Accuracy before novelty
A formulation system has to be correct. BESTMIX does not deploy AI unless it meets a measured accuracy threshold. The cost of a wrong recommendation is a failed batch, a missed specification, or a customer complaint.
The closed loop in pet food production
Close the gap between your formula and your line
For years the formula and the production line have sat on opposite sides of a gap, and quality results come back too late to act on. To cover the delay, formulators pad the formula with safety margins and repeat trials. BESTMIX closes the gap. What the line does and what the lab measures flow back into the next formula, so the next one starts from better information. AI has a role in that loop, and it is the clearest sign the loop works.
Our approach and operating principles
Know exactly what you're using
Statistical analysis is not AI. Pattern matching is not AI. Each capability is described by its real technical category, whether AI, predictive analytics, or data analysis, because each requires different validation and earns a different level of trust.
Validated on production data
Many AI claims in food and feed are validated on small lab datasets or simulated runs. BESTMIX predictive models are validated on years of real production data, across many lines and equipment types, before they inform any live decision.
What is already available today
BESTMIX InsightsInsights predicts finished moisture from your production line's own history and recommends machine settings to hold it on target. Because BESTMIX holds both the formula and the finished specification, it can relate what the line did back to what the formula requires. That link is the loop. In technical terms it is a supervised machine learning regression model, not a chatbot and not a language model. |
BESTY, the knowledge assistantBESTY answers software and setup questions in plain language, drawn only from the current BESTMIX documentation. It gives every team member fast, consistent access to product knowledge without searching manuals or waiting for a colleague. Because it is a closed system, it never sees your formulas or production data. It is included with every BESTMIX subscription. |
Where generic AI falls short
A general AI or dashboard project
A general AI consultancy can build a dashboard from plant data and even predict outcomes. But the prediction stops there because there is no formulation engine to act on it. A formulation vendor, meanwhile, holds the formula but never sees what the line actually produced, leaving nothing to learn from.
What BESTMIX does differently
BESTMIX holds the formula, the quality specification, and the process data, so what the line produces can feed back into the next formula. Without a closed loop between R&D, production, and quality, a model cannot learn from what happened on the line or improve the next production run.
The difference is not simply the model. It is whether production results can improve the next formulation.
AI accuracy depends on the data architecture beneath it
A predictive model needs three things, tied to the batches that ran. All three usually already exist in systems you have.
The formulation
What each batch was supposed to be. From your existing formulation system, or an export from it. You do not need to replace it to start.
The process settings
What the line actually did. From a historian, MES, SCADA, or batching system that already records it.
The finished quality result
What came out, such as finished moisture. Usually already held in your LIMS or your quality records.
Getting started
A proof of concept: Start with one production line
One approach, across your industries
The same principle, predicting a finished quality result that varies from the line's own history, applies wherever a process runs on digital data.
| Process | Industries | Predictive outcome |
| Extrusion | Pet food, aquafeed | Finished moisture |
| Pelleting | Feed | Pellet durability, energy consumption and throughput |
| Any line with digital data | Meat, food | A finished quality result that varies |
Talk to our AI team
Tell us about one line and we will tell you plainly whether it is a fit for a proof of concept, and if it is not yet, what would have to change.
FAQ's
Different tools for different jobs, and each is labeled. BESTY is a retrieval-augmented language model that answers only from BESTMIX documentation. BESTMIX Insights is a supervised machine learning regression model that predicts a production outcome such as finished moisture. The Recipe Analyzer is data analysis, similarity scoring across structured data, not machine learning. Naming the real category is how you know what you can rely on.
No. Our systems are closed. BESTY is trained only on our documentation and never sees your formulas. A predictive 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 if a project ends, your data stays with you. We suggest asking any AI partner the same three questions: who owns the data, who owns the models trained on it, and whether your data is ever used to build something for another company.
No. A proof of concept can work from your own exported data, even from another vendor's system, so nothing has to move to begin and nothing changes in production. Moving formulation and quality into BESTMIX Recipe Management and Quality Control makes the loop stronger, but that is a separate, later choice, not a condition for starting.
No. Nothing is pushed to your process automatically. The model predicts and recommends, and your team decides. The role of AI here is to give a good formulator a faster, more consistent read on what the line will produce, so their judgment is applied to a better picture. The expert stays in control.
Each capability is labeled by whether it is live today or in development, not by date. What is live is live now. For anything in development, your BESTMIX representative can discuss specifics, and current customers follow new capabilities through the in-application release notes.
This page answers the questions that most often influence a decision. Find answers to any additional questions in our FAQ.
Personalisation
See the current personalisation segmentation scoring or apply a manual segmentation to test