Insights

Notes on building AI products

Short, opinionated takes drawn from actually shipping AI products and a decade in the field — not recycled hype. New notes as I build.

Product philosophy

Why I build AI products “additive, not replacement”

The AI products that survive contact with a real org don't fire people — they take the repetitive 80% so humans do the high-judgment 20%. In MB Voice, the AI qualifies and routes; humans still close. That framing is what makes AI adoptable inside a sales floor, not feared by it.

AI PM lessons

Compliance is a feature, not a footnote

In Indian telephony, TRAI/DLT rules can get your numbers blacklisted overnight. So in MB Voice, consent, calling-hours, and DNC checks gate every single dial, and per-call compliance is auto-scored. Building the guardrail into the core loop — not bolting it on later — is what makes an AI product enterprise-ready.

Conversation intelligence

From 2–3% sampled to 100% scored

Most call-QA is theatre: a manager listens to a random handful of calls a week. The real unlock isn't a better sampling method — it's making sampling obsolete. CallEdge scores every conversation automatically, which changes coaching from anecdote to evidence.

Career

The best PM training I had was a sales headset

Ten years of objection-handling taught me more about discovery, prioritisation, and value proposition than any framework. If you're moving into product from a customer-facing role, that experience is your edge — not a gap to apologise for.

Want to learn this properly?

The AI PM Blueprint turns these lessons into a structured, self-paced path — with a real product you build yourself.

Explore the program →