Nudge

Nudge

A Real-Time, Self-Improving AI Sales Coach

Built with Modulate · Airia

Inspiration

Every sales rep has had that call — the one where they talked too long, missed the buying signal, or let a vague objection slide without pushing back. They only realize it in the debrief, hours later, when it's already too late.

We asked ourselves: what if a rep had a silent expert sitting beside them on every call — someone who could read the room in real time, and whisper the right move at the exact right moment? That's Nudge.

What It Does

Nudge is a real-time AI coaching layer that runs silently during live sales calls. It listens, thinks, and surfaces a single precise nudge to the rep the moment they need it — visible only to them.

During a call, Nudge detects the current phase (intro, pitch, Q&A, objection, negotiation, close), identifies critical events like prospect frustration, price objections, competitor mentions, or the rep talking too long, and cross-references the prospect's known preferences from a live insight database. It then surfaces one prioritized, context-aware nudge — short enough to read in two seconds, specific enough to act on immediately.

After the call, Nudge runs a post-call analysis, scores how the call went, and updates rep and prospect insight profiles for next time. It gets smarter with every call.

How We Built It

Modulate / Velma handles real-time speech-to-text transcription. Each utterance comes back with a speaker label, timestamps, and an emotion tag — Calm, Frustrated, Confused — that Nudge uses as a live signal. A [Frustrated] tag tells the agent to deprioritize everything else and address the emotional state first.

Airia is where the intelligence lives. We built two agents on the Airia canvas: a Live Coach Agent that detects events, queries the knowledge base and prospect profiles, and returns a structured nudge on every conversational turn; and a Post-Call Analysis Agent that evaluates the full call record, scores nudge effectiveness, and writes updated insights back to the live data layer. Both agents are called from a Python script running on the rep's laptop via Airia's API interface.

The self-improving loop works by persisting rep and prospect insights in a live Google Sheet. After each call, the post-call agent updates these profiles — tracking communication styles, known objections, what worked, what didn't. The live agent reads them at the start of every new call, making each nudge more specific over time.

Challenges

Getting nudge prioritization right when multiple events fire simultaneously, keeping nudges short enough to be useful mid-conversation, and closing the self-improvement loop with a writable live data layer rather than static files.

What's Next

Rep-facing UI with one-tap nudge feedback, manager dashboards showing rep improvement trends, CRM auto-fill via Airia after each call ends, and Braintrust integration to A/B eval coaching policy variants across real call transcripts.