137. What is digital transformation?

“Software is Feeding the World” is a weekly newsletter about technology trends for Food/AgTech leaders.

Greetings from the San Francisco Bay Area.

SFTW on the web

The Curiosity Flywheel

A few weeks ago, I published a framework called the “The Curiosity Flywheel” and issued a call to action to folks working in the food and agriculture industry to provide their independent perspectives.

There is a need for unique, and objective voices in the industry, which investigate, explore, explain, and imagine a better future. The industry is large and we need to look at problems from different perspectives to understand problems, come up with creative solutions, and make a difference.

The Curiosity Flywheel is definitely worth a read. (On the SFTW blog)

LLMs or CCAs

Are LLMs (large language models) going to replace many human activities? LLMs can already clear certain certification exams. Will LLMs compete with certified crop advisors (CCAs)? I looked at this question in edition 135 of the newsletter and concluded the piece with the following.

How might technologies like robotics, automation, and LLMs augment human intelligence, intuition, and experience and how will it change what it means to be involved with agriculture and farming?

It is CCAs and LLMs not CCAs or LLMs.

What is digital transformation?

As a follow up to the Bayer Crop Science Innovation Summit (June 20, 2023), I did a short one-on-one session with Jeremy Williams, Head of Digital Farming Solutions for Bayer Crop Science. (Here are my original reflections on the Bayer Crop Science Innovation Summit on June 20th, 2023).

Our discussion focused on aspects of digital transformation, cultural challenges, Bayer’s engagement with the startup ecosystem, and what is he most excited about from a technology standpoint? (Full disclosure: I worked at The Climate Corporation (now called Climate LLC), part of Bayer Crop Science from April 2017 to April 2020).

Note: My questions are in bold, and Jeremy William’s responses are in italics. Both are paraphrased and summarized. I have provided additional commentary, which was not part of the conversation in regular font, wherever necessary.

What does Bayer mean by digital transformation?

Digital transformation will help us redefine how our customers interact with our products. Our product definition will include our physical product, our digital products and everything else that goes with it. It will not be about just selling a bag of seed or a gallon of herbicide. This will be aided by digital tools and technologies.

Related reading: Commentary from Thomas Siebel, CEO of C3.AI on the five stages from digital transformation.

True digital transformation only happens when organizations reach clarity—yes, acceptance—on what digital transformation is about. Rather than flashy new programs, it requires focus on tractable problems that, when solved, deliver measurable value, and methodically applying this approach across the enterprise.

This brings up a classic case of Innovator’s dilemma and a potential culture clash between the digital and the traditional side of the business. Is Bayer set up organizationally to achieve digital transformation?

You are right about the Innovator's dilemma. Monsanto set up a team (Climate) but the digital and physical side of the business have to work together as one team. It is not just about selling physical products, but it is about how we structure teams, how we see and pursue funding for different resources, how we provide autonomy and agency to different teams to make decisions, how we enhance our understanding of current & future value drivers for our customers, how we prove customer desirability, and how we iterate and improve on new business models. We are well set up to do so and it is a long journey.

My reflections: Organizational changes and digital transformation are extremely hard. I would like to highlight two books, which talk about the process of digital transformation and organizational changes required.

a. Humanocracy by Hamel and Zanini.

Humanocracy will show you how to launch an unstoppable movement to equip and empower everyone in your organization to be their best and to do their best. The ultimate prize: an organization that’s fit for the future and fit for human beings.

b. The Digital Matrix: New Rules for Business Transformation Through Technology by N. Venkat Venkatraman.

The Digital Matrix will help you understand the three types of players that are shaping the new business landscape; the three phases of transformation that every firm will encounter on its journey to business reinvention; and the three winning moves that will ensure your company’s success along the way.

I had discussed the issue of culture change with Mark Young, ex-CTO of The Climate Corporation in edition 52 of the newsletter.

I noticed Bayer Crop Science executive team does not have anyone with pure digital or technology experience.

As far as pure technology experience goes, the beauty is that our current leadership team at Climate brings that expertise. Our current head of engineering has a pure software background, we have many folks with great product experience, our new product leadership brings entrepreneurial as well experience of working at companies like CNHi.

Why are you publishing “subscribed acres” numbers only for FieldView?

FieldView is seen as a SaaS (Software as a Service) or DaaS (Data as a Service). Yes there is digital value. FieldView does drive customer loyalty, growers consistently give us high NPS ratings. But we realize the intrinsic value of FieldView might be larger. We believe it will help us bring new offers, new carbon business and many other opportunities.

As far as subscribed acres go, we internally track subscribed unique connected acres, actual engagement, engagement beyond planting and harvest events, and try to understand where and how we provide value. Due to this penetration of FieldView is important, and we have a lot of internal debate on which are the right metrics to publish externally.

How do you see Bayer leaning into the startup ecosystem? Having worked at a startup in the past, startups are often weary about big corporations coming into their space. For example, you have a long partnership with Leaf Ag, which is focused on interoperability, and you have talked about building your own capabilities for interoperability. Also, startups are often worried that once they get acquired, they will be lost in the bureaucracy of the big corporation. For example, we have not heard much from Combyne since their acquisition by Bayer.

We realize we have to collaborate with the ecosystem to create value for our farmers and other customers worldwide. As you know, over the last few years we have signed up with 80+ partners within the ecosystem.

We want to support startups, and we want to collaborate with innovators in the industry. We continue to have ongoing conversations to clearly define how our partnerships can create value for our customers, and are beneficial to us and our partner at the same time.

Your point about startups worrying about getting lost in the bureaucracy of a big company is real. We have been thoughtful in providing independence and autonomy to new companies we bring into the Bayer family. For example, Alain (CEO of Combyne) reports to me directly. It will give the Combyne team a direct link into our digital farming efforts.

My comments: This is a challenge often faced by large corporations. For example, in the tech world, the same company can be a competitor and a partner at the same time. Due to germplasm licensing, this is a problem not unknown to the agriculture industry, though Leaps by Bayer has not invested in Leaf Agriculture.

What is the clearest evidence you have seen of digital helping with Bayer's transformation and where do you think you have a thesis but still don't know exactly how to get there?

We have seen a clear impact on customer behavior in the case of purchasing branded corn or soy seed, for FieldView users. It is a clear, significant enterprise value coming from our digital efforts. Digitally enabled customers start to buy more of our physical products.

On the other side, banks are interested in getting granular data to make more precise loans. There are opportunities in the carbon and finance space, through our FieldView partnership ecosystem enabled by API connectivity with FieldView. There is a growth opportunity for us, our partners, and most importantly for farmers.

My comments: This is part of the adjacent value Bayer talked about during the innovation summit, which Bayer hopes to create and capture in the future. It will obviously have to follow all the farmer data privacy rules and protocols.

Finally, what are some of the things you are most excited about for the future, when you think about the intersection of agriculture and technology?

I was going to say LLMs (chuckles), but I am still very much excited about providing more transparency within food and agriculture systems. I believe we can do much better in this area.

Recall is risk tolerance

Dr. Murat Unal, Ecosystem Intelligence Officer - SONEAN, posted an infographic from their de-weeding organizations report.

There are currently, based on our AGCUMEN Ecosystem Intelligence, roughly 1200 organizations in the world providing mechanical or precision technology solutions, offering alternative ways to combat weed in a more sustainable way, by getting rid of pesticides.

Weed is either addressed mechanically (ideally ending up as biomass) or using robots that un-pluck weed with actuators or apply all sorts of treatments (electricity, laser, plasma, hot water, selective pesticides, and many others)

I was surprised that the number of organizations mentioned is as high as 1200. It sounds more like the biological market. It is not surprising though, that most of the organizations are in Europe, due to the EU green deal, with specific targets to reduce chemical usage etc. It is not only a product, and technology challenge, but also a messaging challenge for these organizations to communicate their value proposition to the end farmer.

A common measure used by precision de-weeding technology companies is reduction in herbicide volumes or input costs or weed management costs (emphasis by me). Other benefits mentioned include reduced crop stress, biodiversity improvement, increased crop yield etc.

1. Solix Spray AG Robotics platforms says the following:

In one of our clients' properties, we observed a

97% decrease in herbicide volume during initial trials, particularly in areas heavily infested with weeds

. This underlines our technology's ability to deliver precise applications, precisely where needed."

The Solix sprayer carries out a precision application of herbicides, allowing efficient control of weeds.

2. GreenEye reported data from their trials,

Average herbicides reduction of 78%

from all spraying applications (GoG & GoB)

GoG = Green on Green (spraying weeds present in a crop)

GoB = Green on Brown (spraying weeds, but no crop is present)

GreenEye also talks about a Recall rate (true positive) of 95.7% in multiple field conditions from all spraying applications (GoG & GoB)

(More on recall rates in a minute)

3. John Deere’s See and Spray Ultimate,

See & Spray Ultimate can reduce non-residual herbicide use by more than two-thirds by target spraying weeds.

(Note: 1. Results based on internal John Deere strip trials in corn, soybeans, and cotton in Iowa, Mississippi, Texas, and Illinois, in typical growing conditions, with varying weed size, crop canopy, and field conditions, using targeted spray of non-residual herbicide only, and using current software/algorithm at time of trials. Results vary based on crop; for details see JohnDeere.com/SeeandSpray. Weed-control results based on dual-tank operation, adding an additional herbicide that could not be added to an existing herbicide mix in a single tank. Individual results will vary.)

4. Carbon Robotics LaserWeeder,

The LaserWeeder can kill up to 99% of weeds, weed up to two acres per hour, and eliminate up to 5,000 weeds per minute.

The technology detects and eliminates weeds much sooner than possible with the human eye, killing them earlier in their lifecycle without harming the crop or soil. It can operate on over 40 crops and create and deploy new deep-learning crop models within 24 to 48 hours. Growers utilizing the LaserWeeder have reduced weed control costs by up to 80% and see a return on investment in one to three years.

5. Ecorobotics ARA,

Currently, ARA works with 13 herbicides, fungicides and pesticides on corn, cotton, spinach, onion, rapeseed and several other crops. Ecorobotix says ARA can reduce input costs by 70–95%

in addition to “increasing crop yields” and eco benefits such as preserving biodiversity.

What goes into these savings calculations?

The most common denominator used to calculate cost savings is the cost of broadcast spraying of herbicides and other chemicals. This assumes 100% spraying across the entire field.

If 10% of your field is covered with weeds, and using machine learning models you accurately identify and hit each of the weeds, then you would save 90% of your herbicide cost. (The logic is a bit different for mechanical or laser based weeding, as the numerator is a bit different).

But the reality is a bit different. Any machine learning model is not 100% accurate, and it can often miss real weeds and identify plants which are not weeds as weeds.

Enter precision and recall. This section is about to get geeky, but it is important to understand. From Wikipedia,

Consider a computer program for recognizing dogs (the relevant element) in a digital photograph. Upon processing a picture which contains ten cats and twelve dogs, the program identifies eight dogs. Of the eight elements identified as dogs, only five actually are dogs (true positives), while the other three are cats (false positives). Seven dogs were missed (false negatives), and seven cats were correctly excluded (true negatives). The program's precision is then 5/8 (true positives / selected elements) while its recall is 5/12 (true positives / relevant elements).

Precision can be seen as a measure of quality, and recall as a measure of quantity. Higher precision means that an algorithm returns more relevant results than irrelevant ones, and high recall means that an algorithm returns most of the relevant results (whether or not irrelevant ones are also returned).

For the case of identification of weeds using a machine learning model, one can construct a simple 2 x 2 matrix.

Recall is risk tolerance

In the matrix above,

recall = True Positive / (True Positive + False Negative), which indicates what proportion of the actual weeds were identified as weeds by the model.

If you want to make sure you identify most of your weeds and take care of them, then recall becomes an important metric, as you want it to be high.

A recall of 0.95 means, you are okay to leave 5% of the weeds unidentified and hence untreated.
A recall of 0.98 means, you are okay to leave 2% of the weeds unidentified and hence untreated.
A recall of 1.0 means you want each and every weed in your field identified and treated - a weed free field.

Recall is a measure of your risk tolerance. The lower your risk tolerance for presence of weeds in your field, the higher should be the recall number.

The false positive rate is important as your machine learning model will recommend spraying, even though it is not a weed. In the case of precision spraying of weeds, the false positive indicates wasted chemicals as the model sprayed on plants which were not weeds.

Based on these statistical concepts, it is relatively easy to calculate the amount of chemical sprayed, if we assume what percentage of the field is weedy, also called weed density (This assumes uniform distribution of weeds in a given field for simplicity, which might not be accurate.)

Chemical sprayed = Weed density x recall + (1 - weed density) x false positive rate.

What is the intuition for this formula?

First term: Weed density is the true number of weeds in the field, and recall indicates what percentage of actual weeds were identified by the model. The model will recommend to spray for weed density times recall rate.

Second term: False positive rate indicates when the model recommends to spray even though it is not a weed. The model will recommend to spray the false positive rate times proportion of the field with no weeds (or 1 - weed density).

It is important to note that as the recall rate goes up, the model becomes more sensitive and so its false positive rate might go up. Based on this, one could draw up a simple chart of how much chemical needs to be sprayed compared to a full broadcast spray for different recall rates, weed densities, and assumed false positive rates.

As you can see, chemical usage (and by extension chemical savings) is dependent on many factors, and does not match with the weed density in a given field.

There are many other factors which will go into your actual savings, including factors like:

How many acres do you spray?
Chemical costs on a per acre basis - fixed and variable
Operational metrics like speed, number of hours worked per day, time to refuel and refill your tank

Most of these are more relevant for commodity row crops, but similar principles can be applied to other cropping systems as well.

As a farmer, retailer, researcher or service provider, it is important to know these concepts, so you can evaluate the value of these solutions based on your unique situation.

Show me the money to save some money

All of the new tech mentioned above, along with robotics can be significant investments. The cost of a robot or a sprayer can be upwards of $ 100K depending on the tasks performed by the robot. This is not an unsubstantial investment.

In the case of specialty crops, labor costs continue to rise. In the case of commodity row crops, it is difficult to find human resources, and there is a desire to reduce input costs, and also pressure from regulators to reduce chemical usage.

Oftentimes large investments are cost prohibitive for many farmers. There are different financing models being explored to ease the rate of

Pay-per-use model offers flexibility through subscriptions or a usage fee. In a pay-per-use model, the farmer only pays for usage of the machinery. It lowers the barrier to adoption, and converts a capital expense into an operational expense for the farmer.
Similar to automobiles, another model is an operational lease using fair market value leasing. An FMV (Fair Market Value) lease is a lease that does not define a fixed purchase price at the end of the lease term. Instead, the vehicle may be purchased for the fair marketing value at the end of the term.

Fair market value is defined as the price of a piece of equipment if the equipment was sold at an “arm’s length,” between a willing buyer and a willing seller under similar terms and conditions. The FMV lease gives access to the latest equipment versions at a fixed monthly price. The farmer has the flexibility to extend the lease or purchase the equipment at fair market price, or return or upgrade it to the latest equipment.

Given many of the crop protection offerings are provided as a service by the local retailer to the farmer, the retailer can decide to outright buy the equipment or go for lease option. (similar to how airlines lease aircrafts from the aircraft manufacturers and provide a service to passengers).

Precision Agriculture Survey for Ag Dealers

CropLife provided a preview of their 2023 precision agriculture dealership survey. Some of the key highlights were,

Continued growth in tech to optimize operations.
More dealers offer drone imagery but fewer will offer satellite or aerial imagery. (19% to 57%)

Variable rate fertilization and seeding are cooling off.

Dealers use of on-farm data for decision making has plateaued.
Dealers are still excited about crop input applications with drones.
One of the biggest areas of growth anticipated by retailers has been high-tech pest management, for instance variable herbicide rates based on soils or sprayers responding to the presence of weeds.

These surveys are useful, though in a limited way. Ultimately what matters is the adoption of technology which creates value for the end customer i.e. the farmer. Hopefully, the dealers are making investments in technologies and tools, where they see and/or anticipate real traction and value creation for their farmer customers.

What do you think?

If you like “Software is feeding the world”, please share with a friend.

If you don’t mind answering 3 questions anonymously (2 are optional), I would love to get your feedback.

About me

My name is Rhishi Pethe. I lead the product management and technology delivery teams at Mineral, an Alphabet company. The views expressed in this newsletter are my personal opinions.