AI Conversation Analysis for Trade Show Sales Teams

AI conversation analysis is the use of language models to extract structured data from a recorded conversation, such as pain points, time pressure, sentiment, and follow-up actions. Exporb applies this to trade show booth audio so reps get a scored lead instead of an unread transcript.

Exporb runs three stages on every recording. First, the audio is transcribed on the server. Second, a language model extracts eight signal categories with quoted spans from the transcript. Third, those signals feed a numeric sentiment score from -1 to 1 and a 0-100 lead score that auto-sorts the contact into Hot, Warm, or Fresh.

The AI tags each conversation with eight signal categories: past failures, time pressure, quality and capability gaps, cost concerns, compliance, competitive threats, sample or capability requests, and direct verbal cues. Every tag includes a quoted span from the transcript so the rep can verify the source.

Every prompt instructs the model to extract only what is on the card or said in the audio. If a field is unclear, the model returns null instead of guessing. Each extracted signal includes a quoted span, and business card scans run an additional anti-prompt-injection layer.

Audio recordings of up to 5 minutes are supported per capture. Longer conversations are chunked and processed in parallel on the server, so a 5-minute recording typically finishes analysis in under a minute. Recordings can be combined later.

Capture works fully offline. Audio, photos, and notes save to the device, and the AI analysis queues automatically. The server runs the extraction the moment the app reconnects, so the user can close the app and the analysis still completes.

Each conversation produces a numeric sentiment score from -1 to 1 and a lead score from 0 to 100. Leads scoring 70 or higher are marked Hot, 40 to 69 are Warm, and below 40 are Fresh. Scores combine sentiment with the eight extracted signals.

Sentiment calibration is a numeric score that reflects buying intent rather than just tone. A buyer who calmly commits to a $50,000 order scores higher than a buyer who is loudly enthusiastic but non-committal. Exporb returns the score as a value between -1 and 1.

Yes. Exporb exports leads as Salesforce-ready and HubSpot-ready CSVs with mapped fields, plus Excel, JSON, and a ZIP archive that bundles every business card, audio file, and photo. Real-time API sync is in development.

Yes. The same conversation produces seller fields like lead score and follow-up actions and buyer fields like payment terms, sample availability, and supplier score. Buyer mode also runs a batch material analysis on photos taken at the booth.

Transcription and extraction work for English booth conversations today. Additional language support is on the roadmap. Business card scanning supports any language because OCR is layout-driven.

Lead scoring combines a calibrated sentiment value with eight extracted signal categories, each anchored to a quoted span from the transcript. Reps can review every signal source, override any score, and adjust the Hot, Warm, or Fresh bucket manually before exporting.

Audio, transcripts, and extracted fields live in Supabase with Row-Level Security on every table. Data is encrypted in transit and encrypted at rest via Supabase managed encryption. Production logs are sanitized so personal data is never written to log streams.

Otter.ai transcribes audio. Gong analyzes recorded sales calls after the meeting. Exporb is built for booth conversations: it captures offline, runs the analysis on the server even after the app closes, and produces a scored lead with quoted signal spans, all on a 0-100 lead score scale.