Old rule: more revenue → more people. Agents break the rule. You get non-linear scale: faster cycle times, elastic capacity, lower unit cost.
What changes with agents
- Cycle time: End-to-end workflows run in parallel; handoffs vanish → ~60% faster.
- Elastic ops: Spin capacity up for peak, down for quiet. No temp army.
- Cost: Combining process excellence + agents drives ~40% operating cost reduction.
Examples that travel well
- Hospital discharge: Agent-coordinated steps → –40% wait time, +25% bed utilization.
- GE Healthcare: AI-assisted reads –40% time, +8.5% accuracy.
- Loan processing: Doc AI → –80% cost, 20× faster decisions.
Metrics to track
- Goal completion rate (end-to-end jobs done by the agent).
- Human-in-the-loop frequency (where it still needs help).
- Token/compute cost per job (unit economics).
- Throughput per FTE (this is your margin unlock).
Where it breaks
- Automating broken processes: If it’s messy now, it’ll be messy faster. Fix flow first.
- Multi-agent orchestration: As you add agents, you need an orchestrator and clear roles.
- Governance: Logging, access, and rollback plans aren’t optional.
Quick start
1. Pick a single, repeatable workflow (claims, onboarding, collections).
2. Map steps → remove human-only gates that don’t add judgment.
3. Automate the ugly middle (handoffs, updates, lookups).
4. Report weekly on cycle time, queue depth, human assists, cost per job.