Meridian AG forecasts yield per field, per week , not per county, per season.
A custom TensorFlow model trained on weather, soil, and satellite data. Agronomists get actionable per-field forecasts updated weekly, not regional averages from a government report.
Regional forecasts were too coarse to act on.
Meridian AG's agronomists were making input and harvest decisions based on regional government yield estimates, averages across thousands of hectares that bore little relation to individual field conditions. Soil variability, microclimates, and drainage differences meant the forecast was often wrong by the time it reached the field level. Planning was reactive, not predictive.
A model that sees each field as its own data source.
Weather, soil, and satellite inputs combined into a per-field sequence model. Updated weekly through the growing season so agronomists always have a current forecast, not a pre-season guess.
Data sources unified
Five years of per-field yield records merged with daily weather station readings, soil composition lab results, and NDVI satellite imagery into a single training dataset.
Feature engineering
Growing degree days, rainfall deficit windows, and soil drainage indices engineered as predictors. Domain input from Meridian's agronomists shaped every feature decision.
TensorFlow model trained
Sequence model trained on per-field historical trajectories. Validated on two held-out growing seasons before any production deployment.
Forecast API deployed
REST endpoint updated weekly as the season progresses. Agronomists query per-field forecasts directly from their existing planning dashboard.
Agronomists plan on data. Not instinct.
Forecast accuracy
Improvement in per-field yield prediction vs. the regional baseline used before deployment.
Data to deploy
From first data handoff to a live forecast API serving the agronomist planning dashboard.
Forecast cadence
Model re-runs every Monday with the latest weather and satellite data for the current growing season.
Field coverage
Every managed field in the Meridian portfolio now has an individual weekly forecast, not a regional average.
The smallest stack that solved the problem.
Same pattern. Different industry.
VELOPORT
Gradient boosting model forecasts weekly SKU demand from 3 years of sales history, seasonal signals, and external event data, replacing a manual Excel process.
NORTHBOUND LOGISTICS
Agent reconciles bank statements against 40k+ ledger entries every morning. Exceptions flagged. Report delivered. Finance team reviews in minutes, not hours.
CREDEX FINANCE
Churn scoring model flags at-risk accounts 45 days before cancellation and triggers automated retention sequences, before the customer picks up the phone.
Replace gut feel
with a model.
30-minute call. We'll audit your data and tell you exactly what is and isn't ready for a custom ML model, free, no deck.