Conversation Training — AI That Learns From Your Agents

What Is Conversation Training?

Conversation Training automatically makes your AI bot smarter over time by learning from resolved agent conversations. Every time an agent closes a chat, Velaro extracts the visitor's questions and the agent's answers, converts them into AI-readable knowledge, and adds them to the bot's search index. The next time a visitor asks a similar question, the bot can surface your agents' real, proven answers.

This is different from KB articles (which you write manually) or web training (which scrapes your website). Conversation Training learns from what your agents actually say — in the language your customers actually use.

How It Works

When a conversation resolves, Velaro automatically:

1. Checks the quality of the conversation (minimum 4 messages, at least one agent reply)

2. If CSAT was collected, only indexes conversations rated 4 or 5 stars

3. Extracts visitor question → agent answer pairs

4. Converts them to vector embeddings and adds them to the bot's knowledge index

5. The bot now has this knowledge for future conversations — no manual action required

The bot searches all knowledge sources simultaneously — KB articles, web training, files, and conversation training — and picks the most relevant answer regardless of source.

Enabling Conversation Training

Conversation Training is a subscription add-on. To enable it for your account, contact your Velaro account manager or Velaro support. Once enabled, it activates automatically — no additional setup required.

To view your indexed conversations: go to Bots → [Your Bot] → AI Settings → Training → Conversation Training tab. You'll see a count of indexed conversations, a channel breakdown, and a list of every indexed conversation you can review or remove.

Getting the Best Results

Enable CSAT or thumbs-up ratings — without ratings, all conversations with 4+ messages are indexed regardless of quality. With ratings enabled, only high-rated conversations make it in. This is the single most impactful thing you can do to improve training quality.

More agent conversations = faster improvement — the first 30 days build a foundation. After 90 days with active agents, most bots show measurable improvement in answer accuracy and deflection rate.

Remove bad examples — if an agent gave a poor answer that was indexed anyway, use the Remove button on the Conversation Training tab to pull it from the index.

Removing a Conversation From Training

Navigate to Bots → [Your Bot] → AI Settings → Training → Conversation Training. Find the conversation in the table and click Remove. The Q&A pairs from that conversation are immediately deleted from the bot's knowledge index.

KB Articles auto-push to bot

When you publish or update a Knowledge Base article, Velaro automatically pushes it to the bot's vector index — no manual step required. The bot can answer customer questions using your KB content within seconds of publish.

To check which articles are indexed: go to Bots → [Your Bot] → AI Settings → Training → Knowledge Base tab. You'll see each article's status and when it was last synced.

How Conversation Training Combines With Other Sources

All training sources are searched simultaneously every time the bot answers:

• KB Articles — auto-pushed when published; high-confidence answers for common topics
• Web Training — content scraped from your website
• File Training — PDFs, Word docs, spreadsheets
• Conversation Training — real Q&A from your agents
• Store Q&A — answers from Shopify, BigCommerce, or WooCommerce Bot Knowledge tabs

The bot returns the most semantically relevant answer from any source. You don't need to choose — more sources means better coverage and more accurate answers.

Cost and Plan Availability

Conversation Training is available on paid plans. Indexing costs approximately $0.001 per conversation — effectively free for most customers. Limits by plan: Starter (200 conversations/month), Pro (2,000/month), Enterprise (unlimited). Contact support to check your current plan and enable the feature.