On Value Creation in Digital Agriculture

Note: I wrote this fall 2019, but was compelled to post after the latest Oct 2020 McKinsey reports on digital ag here and here. I found myself wishing (as I always do) that the analyses of digital ag’s role went beyond simply pointing at the potential to raise yields or lower input costs. I think it’s both vague and wrong. And here’s why. - Adam

Preface

One of the highlights of the year around Princeton is the annual Plowing Competition held on Labor Day weekend every year at the Howell Farm. The Howell “living history” Farm is a local treasure, donated to the county by Inez Howell, to maintain a working model of the farm she grew up on, maintained as a time capsule of 1910 rural New Jersey. The plowing competition is the crown jewel of a year filled with plantings, harvests, animal husbandry and food production.

It is a great setting to comprehend technology in agriculture. This particular era witnessed profound changes, not least the introduction of mechanization in the form of steam power. At the turn of the 19th century a farmer could expect to earn 40 cents of every food dollar (cite). In that era, seeds were open-pollinated (not hybrid) so they could be harvested and replanted. Draft power (horses, oxen, mules) did not require fossil inputs. Tillage equipment was largely just cast or forged iron. And of course synthetic fertilizer and pesticides hadn’t yet been invented. A century later, farmers buy seeds, buy fertilizer, buy chemicals for crop protection, buy machinery, buy fuel, and earn less than 10 cents per food dollar (cite).

It is a tautology that farmers are shrewd business owners that don’t part readily with a dollar, so every expenditure must have some return. I accept that people don’t pay money for things unless they perceive that the value of what they are getting is more than what they are spending. But I observe that every generation halves the number of farmers here in the US (cite), which is to say half the farmers go out of business. The farmers that remain double their land under management. Another observation, possibly related: yield gaps. In every county on earth, the average farmer is far less productive than the best farmer (cite).

Walking around the Howell farm, it must be recognized as a museum of a business that went insolvent and had to be taken over by the state (in the form of a county park). They were investing in technology (witness a 1917 Case steam thresher), and the technology delivered some return (imagine threshing by hand!). But in the final analysis, it wasn’t the right investment.

But different investments are possible, with dramatically different results. About the same distance from Princeton to the Howell farm, you can drive to visit Rijk Suydam. Mr. Suydam is a 7th generation New Jersey farmer, currently president of the NJ Farm Bureau (cite). His family moved to New Jersey in 1719 after farming the land around Flatbush Avenue in Brooklyn when they still called it Breuckelen. You can see a Suydam Ave in Princeton, but you can also find one in the Five Points neighborhood of Manhattan. On the ancestral lands, you can see high protein timothy hay growing to serve the equine market, as well as a new building housing a farmer’s market on the ground floor and an insurance office upstairs.

What we witness in the Howell farm and the Suydam farm is evolutionary dynamics at play: survivorship for some, mortality for others. But what traits weight the dice roll of those outcomes? Ultimately it is economic decision making. In another piece (Farming Fast and Slow) I applied a behavioral view of economic decisions on farms. Here though, I want to go back to Ag Econ 201 (Prof Dan Sumner, one of my favorite classes at UC Davis) and take a basic neoclassical view of a farm.

Conceptual Set up

Let’s say the profits of a farm is revenue minus costs:

Where P is the price of the good being sold (say, dollars per bushel), Q is the amount being produced (bushels), and C is the cost of production (dollars), and Pi is the profits (dollars).

Let’s say that quantity produced is a function of people (L) and physical inputs (K) and that these exhibit diminishing marginal returns (this is the classic “Cobb Douglas” production function):

By contrast, let’s assume costs are constant per unit of input:

So to expand on (1) we have:

Classically, you would take the derivative and set it equal to zero, to identify the conditions where marginal revenue equals marginal cost. You might even set a budgetary constraint. We’re not going to do that. I want to use this setup to explore the nature of value creation for innovations in agriculture. Below, we’re going to examine the terms of this equation to understand the playing field where farmers make investments in technology. If I had an extra dollar, where would I spend it?

Movies we’ve seen before

Increase Factor Productivity of Inputs

Affects c. This is the most obvious example of technology. For example: add fertilizer, get more yield. Add improved genetics, get more yield. Spray to eliminate pests, get more yield. The value proposition here is so clear that essentially no contemporary farmer doesn’t use fertilizer or improved seeds, provided they can get a loan (cite). The only uncertainty in this arena is which seeds to get, at what planting density, how much fertilizer to apply without being profligate.

Digital ag plays here to the extent it helps select the right seed. Farmers Edge, FBN, Climate and others have always had a great value proposition for benchmarking genetic performance against like environments. Adapt-N has been developed for choosing the right level of N. Yara has a suite of diagnostic tools for N optimization ranging from on-tractor to hand-held. 

Increase Fundamental Productivity

Affects a. This is land development. Putting in vineyards where once there was rangeland. Deep ripping fields with shallow hardpans. Putting in a center pivot. Poking a straw into a fossil aquifer under the desert floor. With that one big investment, the quantity produced per unit land is dramatically improved no matter what the inputs. I know there is a tech play in knowing where to buy land for certain crops, especially in the context of climate change, regulation, other stressors that are nudging production shifts. Still, the bulk of this is old fashioned iron and wood and machinery. 

Reduce labor costs

Affects d. This obviously has a checkered past in our country and many others. But labor costs remain a critical issue for many of the crops we enjoy. Technologically, this has tended to fall into several camps:

• Breeding, e.g. determinate varieties of tomatoes that ripen simultaneously, and which fall off the vine easily.
• Machinery, e.g. that scoop up the tomatoes, shake them, and sort them directly into a truck
• Architecture, e.g. trellising permanent crops along wires to be easier to machine harvest with machines or even simply without ladders.

We can see examples of this in wine grapes, which are increasingly harvested by machine (and which benefit from architecture changes to facilitate harvest), olives are now essentially hedges that are machine harvested, apples are increasingly grown under high density and never reach much taller than your arm can reach.

As far as new technology, we see a tremendous thirst for hand picked soft fruit such as strawberries and even citrus, which has led to a surge in agricultural robotics. The stated desire for robotics to replace labor may be greater than the actual willingness to pay. For example, (and correct me if I’m wrong) the weed killing bot of Blue River Technologies was eventually taken off the market presumably for lack of uptake by farmers. But then again, Driscoll’s is now working with Plenty to grow strawberries in their vertical farms (cite), so it is possible the economic pressures are so great that there is little future viability without robotics.

Reduce input costs

Affects e. Most digital ag seems to want to play here, and this is what McKinsey cites in their piece. For example, there are myriad companies offering solutions to reduce water costs by way of energy costs. Tule, AquaSpy, Powwow, Fruition (in their original incarnation monitoring xylem sap), Phytech, FloraPulse, even Arable! The list is endless to the point that it is essentially a commodity product (which doesn’t mean the value isn’t real). There is a case to be made that these actually increase factor productivity, which is to say that by watering properly you get better yield or quality. That is not controversial. But I have never seen a product that enables a direct calculation of the additional yield or additional quality associated with the proper use of the product, nor a farmer willing to pay for the additional quality they achieve, attributing the gain to the technology instead of their own innate talent. It’s hard to draw the line.

There is another group of digital technology associated with variable rate technologies that reduce input costs, ie maps for nitrogen, water, and even spraying. Growers Inc provides prescriptions for maps of N, ostensibly to reduce unneeded applications in areas of reduced productivity. They make a persuasive case as far as the value proposition. Valley, Lindsay, CropMetrics and probably others provide variable rate irrigation prescriptions that come with a shape file you can upload to your pivot controller. Most interesting from my perspective are the imaging technologies like Sentera has for mapping weeds with a drone, which in turn can be used for dynamic spray control on a ground rig. Amazone has an amazing technology to stop spraying when the wind kicks up and threatens to drift. Whereas the lettuce weeder that Blue River came out with was framed largely as a technology to address labor costs (replacing a manual job with a mechanized job), Sentera and Amazone are targeted on row crops, with a job that is already mechanized, but uses an expensive input.

Coming Attractions

I would argue that the foregoing have been the areas where digital agriculture as an industry has spent the most time in pursuit of success. And I say that without being at all confident that digital ag has ever found real success on par with, say, fertilizer or drip irrigation as technologies. I am looking for the type of success where the benefit is so clear as to drive adoption on every acre of an operation within a few years of first encounter. The farm revenue equation up above will serve as a map to explore other paths to the promised land.

Fundamental productivity, revisited

Affects a. I know I just said fundamental productivity was old. I have come to see digital ag shows up strongly here, in the form of continuous improvement. Picture a college education: does that not improve the productivity of all labor or capital inputs on the farm? And that’s often how we see people using Arable: reviewing what happened, confronting with outcomes, and discerning how to do better next time. It’s a little hard to quantify, but I nonetheless see it as one of the largest levers of long-term value creation. 

Increase Returns to Labor

Affects b. Why do so many farmers use roundup-ready soy with roundup for weed control? Because it streamlines decision making. In an era when it’s not clear you could live off of farm income at all, and can’t give full attention to the operation, simplicity reduces the time needed to focus on management. Arguably this is the only way to manage a large farm as well, if you derive all income from agriculture.

Why are giant GPS tractors more or less ubiquitous? Again, the focused thought required to run the tractor is considerably less (both if you are the actual driver, or if you are delegating the job to a junior employee). The dividend is an ability to run a larger operation, more fields in more counties.

In current times, the marquee example of digital ag as a work productivity tool is Granular. If it’s astounding to you (as it is to me) that WhatsApp had a billion users and just 63 employees at the time of its acquisition, or Instagram had a billion users and 13 employees during its acquisition, then it should be just as jaw dropping to see teams of a dozen guys managing on the order of 100,000 acres in the midwest. Sure a billion users is big, but they are all basically doing the same thing: posting pictures, making comments. The complexity of what happens in even 10,000 acres is astounding, and yet lots of people do it, and some people make money at it. And to do that means productivity tools, identifying the P&L in each production unit (field) and assimilating all the learnings from this year to make input decisions and even land divestment / acquisition deacons in the small window of time before next season.

You don’t need to tell me there are fewer and fewer people going into agriculture as a profession: I was the last person to get an Agronomy MSc at UC Davis, when they ended the major nearly 20 years ago. The only way we can really imagine that we will maintain or even improve productivity on our land base is if we build tools for people to manage large dispersed land holdings. I could spend more time on this topic, but the only way to get there is if the data from each parcel is accurate enough to be actionable (to provide a true positive / true negative signal for action), to be scalable across all the acres in an enterprise, and to be holistic in terms of the decisions being made (both agronomic and economic). I see some pieces of this puzzle coming together but I don’t see anyone with their arms around the whole thing yet: there is a large opportunity here.

Increase Prices

Affects P. If you are an entrepreneur, you probably heard the adage “You aren’t charging enough for your product.” I think it is a Marc Andreesen quote. And I get it: it is easier for me to increase the price of our product by $100 than to reduce the unit costs of our product by $100. Increasing prices is marketing (yay!). Reducing costs is engineering (gasp!). I think the same holds true for agriculture: the people who are making money are able to create some differentiation around their product, whether it be genetics (GMO free, heirloom varieties, super foods, golden rice, Arctic Apples, etc), unique environment (terroir for winegrapes, Ethiopian hillslope coffee, Juniper Ridge’s wildcrafted aromatics), or management (Fair Trade, Certified Organic, Regenerative, Demeter, Kosher, Local etc). More could be written on this, but the GxExM framework for understanding breeding is just as useful for understanding brand value in agriculture.

Indigo has pivoted hard in this direction of creating value for farmers in the prices the farm products return. Ostensibly their core technology is a seed coating that improves productivity. However, their value proposition is in part based on the fact that they use non-GM varieties, and then facilitate the (digital) marketing of those crops to people who are willing to pay more for them, employing some means of (digital) identity preservation. With the creation of Terraton initiative, those farmers may be able to sell soil carbon accumulation credits as well, backed by a substantial initiative in (digital) tools for monitoring and evaluation of compliance. With this pair of initiatives, they enable revenue from differentiated genetics AND management. Similarly, Benson Hill BioSystems started off as a trait discovery platform, but then bought a seed company to become a vertically integrated play, in other words selling seeds that command higher prices to the growers. 

IBM’s Food Trust (of which Walmart, McCormick, Driscolls and others are members) is another digital play in premium pricing. So much of technology adoption rests on passing a consumer signal (I’d pay more for food with certain characteristics) back to the growers (I’ll make the effort to record my management since I am being rewarded for it). Elliot Grant brought this home to me: again and again you see widespread industry adoption of innovations only where the beneficiary is the same as the payer.

While the blockchain that underlays the Food Trust is of course digital, all of the tools that feed into the block chain must be digital as well. The weather during crop production can be used to understand which fruits at the retailer went bad prematurely, or tasted especially good (Arable plays here of course). New portable assays can read DNA barcodes to verify provenance (SafeTraces). Cold chain tracking can alert buyers to potentially compromised fruit (Parsyl). Digital receipts from the weigh station can be chains of custody (AgCode).

A related piece of this is creating the data that enables food brands to tell their story about sustainable management. Water savings per se may not move the needle on farm incomes, but being able to tell a convincing (and verifiable) story about water savings may help protect margin against risks to reputation. Look inside any food brand’s initiative into regenerative, climate smart, or inclusive agriculture, and you will find a digital record keeping service to prove it.

Decision Making Under Uncertainty

This essentially encompasses all of the pieces of farm revenue. People have to make decisions based on an expectation of what is going to happen. How they formulate that expectation of course varies widely, but it is usually based on a historic baseline. In agriculture, people make early-binding decisions on costs (and sometimes contract prices), and only later does nature take its course and they reap the product that is sold. That is the expected versus the observed, essentially what this blog is all about.

One such example is seeding rates. A high seeding rate suggests anticipation of a good year; if a good year doesn’t happen, then productivity is below average (each plant may not get enough moisture) AND the input costs are above average. It’s more or less impossible to do perfectly when you buy seeds in the fall, but plant in the spring, and seasonal forecasts won’t capture the anomalous weather like this year’s until it is underway. 

Spensa (now part of DTN) made a great product that enabled decision making under uncertainty where they calculated the economic impact of a stressor (like an insect) alongside the price of treatment, and would tell you when it was economical to apply. It was amazing to nerds like me, but apparently it never caught on, because it was too complex to interpret in the moment. Nonetheless, this is I think a really promising direction to build products around because the potential verification of the impact is very close in time to when the advice is given and the decision is made. This is much like Google Maps or Waze: if the map says it will take you 42 minutes to reach your destination, you find out in 42 minutes if it was right or not!

I believe that sensor-driven agtech has huge potential for quick-timescale alerting to nascent conditions and clear corrective action (frost alert, spray weather, pest outbreak). Xarvio’s Field Manager is very strong in optimizing the exact optimal timing of crop protection, both whether it is needed, and when to apply if it is needed. I have always admired Semios for coupling the detection of pests, alongside automated treatment of those same pests — its a closed loop. Because the validation of efficacy is so immediate, these products have the potential to command revenue because they have many opportunities every season to build trust in the user. It took me years to fully commit to the fact that Google Maps gave me better routes than my personal experience. Now that I’ve made the leap, I essentially always use it, whether I need it or not. 

Conclusion

Did we learn anything about the Howell farm and the Suydam farm? What made one prosper and the other flounder? I’d argue that most of the technologies we see on the Howell farm were of the first variety: returns to land, cost savings, the kinds of things that are logical to pitch to any small business owner. But at the end of the day, they were farming commodity grains with intrinsically low prices, and had little opportunity to expand production to achieve economies of scale. Leaving agriculture was their ultimate decision: Inez married a future congressman and did alright. By contrast, I’d argue that the Suydam farm was making choices from the second variety: horse fodder is a specialty product that achieves a higher price, as do all the produce and value added products sold at their farmers market. Furthermore, he literally runs an insurance brokerage, handling decision making under uncertainty in the most direct way possible. And I would argue in his capacity of leading the Farm Bureau he is negotiating for giving farmers more hours in the day in the form of reduced paperwork associated with regulatory compliance.