ClimateAi builds ‘adaptation playbook’ via long-range forecasting: ‘I think we’ll be profitable by Q3 or Q4 of next year’

A flurry of startups offering AI-powered ag insights driven by long-range weather forecasting has emerged on the scene in recent years. But are they delivering actionable info, and what, if anything, does the demise of industry pioneer Gro Intelligence say about the value of this data for food & ag companies?

AgFunderNews (AFN) caught up with Himanshu Gupta (HG), cofounder and CEO; and Will Kletter (WK), VP operations and strategy; at San Francisco-based ClimateAi to explore how it is translating long-range forecasting data into an “adaptation playbook.”

AFN: Give us the brief origins story of ClimateAi

HG: When we [Gupta and cofounder Max Evans] came out of Stanford [in 2017] our value proposition was, if we can forecast weather beyond two weeks, then that is going to be very actionable information. So for the first three or four years, we just focused on developing the technology. We decided to commercialize it first in food and agriculture supply chains because weather drives so much of the output.

AFN: How do you predict the weather further than two weeks out?

HG: The data we use is mostly public data from satellite sensors, radar stations, and weather stations, and in some cases our partners also give us access to their private weather stations. We also tap into oceanic buoys for things like sea surface temperature, salt content and so forth, which we can use as a predictor of long-range weather. We then add that to the output of NOAA [the US government’s National Oceanic and Atmospheric Administration] and ECMWF [the European Centre for Medium-Range Weather Forecasts] which gives out model more reliability.

WK: Most weather forecasts look ahead 10 days or 14 days, we’re looking six months out, or 10-20 years in the future. And we’re not just saying it’s getting warmer, but what does that mean, what are the chances of heat during flowering over the next 10 years for this crop in this region?

There’s a lot of forecasting information. There’s a lot of ground truth information from satellites and sensors. There’s a lot of novel data sources such as oceanic buoys that are monitoring shipping lanes, but can also provide very valuable insights on what’s happening in the ocean, which a couple months later will influence what’s happening on land.

At ClimateAi, the primary innovation is about making these data sources talk to each other in a way that they haven’t, because there is quite simply too much information to be addressed without the use of advanced tools such as machine learning. We’re evaluating all the available forecast streams that exist, bringing in novel data points, such as the oceanic signals, and comparing them to ground truth actuals, seeing which models in which time periods for which variables— temperature, precipitation—have been the best, and then creating a unique forecast for individual crops.

AFN: How do you translate that into actionable insights for food & ag companies?

HG: It’s not just about forecasting risk of heat waves or wildfires or droughts, it’s about translating the impact into specific business or crop metrics. So we have biophysical AI models that explore things like how does a heatwave impact a particular variety of corn at a particular place? Or how might drought or heat at certain times impact almonds vs pistachios?

You have to address the ‘so what?’ factor. What can companies do with this information to reduce their risk or better manage their supply chains? And through that, we’ve developed an adaptation playbook. So let’s say we forecast that yields in certain areas for a given crop will reduce. How can we help people figure out new locations to source from, or help farmers with water management, or how can clients work with farmers to set up new growing regions?

As an example, we had a big seed company that came to us and said can you use your models to figure out new growing locations for tomato seeds in India to supply the Indian market? So we used our tools to figure out the best new growing locations right in front of them during a Zoom call, and they said, Wow! It took us three years to figure out the same locations, and you’ve just done that in a matter of minutes?

WK: We are seeing that climate adaptation is moving to be as important or maybe in some cases more important than carbon reduction in the minds of sustainability leaders, the C-suite and chief risk officers.

Food and beverage customers are interested in understanding the long-term viability of their sourcing regions and what actions they can take to enhance that viability, and also what other actions they may need to take to ensure they have enough quality and quantity in the future.

For long-range planning, we identify a risk to a specific location. You then need to understand how to mitigate the risk that has been identified. So our water index tool can tell you what type of water infrastructure to invest in. If certain events are more likely to repeat in the future than some of the other events, do you need to explore emerging regions that may not be suitable for this crop today, but will be suitable in the future? Do you need to invest in more climate resilient varieties?

AFN: What kind of clients do you work with?

HG: We started working with a lot of seed companies, then we expanded to agrochemical companies and then food processors and packers such as Driscoll’s, Dole, American Sugar, AB InBev and Suntory. We also work with investors in this sector and debt investors such as Rabobank.

[As a point of entry to a prospective client] we focus on heads of procurement, who are constantly trying to manage volatility. So for Suntory, for example, we might say, how might the weather from two weeks and beyond impact barley yields and quality across its sourcing areas? So it might initially be about contract optimization. Then as the season progresses, it also wants to mitigate risks, so let’s say a heatwave is coming up somewhere in one growing region, it can take positions in the spot markets or increase allocation to other suppliers before it’s too late.

WK: In the context of extreme weather, you can manage the impact of an extreme weather event, let’s say a heat wave during the flowering period, by changing planting dates, by increasing irrigation, and so Suntory is sharing insights with its suppliers to help it manage and mitigate the impact of a specific risk.

But they’re also making capex investments in various processing facilities that are based on assumptions around where they will get volume from in the future. So they need information to help them prepare 10 years in advance.

AFN: Are companies typically skeptical when you first talk?

WK: We face three main challenges.

One is the same challenge that any new technology faces – it doesn’t exist as a line item in the budget – yet.

Two, we’re asking users who are used to operate a certain way to think differently: ‘I’m used to thinking about my planting day based on the seven day forecast; now you’re telling me to look three months ahead and change my planting date based on the whole season. That’s a change in behavior. We encourage customers to start with the low risk, high ROI, low hanging fruit types of decisions, use that to build confidence and go from there.

Three, we’re trying to get folks to see us as not purely a sustainability offer, but really about driving strategically important decisions around the bottom line.

And I think that is where we are extremely excited about the movement we’re seeing in the market because there is a realization that from a sustainability perspective, adaptation is as much of a priority as mitigation. And part of that realization is that adaptation has direct bottom line implications that are mission critical to your business. So we’re starting to see the conversation change.

HG: There’s always skepticism, but that’s where transparency from our side really helps. We are very honest about what risks we are very good at forecasting and what we’re not good at predicting, such as hail. It’s very difficult scientifically to forecast hail probability.

Right now, companies have forecasts that go up to 14 days and then they basically just use historical averages and reports from consultants or whatnot, so it’s very fragmented information whereas we are providing a more scientific way of making a decision.

Typically, when the IPCC quotes numbers, they talk about averages, so by 2050 on average, barley is going to be down by X%, or whatever, but that’s not always very useful. The bigger problem for supply chain professionals is the volatility they’re seeing year-to-year. Over the last 10 years, we have seen the highest volatility that we have seen on decadal basis in all the commodities.

AFN: How do you validate your tech and show potential customers that your forecasts are accurate/useful?

HG: If customers want to make a procurement decision on sourcing spinach or barley for example, they want to see how well our models have previously forecast these crops over the last five or 10 years.

We also do what we call hindcasting around climate events. Were you able to forecast a heat wave around the flowering period for this particular crop? And so we can say that yes, we saw it coming two months out. So for example, like the Pacific Northwest heat dome event in 2021, our model picked up signals in February, while NOAA and everyone else didn’t pick up the signals until April.

WK: A lot of the underlying forecast data that we’re using is stored, so we can access decades back and then demonstrate how accurate we would have been at forecasting the present day 10 years ago or three months ago, or whatever the timescale is. So we do a lot of what we call hind cast validation, to show a customer, Hey, this is the region that you produce in, this is in general how good we are at forecasting certain parameters.

If you make your planting decisions in March, so we’ll show you not just how good are we at forecasting six months out, we’ll show you how good are we at forecasting six months out starting in March because that may differ slightly.

And so that hindcast validation process is really important to build confidence, but it also actually helps us identify areas where our models may not be performing as well as we would like. And then we can go in and identify a specific correction. So for example, we identified that we weren’t delivering the granularity of information that we would want around precipitation in certain mountainous regions, so we’ve been able to identify and incorporate additional precipitation data sets that will improve the spatial resolution of our forecasts and improve our accuracy.

So this concept of hind-casting is great for our customers, but it’s also really important for us to continuously improve our modeling approach.

The second piece is we like to speak less about forecast accuracy and more about decision accuracy. Did we give you the right information at the right time to make a high ROI decision? So one customer for example said we moved our planting date because you said it was going to rain. It rained less than you said, but the other models didn’t say it was going to rain at all. So our forecast may not have been perfectly accurate, but we told you there would be a significant rain event that would be damaging to your planting; you moved your planting timing and you save significant yield as a result. That’s what I mean by decision accuracy.

AFN: What kind of progress are you making?

HG: We have worked with more than 56 partners in food, beverage and agriculture. With some of them, it is recurring business. For others, they might work with us to conduct a strategic analysis and then come back after three years.

But right now, we have no qualms about saying we are the market leader in this space for food and beverage spanning 40+ crops in 60+ countries.

We now have sufficient case studies across the value chain on how this partner saw, say, a 5-10% increase in sales because of our forecast, or this partner achieved X million savings in one season. As you look back, as in any technology adoption curve, you start with early adopters, then the early majority, then the late majority, then laggards. I think we are currently in the phase of early to late majority.

AFN: Gro Intelligence went under. How are you different?

HG: We are a problem first company rather than a technology first company. Often, when you’re a venture backed company, you are under constant pressure to grow sales and you end up signing up partners that stress out your resources and hiring more people to serve those customers until you eventually become a consulting organization with very low gross margins.

We began by focusing on specific use cases and products where we could find scores of partners who could benefit. The aim is to be a true SaaS play, where once you have developed the models, you don’t have to hire lots more people to service lots more customers, as it’s scalable.

I don’t know exactly what went wrong at Gro Intelligence, but what you do see a lot of in Silicon Valley is that because of the revenue pressure, you have to sign up any customer just to deliver, and it can end up handicapping you.

WK:  Our customers are already investing in trying to generate risk assessments using very costly consultants. They’re calling growers for anecdotal information, spending thousands of dollars on flights to visit regions, investing in expensive field trials.

We find through multiple tests with customers that their one to two year process gives them the same answer as our software. So we can save them one to two years of time, money, and not to mention speed to market. The ROI story is it will save you time, it’ll save you money. It’ll help you move faster.

AFN: Are new players coming into this long-range forecasting and insights space?

HG: We definitely see more companies coming in as volatility is becoming a bigger problem for supply chains. But it’s good for the market, because it signals to investors that this is the hot market, which we believe will help us when we go out and raise our next round.

AFN: How much money have you raised, and when might you achieve profitability?

HG: We’ve raised $38 million so far, we raised our Series B round last March, and I think we’ll be profitable by Q3 or Q4 of next year. We are a very lean team and we have developed an SaaS model whereby the your marginal cost of adding new customers is minimal. So so we think more in terms of how do we first acquire and prove value to, let’s say, 200 partners, and then we’ll think about going out and raising a Series C after that, probably by the end of next year.

AFN: What is next for the platform?

HG: In addition to food and ag companies, we’re also talking to governments focused on food and water security, and broader supply chain players in transportation, for example.

So we’re looking at things like what is the probability of Mississippi River being lower than normal? Should firms book carriages before [shipping] prices goes up? What is the hurricane risk at the ports at which food companies are importing or exporting crops so they can optimize their logistics?

And the last part is looking at what’s happening in end markets for our clients’ goods. Just as the supply of barley is linked to weather, demand for beer is also linked to weather. So how do these companies think about inventory and sales planning?

From a business standpoint, we started in the US, then moved to Europe, Japan, Australia and India. However, most of the companies we work with have global supply chains and sourcing regions, so they might be sourcing from Africa and South America, and our model can work at any location. The challenge is that in some regions, there’s not much data available.

So we have two product lines, one looks at what we are expecting in terms of yield and quality for crops and regions where we have enough historical data available, either publicly or through our partners.

But then we have a second product called risk outlook, which looks at indices tailored to specific crops that our partners can use in decision making. So as an example, we just launched a risk outlook model on coffee and cocoa. Because of the limited data set, the yield models on cocoa don’t work well so we came up with risk indicators such as heat risk during pollination or drought during harvesting and then we validate them by looking at how markets have reacted in the past.