The Opportunity

As of 2024, 2.6 billion people remained offline globally, while Pakistan alone had 131 million offline users and 188.9 million active cellular mobile connections. That gap showed the core distribution problem: many people cannot be reached through an app, but they can still be reached through a phone call.

The Approach

I led platform and partnership work for a voice-first GenAI service where users dial a telecom shortcode, route through existing IVR infrastructure, and speak with a local-language AI agent. Instead of creating a new pipe, the product layered GenAI on top of telco voice rails already carrying billions of minutes.

I also partnered with engineering on a hybrid optimization stack that reduced per-query inference cost by roughly an order of magnitude and materially improved latency, making the economics viable for low-ARPU emerging markets.

The Impact

• Live commercial deployments with two of Pakistan's three largest mobile operators, with a third active in pitch.
• Order-of-magnitude reduction in per-query AI compute cost, with response latency roughly halved.
• Created a repeatable voice-AI-via-MNO partnership model for a 2.6B-person offline market.
• Gave partner operators a differentiated GenAI VAS proposition with revenue-share economics.

The Takeaway

In markets where mobile phones outnumber internet users, the natural delivery channel for GenAI may not be an app. It may be a voice call. The lesson was to build upward from the constraints of the channel the user already has, rather than downscaling a Western app-first assistant.