Built a voice AI platform for e-commerce about a year ago. Thought latency was the only hurdle. it wasn't. The real pain was the trade-offs between model quality, response speed, voice naturalness, and cost. Stuck in that optimisation loop for months. lost a handful of enterprise deals during that phase too. Hindsight says we tried to sell to premium clients before the infra was ready - classic builder's bias.
The smarter play would've been to push a v1 at break-even to small customers, stress-test the hell out of it, and iterate fast. That's how you learn the actual persona you're selling to and then build a proper GTM around the ideal client profile.
Got past the tech eventually (lots of experimentation and model fine-tuning), but that's when distribution became the real bottleneck. selling to end customers directly is painful - low awareness, slow adoption cycles, and everyone wants proof of ROI before they'll even take a call.
The few customers that converted with minimal friction had a clear pattern: they were tech enablers - agencies, CRM platforms, marketing tools that already had distribution channels. They didn't want to rebuild voice infra. They just wanted to plug ours into what they already sell. Once we leaned into that partner route, things started clicking.
so now I'm convinced voice AI is less of a tech problem and more of a distribution problem. Curious how others are navigating this - direct sales or through enablers? Where does the stack still break for you: latency, cost, or UX? And if you are selling directly, are you going vertical or horizontal? Happy to swap notes.
