AI in Pet Food Operations: 10 insights from IPPE 2026
At the AFIA Pet Food Conference held during IPPE 2026 in Atlanta, a panel dug into how AI is being put to work across pet food operations - from formulation to extrusion to marketing. Host Eric Altom, Ph.D. (Balchem) led the discussion with Hana Bieliauskas (Inspire PR Group), Johanna Ballesteros (SWARM Engineering), Filip Snauwaert (BESTMIX Software), and Tara Zedayko (Ollie Pets).
If you run plants, supply chains, or product lines, here are the insights that matter - and how to act on them.
1. Data readiness shouldn't delay adoption
The myth: "we're not ready - our data isn't perfect." The reality: waiting on perfect data just delays value. As Ballesteros put it, postponing for data "readiness" pushes back results you could be capturing now.
- Action: Pick one use case. Start with available data and define what "good enough" looks like. Improve data quality as you deploy.
2. AI amplifies expertise rather than replacing jobs
AI scales the judgment you already have. Zedayko noted there are fewer than 100 veterinary nutritionists in the U.S.; AI helps extend that scarce expertise across product development and support without replacing it.
- Action: Identify where expert time is the bottleneck (formulation reviews, spec checks, root-cause analysis) and deploy AI as a force multiplier.
3. Extrusion process optimization delivers measurable results
In a proof-of-concept, Snauwaert reported faster startups on new products, up to 33% fewer reworks, and 50% less moisture swing. That improvement shows up as throughput gains, scrap reduction, and tighter spec adherence.
- Action: Start with one line and a narrow KPI set (e.g., startup time, moisture variability). Instrument, model, and iterate weekly.
4. Trust and accuracy are critical barriers
Pet food operations have zero tolerance for prediction errors that could touch safety or quality. Snauwaert underscored the stakes: a single error can halt production or compromise product health. Data security and protecting proprietary recipes are equally vital.
- Action: Build guardrails - validation gates, versioned models, human-in-the-loop approvals, and read-only access to sensitive formulas.
5. Start with clear, focused use cases
Broad AI initiatives stall. Narrow ones pay back. Target problems where expert judgment is scarce or variability is expensive: production planning, quality assurance, or formulation optimization.
- Action: Define the metric that matters (OEE, rework rate, forecast accuracy), baseline it, then run a 60-90 day pilot.
6. System integration solves data silo challenges
AI only helps if it can see across formulation, production, QC, and inventory. Ballesteros emphasized integrating with existing systems so optimization isn't trapped in one step of the process.
- Action: Map your data flow across ERP, MES, QMS, LIMS, and formulation software. Prioritize a few high-value integrations first.
7. Domain-specific AI ensures safety and compliance
General chatbots are risky in regulated environments. A domain-scoped approach is safer. Example: AAFCO's "Ava" assistant draws only from AAFCO source materials, ensuring accurate guidance and attribution.
8. Simple applications offer quick entry points
Not every win requires a plant-wide rollout. Start with report summaries, content ideas, consumer sentiment from reviews, and pattern detection in feedback. Quick wins help teams build confidence before tackling formulation or scheduling optimization.
- Action: Set a 30-day sprint: pick two low-risk workflows, measure time saved, and share results to build momentum.
9. Marketing applications require ongoing training
Bieliauskas pointed out many teams aren't trained well on modern AI tools yet. That's fixable. Clear prompts, review checklists, and brand guardrails are the difference between average outputs and production-ready assets.
- Action: Provide short, role-based enablement on prompts, review standards, and compliance. Consider a focused certification for ops-led automation: AI Automation Certification.
10. Photo-based health screening enables personalized nutrition
Ollie processes photos of stool, skin/coat, body condition, and teeth to deliver personalized plans - more than 100,000 images from 54,000+ dogs. That consumer signal can inform product iteration, service design, and proactive support.
- Action: Treat image inputs like any other QC signal: standardize capture guidelines, label outcomes, and close the loop with product and CX teams.
Operations playbook: get started in 30-90 days
- Pick one use case with a clear KPI and a single process owner.
- Map data sources and minimum integrations required (keep scope tight).
- Set guardrails: validation steps, audit logs, and human approvals.
- Pilot on one line or product family; review results weekly.
- Document wins and gaps; scale only after you hit the KPI target.
If you want structured upskilling paths for operations, planning, and QA teams, explore role-based options here: AI courses by job.
Your membership also unlocks:
