To approve credit for a farmer, they sent a surveyor out to inspect the crops, the boundaries and the water. It took weeks, it cost a fortune, and it ate every last bit of margin on the loan. So the deals kept stalling, and whole rural regions stayed out of reach.
Assessing a farm the old way meant boots on the ground: an expert driving out to a remote property, collecting data by hand, and working from reports that were often out of date. Underwriting a single agricultural loan ran three to four weeks, and the cost of the trip devoured the profit on the deal.
It was inaccurate as well as slow. Banks struggled to judge the real value of a future harvest, or to notice weather damage and pests in time, which meant risk was mispriced and bad debts piled up on collateral no one had truly checked.
And it could not grow. Because everything depended on people making physical visits, the number of applications the company could handle in a day was capped, and expanding into remote rural areas was simply off the table.
Satellite imagery and computer vision in place of the field trip, with the underwriter still on the call.
We connected the underwriting system directly to feeds of satellite and drone imagery and real-time weather data, so a farm could be assessed the moment an application arrived.
We trained models to analyse the imagery: detect the farm's boundaries, classify the crop, and compute a vegetation-health and soil-moisture index that stands in for a physical inspection.
A predictive engine turns the geographic findings into a precise financial risk score, with an estimated value for the harvest that would serve as collateral.
Instead of waiting for a surveyor, the underwriter gets a detailed report within minutes, with a valuation and a clear recommendation to approve or decline.
The system recommends; a person approves. The underwriter keeps final judgement on every loan, now backed by evidence rather than a stale report.
Because nothing depends on a site visit, the same pipeline scores thousands of farms a day, opening rural markets the old model could never reach.
Faster, cheaper, and better at pricing the risk.
The company went from analysing a few dozen applications a week to scoring roughly 25,000 farms and fields in a single day, and cut underwriting and operating costs by more than 70% by ending the expensive field trips.
Loan approval dropped from weeks to a few minutes, and accurate, data-based collateral valuation cut the default and bad-debt rates the financiers had been carrying, so lending to farmers finally made money again.
The point was never to remove the underwriter's judgement, only the three-week wait in front of it. Replacing the field trip with satellite imagery and a model that reads the land gives the banker better evidence, faster, and lets it scale to markets a human team could never cover. It is the same pattern as our credit-underwriting work, pointed at fields instead of files.
Toxic chat was driving players off a gaming platform while regulators circled. Context-aware moderation across 30+ languages told banter from abuse, cutting toxicity 45%, lifting new-player retention 20%, and dropping false blocks 75%.
Read the case → Trust & safety · E-commerceBots and paid actors flooded a marketplace with fake reviews and harmful content. An anomaly engine plus human review blocked 95%+ before shoppers saw it, cut moderation cost sharply, and rebuilt trust.
Read the case →
Reef TRH