Early adopters have been amassing and recording data for a quarter-century or so. How do we turn that data into information, and information into knowledge?
Early adopters of precision farming in the internet age have been amassing and recording data for a quarter-century or so. For a good while, these datasets seemed like the Beanie Babies of agriculture: Collectors weren’t exactly sure what so many numbers would be worth in the end, but they bet on value increasing over time.
While the market value of those little plush toys didn’t hold up, the mountains of data are poised to pay off, thanks to machine learning and artificial intelligence (the “other A.I.” in swine spaces) — and trained data scientists. One such professional, Caleb Grohmann, works with me in Carthage Veterinary Service’s (CVS) Data Analytics Division.
Having grown up on a swine farm, Grohmann has said he wants to help producers of all sizes maximize profitability by digging into the information they’re recording. We can do this by looking at a producer’s supply chain and figuring out ways to try to take advantage of opportunities to optimize the relevant parts of it for them using the data they routinely collect. This is where data analytics differs from data collection. Our overarching goal is to turn data into information, and information into knowledge.
Data capabilities are always followed by questions of security, and rightfully so. Our team and tools follow strict standards to keep client data safe when shared between companies. When sharing data, it’s vital to know: is there access to your internal systems, who can view it, how much visibility they have, who controls permissions, what layers of protection are in place, and whether access and actions are tracked.
A Model Case Study
To give one example, using advanced machine learning methods, Grohmann has designed a model that projects the optimum date to ship finished hogs to packers for the best possible profit.
This approach considers several data streams, including historical feed delivery and mortality patterns and site-specific weather conditions and their complex relationships with pig growth. Taken concurrently with real-time grain and lean hog pricing data, producers can use fully automated insights to make informed decisions regarding when and where to schedule loads to the packer.
Not only has this model been incredibly accurate to within an average of 2 to 3 days based on target load weight for our clients, it has shown that in the real world, had the client been able to sell loads on the predicted dates, doing so would have added anywhere from 30 cents to $3 more in net income per pig. It is, of course, not always possible for a real producer to ship or deliver on a specific date, but if they can, maximum added value could be realized.
“What makes these machine-learning models progressive compared to traditional approaches is that they are entirely data-driven, reducing the need to rely solely on assumptions or past practices,” Grohmann noted. “Producers have historically made sound decisions based on their experience, such as pulling the first load of pigs at 140 days on feed. However, data now allows us to see that individual herd cycles can differ, and adjusting to those differences can create additional value.”
Why the model works is because the A.I. program we are training can efficiently sort, learn and derive relationships from hundreds of different data sources that all work together to influence an outcome. It’s not enough, for example, to look at broad effects of season on growth rates or when you sold first cuts in the past — other factors and their interactions, such as actual temperature and humidity or long and short term feed delivery patterns, have proven to be stronger predictors of pig growth as well.
Producers have always had to rely on their experience and knowledge to find these complex information relationships, while additionally trying to synthesize their on-farm data with market prices, the political landscape and much more to make decisions. The fact is that no matter how proficient one person is with numbers, machine learning can parse all this information faster than any human with a spreadsheet will ever be able to.
“There’s no way we were able to do any of this stuff that we’re talking about when I was a kid,” Grohmann pointed out. “Being able to do this for producers is kind of my dream.”
Use What You’ve Got
It is possible for models to analyze years of information — such as herd-specific diets and rations, medications, protocols and more — to create useful predictive outcomes for an operation. We are also learning that we can pull historical data from a certain defined time period and, using advanced statistical techniques, begin deriving root causes of some on-farm issues to better propose future solutions. The more we do this, the more our producers can take prescriptive action, and we can directly quantify the outcomes of this action. In this way, the results of our analysis become even more refined, which translates into better solutions for producers in the long run.
Going back to Grohmann’s model’s impressive results with respect to projected income gains based on optimum marketing dates, it’s worth noting that it was able to achieve such accuracy using on-farm data already at our fingertips. So far, there has been no additional requirement for feed bin or other sensors to yield a workable solution.
This is not to say sensors don’t have their uses, nor that our models won’t be improved by incorporating such data in the future. But if you’re a producer who doesn’t use or isn’t ready to invest in additional sensors in barns, you will surely appreciate that you can still benefit from data analytics without that added expense.
And, it’s exciting for us to prove every day that there’s still plenty of untapped potential in the years of data already collected!
Live Link: What to Do With All That Data? We’re Finding the Answers
