AIGameskraft
AI that routes, not decides
-60%agent-call attempts
The problem
About 1,750 users a day were trying to reach a human agent, most of them not tech-savvy and not comfortable in English.
What I did
- —Killed my own full-AI-agent idea. Automating judgment was the wrong bet.
- —Built an audio LLM: speak a query in six vernacular languages, get routed to the right page.
- —Automated access, not decisions.
Impact
- —Agent-call attempts dropped from about 1,750 to about 700 a day, a 60% cut, across 3 of 8 apps.
Do not automate judgment, automate access. The highest-ROI AI routes and surfaces, it does not decide.