before the season began. When those measurements diverge from experienced reality, the farmer loses, and there is nobody to argue with. The adjuster, who was slow and sometimes corrupt and occasionally wrong, could at least be contested. The satellite cannot.
Consider who sits in the room where thresholds are designed for the Mekong Delta products.
Igloo, incorporated in Singapore, designs and operates the insurance product. Hillridge, incorporated in Australia, owns the satellite-data platform. MSIG, headquartered in Japan, underwrites the risk. SCOR, headquartered in France, acts as a reinsurer. Vietnam’s national meteorological agency provides rainfall data. Australia’s Department of Foreign Affairs and Trade provides the catalytic grant funding that makes the economics viable at the premium levels small farmers can afford.
The farmer in Bến Tre pays the premium. She owns the paddy. She has no seat in the room where reference crops are chosen, satellite grid resolution is set, or trigger thresholds are calibrated.
Robert Bergsvik and Sanneke Kloppenburg, writing in Earth System Governance in December 2024, find that the combination of satellite monitoring and algorithmic payout reduces on-the-ground complexity in how disasters are perceived and governed, intensifying what they call the depoliticization of climate disasters. The political question of who owes whom becomes a technical question with a binary answer: did the index cross the threshold? When the answer is yes, value flows to the farmer. When the answer is no, the farmer has no standing to challenge the measurement because the measurement is objective. The question of what happened in her field has been replaced by the question of what the satellite recorded. These are not the same question.
A further twist: when the algorithm fails, a practice has emerged in several African parametric programs that makes the politics visible again. Researchers documenting parametric insurance programs across the continent have found what insurers call “ex gratia” payments, from the Latin for “from grace”: informal transfers made to aggrieved clients even when nothing is contractually due, deployed to preserve trust in markets where that trust is fragile. These discretionary payments reintroduce human judgment, informally and without transparency, precisely when the formal algorithm has collapsed. The adjuster, the system was designed to eliminate returns quietly, through the back door, now with no name, no face, and no accountability.
In an earlier piece on this Substack, I followed a related question through a different set of streets.
That essay looked at Voice AI and the informal economies of Nairobi and Delhi: the Sheng-speaking matatu conductors who run Kenya’s minibus networks and the Hinglish-speaking kirana owners who run India’s corner stores. These are people who have built deliberate opacity into their operating languages, partly as craft, partly as self-protection. Voice AI trained on those languages makes the opacity legible to the platforms that want to reach these markets. The question the piece asked was simple: when an informal, opaque system is made legible, who captures the value of that legibility?
Parametric insurance poses the mirror image of that question.
The conductor’s opacity was his own. It was designed for nobody’s benefit but his. When legibility is extracted, value flows upward toward the platform.
The farmer’s disaster is also made legible, but the architecture runs in the opposite direction. She contracts for legibility. She pays specifically so that her loss will be seen, measured, and paid. The promise of parametric insurance is that legibility flows downward: the satellite sees the flood, the algorithm reads the salinity, the mobile money account fills before the next season starts.
Both stories share the idea that the instruments of measurement are controlled by someone else. In the informal economy, legibility was extracted without consent. In parametric insurance, legibility is contracted for. But in both cases, what gets measured, how, at what resolution, and against which baseline, is decided by someone who is not standing in the field.
Above the Mekong Delta, several commercial satellites pass daily, measuring salinity, soil moisture, and vegetation indices. They are extraordinary instruments. The data they generate is dense, precise by any historical comparison, and genuinely useful.
They do not record nine dead cows.
They do not record a farmer who has not planted a third season in five years because she cannot absorb another loss. They do not record the fact that she switched to a shorter-season variety years ago because the rains became unreliable, or that the satellite’s five-kilometer grid does not distinguish her half-hectare from the better-drained plot beside it. They record what they were calibrated to record, at the resolution someone paid for, against thresholds set by someone in Singapore or Melbourne before the season started.
Parametric insurance is a genuine response to a genuine crisis. The speed advantage over indemnity insurance is real. The ability to reach farmers that the adjuster never could is real. The potential for AI and improved sensor networks to further narrow basis risk over time is real.
None of that resolves the question that the Malawi case makes concrete. The farmers had already adapted. The model was watching the wrong crop. Six million people were hungry, and the algorithm said 20,594 were affected. The gap between those two numbers is not only a data quality problem, though better data would narrow it. It is a question of who controls the instruments that define the boundary between disaster and ordinary hardship, and what recourse exists when those instruments are wrong.
Somewhere tonight, above Bến Tre province, a satellite is passing. Lê Thị Phương’s field lies fallow. The instrument overhead has no entry in its database for a farmer who learned not to plant.
Sources: Saline intrusion cost data and farmer testimony (Lê Thị Phương) from Dialogue Earth/Mekong Eye (May 2025). Igloo product details from Igloo press release, November 2022. Hillridge/MSIG market report from MSIG Vietnam, May 2024. Rice salinity tolerance threshold (4 g/L) from “Interacting effects of land-use change and natural hazards on rice agriculture in the Mekong and Red River deltas,” Natural Hazards and Earth System Sciences (2021). Malawi case: ARC Ltd press release, November 14, 2016; ActionAid, “The Wrong Model for Resilience” (May 2017); Bloomberg Green (December 2024). Depoliticization argument: Robert Bergsvik and Sanneke Kloppenburg, “The depoliticization of climate disasters: Unpacking the entanglement of satellites with parametric climate risk insurance,” Earth System Governance, Vol. 22, Article 100221, December 2024, DOI: 10.1016/j.esg.2024.100221. Ex gratia payments: preprint documented in ResearchGate (2024).
I will be thinking for a long time about your argument on the direction of legibility - and perhaps also (is it possible?) the intersection of policy to recalibrate the power balances. Informal practices are how people find their power when the economy excludes them. We imagine formalising so that they can be included. I have always had a problem with the notion of “inclusion” - it presumes a direction, when in fact the opposite may be more productive, more efficient, indeed more ‘inclusive’ at least from an outcomes perspective. We have a project in a tenement area of Nairobi known as Pipeline. The entire area is informal by definition - tenure, building control, rental arrangements, even governance. The buildings are solid, thankfully, but small, dark, with very poor access to water and sanitation. Eight-storey walk ups, no lift; quality is poor. But it’s serving a segment of the Nairobi population that no other housing form serves and the location is brilliant. The study is asking - can we align the incentives differently to achieve a better housing outcome? I think it’s the same question as what you’re asking here…. Excellent article. Thank you.