AI in Pet Food Production: Cutting Through the Hype

Bea: Let’s start at the very beginning. When you hear ‘AI in food and agriculture,’ what does that actually mean to you?

Filip: The first thing I want to say is that AI is not a goal in itself. It’s a powerful technology that allows us to become more productive and more efficient. In practical terms, it means using the data you already have to support better decisions. In manufacturing, that usually comes down to predicting a key outcome - say, moisture or density-  and recommending the settings most likely to keep you within spec. The humans still decide. AI just gives them sharper information to work with.

Bea: There’s a lot of fear floating around. What’s one misconception about AI in this industry that you’d like to clear up right away?

Filip: That AI equals automation that replaces people. That’s the Hollywood version. The reality is much more practical. Most of the value we see today is in decision support: making expert judgment more consistent and easier to repeat across shifts, across teams, across plants. It’s not about removing people from the equation. It’s about making sure the best thinking doesn’t walk out the door at the end of a shift.

Bea: So where is AI actually creating real value today? Not in theory, but in practice.

Filip: In operations and quality. Early warning systems, fewer surprises, and better root cause clarity. Especially in processes like extrusion, where dozens of variables interact in ways that are hard for any single person to track in real time. Another example we’re proud of is our AI agent for documentation. It helps teams get to the right information faster, without digging through folders or asking around.

Bea: Now let’s go deeper into production. Where can AI bring clarity in complex workflows like extrusion, drying, and coating?

Filip: Clarity comes from linking data that normally lives in separate places: line settings, QC checks, and product context. Once you connect them by time and product reference, you start to see which variables actually drive moisture, density, and other quality outcomes in your specific process. That’s where a lot of internal debates stop being opinion and start being evidence. It’s surprisingly powerful when people can finally point to data instead of gut feeling.

Bea: Operators are already juggling a lot. How can AI reduce the cognitive load when dozens of parameters are shifting during a run?

Filip: By narrowing the field. No operator can realistically watch fifty signals at once. Nobody can. A good model highlights the few variables that matter most right now, in this moment, and suggests a small set of adjustments that are most likely to bring the product back inside tolerance. Think of it as having a very consistent second set of eyes. One that never gets tired, never gets distracted, and doesn’t change its approach depending on who’s on shift.

Bea: Most companies already have MES, QMS, LIMS, ERP systems in place. What gaps can AI close quickly, without ripping everything out and starting over?

Filip: Two gaps come up almost every time. First: time alignment and context. Second: turning specs into actionable guidance. Many systems store the data perfectly well, but they don’t connect a QC result back to what the line was actually doing minute by minute. And they don’t translate a tolerance range into a recommended action. AI projects often start by building exactly that bridge, not replacing the systems you already have, but making them talk to each other in a way that’s useful on the floor.

Bea: Here’s the big one for quality teams: can AI actually predict out-of-spec risk before it happens and reduce rework?

Filip: Yes! If you define ‘good’ clearly, using your tolerance ranges. Once you’ve done that, you can predict drift before you get a failed lab result, and you can prioritize checks where the risk is highest. In our extrusion work, for instance, the goal was to recommend settings that keep moisture and related quality parameters stable, so the team spends less time chasing problems after the fact and more time running a smooth line.

Bea: A lot of people worry about ‘black box’ AI - decisions nobody can explain or trace. How do you design systems where AI strengthens traceability rather than undermining it?

Filip: Three rules. First, keep it advisory: the operator can accept, adjust, or ignore any recommendation. Second, make the inputs visible, so you can always explain what the model ‘saw’; which variables, which recent trends. Third, log the decision. That means QA can review what was recommended and what was actually done, tied to a specific lot or run. Follow those three rules and you end up with something that’s not only useful. It’s audit-friendly.

Bea: Looking at the next twelve months: what does practical AI look like for production, quality, and regulatory teams?

Filip: Smaller, targeted use cases. Moisture and density stability. Early warning systems. Standardized operator guidance that doesn’t depend on who happens to be on shift. Plus better data contracts between MES and QC systems, so you can actually trust the foundation you’re building on. We’re not talking about full autonomous control, not yet. We’re talking about repeatable decision support that reduces variation and makes life easier for the people doing the work every day.

Bea: Zooming out a bit further. What AI capability do you think will be standard in this industry within the next three years?

Filip: AI-driven early warning and operator guidance for key quality parameters, integrated into day-to-day workflows. Think ‘predict drift and recommend actions’ rather than ‘robots running the plant.’ It won’t be exotic anymore. It’ll just be part of how a well-run operation works.

Bea: Where’s the biggest untapped opportunity, the place where AI could create disproportionate impact?

Filip: Reducing variability. Not by adding more controls, but by making best-practice decisions repeatable across sites, across shifts, and across new staff. That’s where most of the hidden cost lives in this industry. When your best operator retires or transfers, the knowledge shouldn’t leave with them. AI can help lock that expertise into the way the line runs every day.