Fine-Tuning Custom AI Models

Overview

Fine-tuning trains a custom AI model on your actual call data. The result is an AI assistant that understands your business's specific language, products, and customer interactions better than a general-purpose model.

Enterprise plan only. Requires at least 50 completed call transcripts.

How It Works

1. AIVO collects transcripts from your completed calls.
2. The transcripts are formatted into training data (prompt/completion pairs).
3. A fine-tuning job is submitted to AIVO's AI platform.
4. The platform trains a custom model based on your data.
5. Once training completes, your assistant is updated to use the new model.

Requirements

• Enterprise plan subscription.
• At least 50 completed call transcripts with good quality audio.
• Calls should cover a representative sample of your typical interactions.
• Training typically takes 2-4 hours depending on data volume.

Starting a Fine-Tuning Job

1. Go to Voice & AI > Advanced > Fine-Tuning.
2. AIVO shows how many eligible transcripts you have.
3. Click Start Training when you have at least 50.
4. Choose training parameters:

• Epochs: How many times to train over the data (default: 3).
• Model suffix: A name for your custom model (e.g., "dental-clinic-v1").

1. Click Submit Job.
2. Monitor progress on the same page. Status updates include: pending, running, succeeded, or failed.

After Training Completes

1. AIVO automatically updates your assistant to use the fine-tuned model.
2. Make a test call to verify the improved responses.
3. Compare call quality metrics (confidence scores, resolution rates) before and after.

Best Practices

1. More data = better results. 50 calls is the minimum; 200+ calls produce noticeably better models.
2. Retrain periodically. As your business evolves (new services, seasonal changes), retrain every quarter.
3. Review bad calls first. Before training, review low-confidence calls and update your knowledge base. Training on bad data produces a bad model.
4. Keep the base knowledge base current. Fine-tuning supplements your knowledge base; it does not replace it.

Monitoring

View all fine-tuning jobs in Voice & AI > Advanced > Fine-Tuning > Job History:

• Job ID, model name, status, start time, completion time
• Training metrics (loss curves)
• Option to roll back to the base model if the fine-tuned version underperforms