Aerobotics: Farmers Buy Decisions, Not Pictures

South African agritech has produced one genuinely instructive company, and its lesson is about what customers pay for. This story covers the technology, the tree crop focus, per-tree analytics, the shift to yield forecasting, the business model, international expansion and why agritech is hard — part of the South Africa Company Stories hub.

Why are tree crops the right starting point?

Because each tree is an individual asset worth tracking. An orchard represents years of investment per tree, trees remain in place for decades, and problems affecting a single tree can be identified and treated individually.

Technically, discrete objects in regular patterns are far easier to detect and measure from imagery than a continuous crop canopy, which makes per-unit analytics genuinely achievable rather than aspirational.

Commercially, the value per hectare is far higher than in field crops, so a grower can justify spending on analytics that would never pay back across thousands of hectares of low-margin grain.

What does per-tree analysis actually detect?

Missing or dead trees, canopy size differences indicating stress or poor establishment, irrigation problems visible as patterns across blocks, and areas where growth diverges from the surrounding orchard.

The economic argument is straightforward: a gap in an orchard produces nothing while occupying land, water and management attention, and identifying gaps across thousands of trees is impractical on foot.

Detecting problems early also compresses the response time. A stressed block identified in weeks rather than at harvest can often be corrected within the same season, which is where the return sits.

Why did yield forecasting become the key product?

Because it answers the question that determines money. A grower who knows the expected volume and timing of a crop can schedule labour, book packhouse capacity, negotiate sales contracts and plan logistics.

The downstream customers care even more. Packhouses, marketers and retailers plan capacity and commitments months ahead, and inaccurate crop estimates cause either wasted capacity or unmet contracts, both of which are expensive.

It is also a product with a clear, measurable benchmark. A forecast is either accurate or it is not, which makes the value provable in a way that health monitoring is not — and provable value is what sustains a subscription.

Why is agritech commercially difficult?

Because farmers have one buying cycle a year, are appropriately sceptical of technology claims and evaluate everything against a return per hectare that must be visible within a season.

Sales cycles are correspondingly long. A grower typically trials on a small block, evaluates over a full season and expands only if the result is clear, which means customer acquisition takes years rather than quarters.

Seasonality also distorts the business. Revenue and workload concentrate around specific growth stages, and a company must fund a full year of engineering from a compressed selling and delivery period.

The additional constraint is that most farms are not large enough to justify enterprise pricing, so the model must work at a price point closer to software for small business than to agricultural equipment.

How does the drone-versus-satellite question resolve?

Practically, by combining them. Satellite imagery is cheap, frequent and covers large areas at moderate resolution; drone imagery is expensive and infrequent and resolves individual trees in detail.

The sensible architecture uses satellite data for regular monitoring and change detection, and drone flights where the detail justifies the cost — a specific problem, a high-value block, or a formal yield estimate.

Satellite resolution and revisit frequency have improved substantially, which has shifted more work to the cheaper channel over time and reduced the operational burden of flying.

What is the business model?

Subscription per hectare or per tree, sometimes with additional charges for specific surveys or formal yield estimates, sold to growers directly and increasingly to packhouses and marketing organizations covering their supplier base.

Selling to aggregators is more efficient than selling farm by farm. A packhouse representing hundreds of growers can adopt across its supply base in one decision, which shortens the sales cycle dramatically.

It also aligns the product with the customer who benefits most, since the packhouse gains directly from better forecasts across its whole intake rather than from any single orchard’s performance.

Why expand internationally?

Because the addressable market in South African tree crops is limited, while the same crops are grown at far greater scale in the Americas, Australia and southern Europe under identical technical conditions.

The product transfers cleanly. A citrus or almond tree behaves the same way regardless of jurisdiction, so the analytics require adaptation rather than reinvention — which is unusual and valuable in a market-specific industry.

The commercial challenge is distribution. Selling into large agricultural markets requires local presence, agronomic credibility and relationships with the packers and cooperatives that aggregate growers.

What are the risks?

Competition from equipment manufacturers and input suppliers who bundle analytics with products growers already buy, and who can afford to give data away to sell chemicals or machinery.

Forecast accuracy is an existential product risk. A yield estimate that proves wrong in a commercially damaging way undermines the entire proposition, and agricultural customers share such experiences quickly among themselves.

Funding cyclicality is the third. Agritech raised heavily during a period of enthusiasm for the category, and companies that scaled costs against that funding environment have had to retrench as it tightened.

What is the lesson?

That customers buy decisions, not data. The imagery was impressive and the product that sold was a number a grower could act on commercially.

The second lesson is about finding the customer with the most to gain. The grower experiences the benefit; the packhouse and marketer experience it multiplied across a whole supply base, which makes them the better buyer.

The third is that some technology is genuinely global from the start. A product built for South African citrus works in California and Spain, which is rare and is exactly the characteristic that makes a small-market company internationally viable.

What data does a grower already have?

More than outsiders assume: irrigation records, spray programmes, soil analyses, harvest volumes per block and years of accumulated personal knowledge of which parts of the farm perform.

The value of aerial analytics is therefore incremental, and it must add something the grower cannot already see — consistency across a large area, early detection, or a quantified estimate rather than an experienced guess.

This is why integration matters. Analytics that sit in a separate application the grower must remember to open compete poorly against a system already embedded in how the farm records its season.

How does water scarcity change the value proposition?

It raises it substantially. Where irrigation water is allocated and expensive, identifying blocks receiving too much or too little translates directly into cost saved and yield protected.

Aerial imagery detects irrigation failures quickly — a blocked line or failed emitter shows as a pattern of stressed trees long before it would be noticed on the ground.

In regions facing recurring drought and tightening allocations, this is the argument that most reliably converts a trial into a subscription, because the saving is measurable within a single season.

How is a yield forecast actually produced?

By combining per-tree canopy measurements with fruit counts sampled from high-resolution imagery or ground surveys, then modelling the relationship between observed characteristics and eventual harvested volume for that crop and region.

Accuracy improves with history. A model calibrated against several seasons of actual harvest data for the same orchards performs far better than one applied to a farm for the first time.

This creates a genuine data moat: the accumulated relationship between imagery and realized yield across thousands of blocks is an asset a new entrant cannot buy, only build over equivalent seasons.

What is the packhouse’s perspective?

A packhouse must schedule labour, packaging, cold storage and shipping against fruit it does not control, arriving from many growers whose own estimates are frequently optimistic.

An independent, consistent forecast across its whole supply base lets it commit capacity and sales contracts with confidence, and the cost of being wrong — idle lines or unfulfilled contracts — dwarfs the price of the analytics.

This is why selling upward through the chain works better than selling farm by farm: the buyer with the largest exposure to forecast error is the buyer most willing to pay to reduce it.

Why did agritech funding cool?

Because early enthusiasm assumed technology adoption curves from software applied to an industry with annual buying cycles, thin margins and a customer base that changes practice slowly and for good reason.

Revenue growth accordingly took longer than investors underwrote, and companies that had scaled sales and engineering against optimistic projections had to reduce sharply when funding tightened.

The businesses that have continued are those whose product produces a measurable financial result within a single season, which is the only claim that survives a grower’s scrutiny in a difficult year.

Which crops does the model extend to?

Naturally to other permanent tree and vine crops — citrus, nuts, pome and stone fruit, olives, avocados — where individual plants are discrete, long-lived and valuable enough to justify per-unit analysis.

Extension into row crops is harder, since the unit of analysis becomes an area rather than a plant, the value per hectare is lower, and established competitors already serve broadacre farming with satellite and machinery data.

The commercially sensible path has been depth rather than breadth: more accurate forecasting and more crops within the permanent category, where the existing models, data and customer relationships all transfer.

Frequently Asked Questions