01 · Ongoing · Agentic AI · Global scale-up
Smart Self-Improving Lines: scaling one fix to many lines, in parallel
38%of common failures, across 8 problems
$500Kper labeller instance, 20+ lines
- The goal
- Take proven solutions for our most common losses from one line to many lines in parallel. Single-line fixes stay hard-coded and specific, so we bucketed the top losses into two scalable themes: auto-centerline closed loop and equipment performance.
- The numbers
- 8 common problems (labellers, filling, capping, glue and more) that account for 38% of common failures. Labellers alone scale to 20+ lines at about $500K per instance; pilot and demo lines set the exact average value per instance before scale-up.
- How I did it
- Run a pilot and a demo line per problem, then capture the solution in a Common Best Approach (CBA) template for fast reapplication. An LLM agent, the AI Use-Case Finder, helps sites find the right use case and hands them the CBA: team charter, inputs and outputs, model artifacts.
- Why it mattered
- These are the biggest shared sources of loss across the network, so every reapplied solution compounds value instead of starting from zero.
- Timeline
- Started in June 2025 (FY26). All lead lines and CBAs are in place; next is parallel reapplication across lines and sites.
Lesson learned “A pilot isn't a product until someone else can run it without you. Write the playbook while the knowledge is still fresh.”
Agentic AILLMClosed-loop controlScale-up strategyKnowledge managementStakeholder alignment
