Case Studies

Real-world examples of ACGP protecting organizations from AI agent risks.

  • Financial Services

    How a trading firm prevented a $2.3M loss with ACGP governance.

    Read case study

  • Customer Service

    Reducing refund fraud by 94% while maintaining customer satisfaction.

    Read case study

  • Enterprise Deployment

    Scaling ACGP across 500+ agents in a Fortune 500 company.

    Read case study

Impact Summary

Metric Before ACGP After ACGP Improvement
Unauthorized actions blocked 0% 99.7% -
Average incident cost $47,000 $1,200 97% reduction
False positive rate N/A 2.3% Minimal friction
Time to detect issues 4.2 hours 45ms Real-time

Common Patterns

What Works

  1. Start with ACL-2 - Most production agents perform well with standard oversight
  2. Configure tripwires early - Set financial and operational limits before deployment
  3. Monitor trust scores - Agents with declining trust often signal problems
  4. Use governance contracts - Explicit latency/quality trade-offs prevent surprises

What to Avoid

  1. Over-governing - ACL-5 for a simple FAQ bot adds unnecessary latency
  2. Ignoring nudges - Repeated nudges often predict future blocks
  3. Static configuration - Agents should earn (or lose) autonomy over time
  4. Skipping testing - Conformance tests catch issues before production

Share Your Story

Using ACGP in production? We'd love to feature your success story.

Submit your case study

Implementation Guides Contributing