Score every AI agent on your line.
Vision QC, predictive maintenance, scheduling optimizers, supply-chain agents. Defect-detection accuracy, downtime reduction, throughput impact. AgentScore reads your operational data and tells you which agents are catching defects and which are producing them.
What you see in the dashboard
- Per-agent score across defect-detection accuracy, downtime reduction, production throughput impact.
- ImplementationScore for the plant — data foundation, governance maturity, learning loops.
- Hard grade cap on agents with policy-violation events — an A is impossible without clean audits.
The agents
What AI on a manufacturing floor typically looks like.
Vision, planning, and maintenance — same eight factors as every other vertical, weighted for plant operations.
QC vision agent
Inspects parts on the line, flags defects, sends rejects to a human reviewer. Plays nicely with MES + SCADA.
Predictive maintenance
Reads sensor + downtime telemetry, calls maintenance windows before a line goes down.
Scheduling optimizer
Sequences orders by changeover cost, due date, and machine availability. Proposes — humans confirm.
Supply-chain triage
Reads PO + shipment exceptions, prioritizes which suppliers a planner needs to call this morning.
OEE copilot
Surfaces the loss bucket dragging OEE this week — availability, performance, or quality — and points at the line.
The signal
What we read from your MES, SCADA, and QC log.
CSV upload for the QC log. Connectors for MES + SCADA on the integration roadmap.
Defect-detection accuracy
Vision agent's catch rate on labeled rejects, peer-relative.
Downtime reduction
Hours of unplanned downtime avoided where the agent triggered the call.
Throughput impact
Production-rate delta on lines the agent operates against the human-only baseline.
Override rate
How often a QC inspector overrides the agent's defect call.
Policy-violation rate
Audit failures or out-of-spec passes — the governance hard cap.
Tier-2 automotive supplier · 4 lines · 2 vision agents
Vendor-reported 99.4% accuracy held to the peer benchmark.
Pre-AgentScore: vision QC vendor reported 99.4% accuracy. AgentScore's peer-relative z-score showed the agent was 1.8σ below peer-group median on defect-detection accuracy.
Vendor retrained the model on plant-specific data, accuracy moved to a true 99.5%, scrap rate fell 8% the next month.
SOC 2 in progress
Type II audit kicked off Q3 — interim report on request.
Read-only data access
OAuth scopes never request write to your CRM, jobs platform, or call data.
Algorithm-only scoring
Every score computed by ASAM. No manual overrides — not for customers, not for us.
Anonymized at ingest
Identifiers stripped before any aggregate analytics. We never train on identified rows.
Upload your QC log — score in 5 minutes.
The number on the vendor's slide isn't the number on your line. AgentScore tells you which one is real.