JBS delivers accurate, large-scale beef yield forecasts with AI
JBS S. A., Brazil’s leading meatpacking company, partnered with DSM Nutritional Products Brazil to predict and optimize beef production across more than 4 million animals from 5,204 farms. The real-world project tackled the complex challenge of forecasting beef carcass weight, maturity, fat, and quality by integrating animal traits, nutritional inputs, market prices, soil fertility, and climate data from 12 states. The team developed and tuned advanced machine learning models (Random Forest, Generalized Linear Regression, Neural Networks) in R, orchestrated via high-performance computing. The Random Forest model performed best for continuous variables, informing yield prediction and business decisions. Integrated predictions allow better resource allocation, increased sustainability, and strategic production planning. This approach is paving the way for more efficient and sustainable beef production at a national scale and demonstrates the business value of large-scale AI adoption in agriculture.
- Organization
- JBS S.A.
- Industry
- Agriculture
- Location
- Brazil
- Published
- April 2020
Reported outcomes
Strategic outcomes
Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks
- Customer
- JBS S.A.
- Provider
- Microsoft
- Maturity
- Production
- Linked source
- PMC (US National Institutes of Health)
Variable market prices, changing climate, and inconsistent soil attributes affected predictive accuracy and operational decisions
Primary read
Use case focus
Showing 3 of 3
- 1Yield prediction
- 2Resource optimization
- 3Large-scale machine learning
- Difficult to forecast beef yield and quality across a vast, heterogeneous supply chain comprising over 5,204 farms and more than 4 million animals.
- Variable market prices, changing climate, and inconsistent soil attributes affected predictive accuracy and operational decisions.
- Integrating diverse datasets (animal, nutrition, economic, soil, climate) from multiple sources was complex and labor-intensive.
- Optimizing sustainable growth without increasing environmental impact was a core business objective.
- Aggregated data on animal traits, farm nutrition, tech adoption, economic variables, soil fertility, and climate for each farm.
- Developed and compared machine learning models (RF, GLR, NN) using large-scale R workflows and high-performance computing.
- RF model applied for most accurate continuous yield prediction; GLR tuned for categorical carcass features.
- Used farm-matching algorithms and imputation to ensure complete, reliable training data.
- Enabled accurate forecasting of beef carcass weight (R2 = 0.61), maturity, fat, and quality at a national scale.
- Data-driven predictions now inform resource allocation, planning, and sustainability decisions.
- Production forecasting supports efficient use of feed, nutrition, and land, improving profitability.
- Demonstrates a scalable, real-world agricultural AI deployment.
Architecture
Data from over 5,204 farms—their animals, nutrition, economic variables, soil, and climate—was integrated using custom farm-matching algorithms. Machine learning models (RF, GLR, NN), managed and tuned in R, were trained on high-throughput computing infrastructure for rapid experimentation and validation. Data preprocessing included imputation, scaling, and validation routines. Predictions inform production planning and resource allocation at scale.
Sources & evidence1
- Customer explicitly identified
- Deployment status explicitly supported
- Independent source available
- Technical implementation details available
- Recent evidence check available
- Last evidence check: Jun 1, 2026.
The case's original source is still reachable.
- Cited source last checked Jun 1, 2026 — ok (0/1 broken).
Measures whether this deployment's public evidence persists — not whether the system is still in production.
AI-generated summary. Verify important details with the linked sources before relying on this case.
Explore related AI use cases
Was this useful?
Community
Comments
Loading comments...