A Roadmap of AgTech

In the past decade, there have only been a handful of large AgTech exits. Most notably, Climate Corp was bought by Monsanto in 2013 and Blue River Technology was bought by Deere in 2017. Companies like IndigoAg and Farmers Business Network have both reached unicorn status in their recent rounds, and many other AgTech companies are well on their way. In order to figure out what makes an AgTech company thrive, we took a look at AgTech 1.0 and the digital transformation underway.

Challenges from AgTech 1.0

If it isn’t broken, don’t fix it

The agriculture industry has been historically slow to adopt new technologies. One of the biggest challenges that AgTech 1.0 companies faced was a legacy mindset. The average U.S. farmer is 58 years old and their average farm tenure is over 20 years. Many farmers have been running their farm in the same manner for years. While many growers are eager to reduce their input costs, they are simultaneously hesitant to make too much change. Many growers I spoke with shared stories about trying new technology only to have the equipment malfunction in the field, unable to get the attention of a customer service team. Without the knowledge, ability, or time to repair the equipment on their own, growers lost time and revenue while they waited, sometimes unable to find a solution until the end of the harvest season due to the compressed labor market and labor shortages stemming from stricter visa policies.

Correlation or causation?

Another barrier to tech adoption is the lack of a clear return on investment. Products and services that provide a cost reduction (e.g. precision spraying) tend to easily build traction among growers, most of whom are acutely aware of their costs. In very few other industries can companies build so much traction selling into a cost center. However, while growers are aware of their costs, many find it challenging to assess the magnitude and impact of each product on revenue (yield). Many growers I spoke with did not know if they had been or would be profitable, until many months, if not years after the growing season.

Only one shot on goal

Unlike many other industries, revenue is highly seasonal: growers harvest only once per year, and in the case of permanent crops such as tree crops (e.g. nuts, citrus) the same trees will provide yield for 15–30 years. This makes the risk/reward trade-off significantly higher than in other industries. Adopting a new technology that fails could cost a grower one whole year of revenue, or in the case of tree crops, a few decades of revenue.

Many large farms combat this by partitioning their fields and testing new technology on a small subset so as to limit the risk of a bad harvest. With an average U.S. farm size of 444 acres, this strategy is reasonable for most medium and large farms, allowing technology to slowly flourish. However, even larger farms still have one revenue shot each year.

The long tail of farms

Additionally, the long tail of small family farms, while declining in total number, still represents a huge portion of total growers. The USDA estimated that as of 2015, family farms represented 96 percent of the two million+ U.S. farms and 89 percent of U.S. farm production. As a result, widespread technology penetration across the U.S. is not easy.

What’s Changing: The Digital Farm

Recent advancements in machine learning and data labeling, combined with the development of edge computing have created a new frontier for technology. Across the value chain, many parts of the farm are slowly beginning to automate. From autonomous tractors that identify weeds, to robotic harvesters that distinguish between strawberries and pests, to drones that swarm and spray precision pesticide, technology is revolutionizing the farm.

Underlying this trend is a huge amount of data collection. Growers are collecting more data than ever; however much of this data is not used for decision-making. The average farmer makes over 10,000 decisions each season. As a result, growers intuitively understand the impact of each decision on yield and profitability, but it is rarely measured. The potential impact of the data is reduced by this lack of clear causality, and many software solutions are seen as a “nice to have” rather than a “need to have”.

For most growers, making better, quicker, more accurate decisions matters more than collecting additional data. An example of this could be reducing fertilizer; one study found that 40 percent of fields are over-fertilized which not only increases the cost of inputs but also reduces yield by 15–20%. Companies that use computer vision and sensors to identify the areas of over-fertilization and offer clear, ROI-driven, and prescriptive steps to growers, could add tremendous value. Technology solutions that fit seamlessly into a grower’s workflow, requiring little to no additional work, will likely win out.

Challenging the growth of Wi-Fi enabled technology adoption is the significant acres of rural areas without continuous broadband. In the U.S., 24 million Americans live without access to fixed broadband providers, most of whom live in rural areas where agricultural lands lack consistent coverage and are not well enabled for GPS technology. In many developing countries, internet access can be a huge challenge to building on-field technology. Microsoft FarmBeats is solving this by using wireless networking technology called TV White Spaces, which exploits unused TV spectrum data to transfer data when the spectrum has low utilization. FarmBeats connects wireless sensors to a base station and a base station to an edge computer in a farmer’s home with a slow internet connection

Software

Many software platforms exist to help value chain participants increase profitability through data transparency. Some data platforms provide information directly to input providers to better understand how their inputs perform in the field, while others provide growers solutions to better navigate their costs and market their grains. Likewise, an increasing number of companies are focused on crop yield data to support both growers and buyers. For growers, the first 90 days can influence yield by up to 35%. So collecting and acting upon accurate and real-time data even more important.

I reviewed a ton of companies with sophisticated computer vision (CV) algorithms. CV is mainly used in crop production for activities such as detecting and differentiating weeds, enabling autonomous weeding and spraying, and identifying fruit for harvesting. It is also being used off the farm at processing centers for sensing and informing fruit ripeness. Overall, CV technology delivers consistent, strong accuracy. However, the data needed to support these CV models can be cumbersome and challenging to collect at scale and limited training sets exist. Furthermore, CV models are not commutable across crops, making it challenging for start-ups to build strong predictive accuracy across crops, and across a large market.

For example, a particular challenge in harvesting is identifying fruit that is hidden under the canopy. Specific crops, such as many citruses and berries, have 20%+ of the yield under canopy leaves. This means that despite using an automatic harvester, growers will still require a manual walk-through and therefore only reduce costs slightly.

It’s clear that the future of the farm is automated; certain growing practices have already started changing to reflect that belief.

For example, some growers have begun to grow both hazelnuts and chestnuts in a hedge formation, to make them easy for automatic harvesting via shaking and drying, like almonds. New seeds will emerge to help plants grow at a speed and shape that enables easy automation throughout the plant’s life cycle.

Hardware

From sensors to robots, to drones — hardware is core to enabling much of the software on the farm. While the quality has continued to improve dramatically, the costs for many systems have made them cost-prohibitive to growers. As a result, many start-ups have pivoted to a RaaS (robots as a service) model, where growers procure robotic harvesting or weeding services only when needed, and do not own the equipment. This model is still being tested but is likely to win until hardware costs further decrease.

Thank you to everyone who spoke with me this summer. This research comes from dozens of calls with farmers, entrepreneurs, academics, and other industry participants. Stay tuned for Part 2 coming next week that looks at The Playbook for AgTech 2.0.