Dental manufacturer's $25K AI vision POC avoids $150K+ failed pilot, pivots to Andon system

An FDA-regulated dental products manufacturer spent significant time daily on mandatory dual-signature verification of custom 5-digit expiry date stamps on syringes. With 4 changeovers per day averaging 30 minutes each, the 100% manual paper-based process was a costly compliance bottleneck with no automation in place.

A $25,000, 3-month POC deployed a specialized lightbox with a high-resolution camera, a locally-processed neural network trained on 500 syringe samples, and JSON-over-OPC-UA integration with the existing MES for automated pass/fail date-stamp verification. The goal was to eliminate dual wet-ink signatures while meeting FDA traceability requirements.

The system achieved zero false positives and 15-second processing speed but failed on false negatives due to production lighting variation, magnification challenges, and insufficient training data. The team pivoted to a sub-$5,000 Andon cord system that reduced changeover verification time quickly, saving an estimated $125,000+ by avoiding a failed pilot.

• Establish clear, measurable success criteria before technical work begins — especially in FDA-regulated environments where error tolerance is near zero.
• A disciplined POC-first approach (think big, start small, fail fast) can deliver more value through risk avoidance than a successful but overly ambitious rollout.
• Human expertise is sometimes the optimal solution; AI automation requires controlled, representative conditions that production environments may not provide.