Real IQ — Voice-Based Customer Feedback SaaS

Surveys have a response-rate problem and a depth problem — people skip them, and the ones who don't give you a number without a reason. Real IQ replaces the survey with a voice conversation, then infers CSAT, NPS, sentiment, themes, and root-cause drivers from the transcript. The inference pipeline was the core product bet, not a feature bolted onto a voice agent.

What I did

• Compressed a typical 6-month build into 6 weeks using Lovable and Claude Code, with strict prompting discipline to prevent regressions
• Separated the live-conversation model (optimized for cost and latency) from the post-session inference model (optimized for accuracy) to protect unit economics without giving up insight quality
• Priced subscription tiers with conversation-based quotas over per-call credit packs, based on cost modeling per conversation
• Set the data-retention rules for the platform — transcripts only, no voice data at rest — reducing breach exposure and simplifying enterprise security review