It's Verdus day 46. Feels like the second derivative in the last 30 days was too low. At this rate we'll die out. Luckily, a lot is changing. We spent the last week in Napa driving everywhere trying to desperately validate our product, find new opportunities, and cultivate customer relationships. These are my notes.
The first winemaker said that they would have a strong use-case for drones — especially to scout vigor via IR — but the current products are too expensive or complicated (made for commercial operations). They developed zone-based scouting because of their farm's 20-year experience and the (labor) challenge of operating at the acreage of their whole vineyard. If our product offers stuff like this: “Block 14, row 7 has early mildew pressure: spray here only," I think they'd use it because the farming team wouldn't need to scale labor to do things like spraying or dusting. We can easily 100x their labor productivity for tasks that involve scouting.
The second guy we spoke to said that their farm did not have any pest or disease challenges. I think that's because that specific winery is unique — no one we've spoken to, from professors to experts, said anything like that. But who is really the "expert"—the "expert" or the actual guy on the ground? I think it's safe to disregard this conversation because it's an outlier (until we chat with another person that also says something similar)
The viticulturist from the VMC (vineyard management company) I spoke to on the phone makes me the most confident. He said that powdery mildew is a major think they look out for — especially anything that involves mealybugs. (They only spot those super late since wet-like symptoms appear on the trunks because mealybugs release honeydew after eating the phloem.) He also pointed out that their #1 problem is operating at large scales, mostly due to labor. If our tool 10x's their productivity (I'm confident marginal improvements won't convince them to buy it) then they'd use it.
Since labor seems to be the most universal pain point, I think any tool we make that increases the productivity of their laborers would be adopted. But the increase should be >10x. A small improvement won't be adopted bc of the friction of adopting/integrating a new product into their operations.
The winemaker and the ex-Pixar scout at the aggregate wine production facility we spoke to showed us the extreme scientific precision in the wine production process post-harvest coupled with the extreme intuition and experience-driven process from seed to harvest. And they emphasized how the most painful thing they deal with is labor — with rising costs and hard-to-find skill. Workers walk through the field throughout the season doing four main tasks: (1) crop maintenance (pruning suckers, training vines, etc.), (2) scouting for mildews, pests, and viruses, (3) taking random samples of individual berries from clusters throughout the vineyard to assess Brix, and (4) harvesting at the end of the season. Since disease and pest pressure could be significant if left untouched, the farmers spray liquid treatment (fungicides, insecticides, nutrients) and dust dry powdered material (sulfur dust) on everything. This costs a lot—one liquid treatment is $50/acre and applied eight times a year.
There are 47,216 Napa acres under cultivation with 475 wineries. So that's around 100 acres/winery, leading to a cost of $40,000/season just for a single liquid additive. With precision spraying, if that is cut by half, our sensing product can already provide $2,500 of value for one additive per month of the growing season not including benefits of vigor and ripeness information.
They also mentioned that most technology for grape growing is made for large commercial wineries in-house but smaller wineries aren't able to use the tech (95% of Napa wineries are family-owned). And grapes are grown on the sides of hills rather than flat ground for wind circulation benefits, making labor especially hard. Some old vintner sprained their ankle since the soil is clay-loamy, making it crack under low moisture conditions.
The obvious thing from all our convos till today is that farm labor needs to be automated. Scarcity and cost are driving this demand and it’ll happen sooner or later. Building tools to increase labor productivity or automate it is key — especially if we are the first. The question is how do we get there?
The main vineyard labor tasks from hard to easy are: harvesting (and other complex physical manipulation like suckering or pruning), Brix testing, and scouting. Precision harvesting is the hardest agricultural technology challenge. It requires a lot of capital and data to execute. Sensing is the most optimal entry point since it is foundational to harvesting and Brix testing, it would provide enough data, and establish contracts with farms. So how can we make something useful with sensing?
Farms really care about scouting for three things: vigor (how healthy the crop is), ripeness (how developed the crop is), and anomalies (hard-to-predict outliers in the features of a crop — color, size, volume, etc., generally indicating pests, disease, mismanagement, etc.). With that information, they can adjust spraying and dusting to protect crop from pests and disease, figure out when to prune, train, and treat crops, and choose when to harvest crop at the end of the season. This can save costs on two major inputs: labor (by knowing when/where to direct labor, you can increase the productivity of workers and reduce time on the field) and material input (by knowing where the anomalies are, you can more precisely spray/dust materials, saving money)
Also—to get an MVP off the ground—we need to look at what the product actually is. The product is three things: hardware, ML, and software. The hardware doesn't really matter—could be a drone, human-mounted camera, ATV-mounted camera, or even a humanoid robot. The software is flexible—could be an iOS app, web dashboard, or even a Telegram-texting-bot. But the output of the ML is constant since farmers care about vigor, ripeness, and anomalies—that won't change. A conversation with Callum at EF changed my mentality on this—walking through vineyards with a camera and showing farmers the output could be good enough to get some approved pilots. Since the drone stuff is trivial. Just need to prove to them that something useful is outputted from a camera looking at their crop.
Actually sensing might not be the move.
Just talked to a vineyard manager of the second or third largest VMC in Napa. He said they use a ton of technology—soil moisture sensors in each block/farm, documentation tracking what the workers do, satellite imagery with NDVI seasonally, etc. But they're overwhelmed. There's a ton of data but it's hard to convert that into effective labor management. They care about yield but shit like disease/pests aren't a major concern. Labor is the biggest.
Talked to another one and they said the same thing. They have a ton of data but nothing combines the insights. It's like having a home smart thermostat, fridge, kettle, lights, but they all have their own apps. They should have a Google Home.
It's obvious that data is only going up in agriculture. But the data isn't really that useful unless it works together. And works with the boots on the ground. It might be better to work on that — an operating system for agriculture — and then later build the hardware around it.