59,000 Hours of Real Business Coaching & Sales Calls for AI Training

Data Product Features

• Real experts, real stakes: No actors and no scripts. People bring real businesses, real money worries and real decisions.
• Group session structure: Hot seats, Q&A and teaching blocks are marked with start and end times and the student in focus.
• Speaker roles: Every speaker is labeled coach, student or other participant. Pseudonyms are unique to each call.
• Stages, summary and themes: Each call has a summary, a main theme and supporting themes, and is split into titled stages with times and a short summary of each.
• Tone on every segment: Each transcript segment has a tone label (such as encouraging, frustrated, curious, empathetic or relieved) and a strength of low, medium or high.
• Conversation flow: Each segment is labeled with what it does in the conversation, such as check-in, exploration, teaching, problem statement, reframe, commitment or close.
• Coaching techniques and their effect: Each technique the coach uses (such as a diagnostic question, normalizing, giving an example, a reframe or celebrating a win) is labeled. The data shows whether the student's emotional score then went up, stayed flat, went down, or drew no response.
• Key moments and turning points: Insights, breakthroughs and resistance are marked, along with coach redirects, topic shifts and student pushback, each with a short note.
• Emotional arcs: Each speaker's emotional path through the call is scored from −3 to +3. Notes explain each emotion shift and the moment that triggered it.
• Voice measurements: Speaking speed, average pitch, pitch variation, relative loudness and the pause before each segment.
• Quality data: Every call records its signal-to-noise ratio, clipping, speech recognition confidence and the privacy checks it went through.
• Living dataset: More hours are added as we process our archive.

Distribution

• Format: Each call ships as four files: audio (mono, from the original Zoom recording, AAC at 32 kHz); a JSON file with every label; a Markdown transcript with tags on every segment; and a self-contained HTML viewer with a table of contents and an emotional-arc chart.
• Structure: One folder per call, named by a random call ID (e.g. CC-WYKCVAP5), grouped by year.
• Size: About 3.4 TB, almost all of it audio. The three text files add roughly 25 GB.
• Delivery: S3 bucket.
• Data Volume: About 59,000 hours; about 50,000 calls; roughly 13 million transcript segments (about 220 per hour); 11 fields per segment, plus call-level data.

Usage

• AI coaching and support agents: Train agents to ask good questions, notice how someone feels, and respond with care.
• Technique-to-outcome modeling: Learn which coaching techniques help people move forward, using the before-and-after emotion scores.
• AI sales assistants: Train and test sales agents on real conversations where people voice doubts, push back and decide.
• Sales call scoring: Build tools that review sales calls and coach reps.
• Emotion recognition: Recognize feelings from both words and voice in real, multi-person conversations.
• Speech recognition and speaker separation: Improve accuracy on real Zoom calls with many speakers, microphones and accents.
• Expressive voice models: Help synthetic voices sound warm and natural.
• Evaluation sets: Test how well models follow long, real conversations.

Coverage

• Geographic Coverage: Global; most participants are in the United States, Canada, the United Kingdom and Australia.
• Time Range: About 2015 – 2026.
• Demographics: Adult entrepreneurs and small-business owners, from first-time founders to experienced owners, talking with expert business coaches and a sales team. Language: English.

License

AI Training Rights

• Use the Data Product to train, fine-tune, and evaluate machine learning models, including large language models.
• Incorporate Data Product content into models and commercialize resulting model outputs.
• Create derivative works (model weights, embeddings, etc.) for any lawful purpose.

• The Data Product itself may not be sold, redistributed, or shared outside of licensed usage.
• Licensee may not attempt to identify any person in the data.
• Licensee must comply with all applicable laws, including data protection and privacy regulations.

Who Can Use It

• AI labs and model builders: For training and fine-tuning conversational and voice models.
• Data scientists and ML engineers: For building emotion, speech and dialogue models.
• Sales technology companies: For AI sales assistants and call-scoring tools.
• Coaching, education and support platforms: For AI coaches, tutors and support agents.
• Researchers: For studies of conversation, emotion, group dynamics and how people change.

Data Dictionary

• How the labels are made: Our own automated pipeline makes every label, running entirely on our own computers. Transcripts come from NVIDIA Parakeet speech recognition, with Whisper as a backup. The labels are made by AI models, not typed by hand, and each call records whether a person has reviewed it. Buyers can check label quality in the sample before licensing.
• Privacy: No video is included. Names of people, businesses, products and places, plus dates, amounts, links and contact details, are replaced with numbered placeholders like [Person 3] or [Business 2]. Spoken names are covered in the audio before delivery. Calls are dated by year and quarter only, and pseudonyms are unique to each call, so no one can be followed from call to call. Every call goes through automated privacy checks, including an AI privacy review.
• Consent: Every coaching call is covered by signed agreements from everyone on it. Every sales call starts with a notice that the call is recorded and may be shared or used for training. A consent summary is available on request, and full terms can be reviewed under NDA.
• Slices: License the whole library, or just part of it: coaching only, sales only, certain years, or text only.
• Sample: 5 full calls (3 coaching, 2 sales), each with all four files.