We let our last support agent go. Then realized we didn't need to replace them. Our AI was already answering tickets better and faster than they were. So Adam and I made the call: no replacement hire.
In August, that AI worked 182 hours. That's 109% of the human we decided not to hire to replace the last guy.
We're trying to be transparent about leaning into an AI-first support model.
Here's our August:
- 168 possible working hours at RB2B (or 128 less our AI-only week)
- 182 hours of support handled by Intercom's Fin
- 552 tickets resolved
- 70.1% resolution rate
The money math:
- Human salary we're not paying: $7,449
- What we paid AI: $552
- Savings: $6,897
- ROI: 1,249%
Here's what actually happened:
Our paying customers, who have direct access to human support, clicked the AI option first. By choice. Repeatedly.
Resolution rate climbed 1.7% from July. The AI is literally getting better by handling our customers' real problems.
I run AI at a company that publishes every metric.
This is one of them: we discovered our AI was outperforming the human it was supposed to assist.
Other companies are debating this behind closed doors.
We're building in public and the data speaks for itself.
