~43%
Less production time wasted
BESTMIX Insights. Predictive analytics.
BESTMIX® Insights
Predict finished moisture and formulate to the true target.
BESTMIX Insights predicts finished moisture from your production line's own history and recommends machine settings to hold it on target. It is a supervised machine learning regression model, not a chatbot and not a language model. Its purpose is to remove the safety margin you are paying for on every ton.
The safety margin you are paying for
Finished moisture moves from batch to batch. A wet batch dilutes the nutrient concentration, so to protect the label on every batch, formulators pad the formula with a safety margin sized for the worst batch. That margin is paid for in raw material on every ton you run.
Insights predicts finished moisture, so the formula can be set to the true target instead of padded for the worst case. Steady moisture is the mechanism that lets the margin come down.
What kind of AI this is
Insights is a supervised machine learning regression model. It learns the relationship between the line's settings, the formula, and the conditions from the plant's own history, and predicts a number: finished moisture on the next run. It is not a chatbot and not a language model.
Naming the real technical category is our credibility position, because it tells you exactly what you are relying on.
How it works
Insights reads the plant's process data and formulation specifications together. Because BESTMIX holds the formula and the finished specification, it can relate what the line did back to what the formula requires. That link is the loop.
It predicts finished moisture for the next run, and it recommends machine settings from the same history to hold moisture on target. It does not monitor your process in real time, and it does not push changes to the line. It predicts and recommends, and your team decides.
Every prediction is explainable through SHAP values, so a formulator or engineer can see why a setting was recommended and audit it before relying on it.
Validated in production
In an aquafeed extrusion operation, the same process pet food kibble runs, Insights predicted finished moisture with an R squared of 0.92 and a mean absolute error of 0.51 percent. It was 50 percent more accurate than experienced operator predictions, and 87 percent of its predictions landed within a 1 percent tolerance.
Across two production lines, tighter moisture control produced savings of about $566,000 a year, most of it from formulation, once the team pulled back the safety margin.
These figures are from one validated implementation. They are directional for pet food and feed rather than guaranteed, and are converted from euro.
Some important figures we want you to know!
~50%
Reduction in nutrient swing
~$500K
Annual savings potential
ⓘ Validation note: These figures reflect results measured in specific customer implementations. Actual results vary with line conditions, product mix, formulation targets, and data quality.
FAQ's
Find quick answers to the most common questions about BESTMIX® Insights
Three things, tied to the batches that ran: the formulation, the process settings the line used, and the finished quality result you care about, such as finished moisture. Insights relates them to each other. More history improves accuracy.
No. A plant does not need new instrumentation to begin. It needs the records it likely already keeps, in a digital, exportable form, from a historian, MES, SCADA, batching system, or structured logs.
A model is validated before it informs any live decision, and it is checked against new production results as they accumulate. Ingredients drift, seasons change, and lines are adjusted, so a model is not set once and left alone. Accuracy improves as more production data comes in, and SHAP explainability lets a formulator or engineer see whether the reasons behind a prediction still hold.
No. A model built for you is trained on your data, for your use. We do not pool your data to train models we sell to others, and your data stays with you.
A proof of concept on one line, built on your own historical formulation, process, and quality data. It delivers a measured read on how much variability is costing that line, how much a model can recover, a joint ROI built from your own numbers, and a written feasibility report. Nothing touches live production.
This page answers the questions that most often influence a decision. Find answers to any additional questions in our FAQ.
Evaluate it on your own line
Because Insights is proven on your own data, a proof of concept on one line is the honest way to evaluate it. A model that fits your plant does not exist until it is built on your production history.
Important information
The savings figures on this page reflect results measured in a specific customer implementation, an aquafeed extrusion operation running two production lines, and are shown in US dollars converted from the validated euro result. Actual results vary with line conditions, product mix, formulation targets, and the quality and completeness of the available data. Nothing on this page is a guarantee, warranty, or promise of results.
Model performance figures, including R squared, mean absolute error, and tolerance rates, were measured on the validated aquafeed dataset described above. They are not a prediction of performance on any other line.
Capabilities described as in development or in validation are not commitments to deliver. Their scope, availability, and timing may change.
This page is provided for general information only. It is not an offer and does not form part of any contract. BESTMIX is a registered trademark. Copyright 2026 BESTMIX. All rights reserved.
Personalisation
See the current personalisation segmentation scoring or apply a manual segmentation to test