100% model coverage across 12 production lines - without adding headcount.
The Challenge
Machine learning had spread across Nordic Manufacturing Group's plants the way it does in most industrial companies: organically. Predictive maintenance models in one plant, vision-based quality inspection in another, demand forecasting run by a central team, each deployed by whoever needed it, on whatever infrastructure was closest. There was no central inventory, no shared validation standard, and no way to answer a simple board-level question: how many models are running in production, and who is responsible for each one?
The Solution
Thndr AI was connected to the company's deployment pipelines and edge management platform, and its registry auto-discovered the full model estate — including models no central team knew existed. Within three weeks every model had an owner, a risk profile tied to the production line it served, and drift monitoring linked to quality KPIs. Governance became a property of the deployment pipeline itself, not a separate process the plants had to remember to follow.
Discovery Before Control
The first result wasn't a policy — it was a map. Auto-discovery surfaced 14 models running in production that didn't appear in any central documentation, including a quality-inspection model that had been retrained locally by a plant team eight months earlier. Each shadow model was registered, assigned an owner, and risk-tiered before any enforcement began. Governing what you can't see isn't governance; the inventory had to come first.
Governing at the Edge
Most of the estate runs on factory-floor edge hardware with intermittent connectivity — a deployment pattern most governance tooling silently assumes away. Policy checks were moved into the release process, so a model version is validated before it ships to a line, and every edge deployment reports its exact model version and configuration back to the registry when connectivity allows. Rollouts became versioned and reversible per production line.
Scaling Without Headcount
The central ML platform team is four people, and that didn't change. Policy templates per model class — vision inspection, predictive maintenance, forecasting — let plant engineers onboard their own models through a self-serve flow, while the central team sets the guardrails instead of reviewing every deployment. Governance capacity now scales with the template library, not with the team's calendar.
Results
Every production model is registered, owned, risk-tiered, and monitored across all 12 lines.
Auto-discovery surfaced models absent from all central documentation and brought them under governance.
From first pipeline integration to a complete, continuously maintained model inventory.
The four-person ML platform team governs the full estate through templates and self-serve onboarding.
See what Thndr AI can do for your team
Talk to our team about your specific AI governance challenges.
