Over the last two years we’ve put more than 40 agents into production across logistics, healthcare, finance and commerce. We went back through the numbers to see what actually paid off.
Speed mattered more than savings
Labour savings were real but modest. The bigger wins came from speed: replying to leads in minutes, resolving exceptions the same day, closing the books faster. Speed compounds.
The median payback was 11 weeks
Fastest payback: 3 weeks (inbound lead qualification)
Slowest: 9 months (a bespoke research copilot)
Projects with a clear baseline paid back 2.3× faster
Adoption beats accuracy
A 92%-accurate agent the team trusts outperforms a 98%-accurate one they route around. We now spend as much time on hand-offs and explanations as on prompts.
If you’re planning your first deployment, our ROI calculator uses the same assumptions we use internally.
Written by
Daniel Reyes
Built ML platforms for a decade before agents were cool. Believes every prompt deserves a test.
