How Much Does It Cost to Train AI on Your Data

The Two Costs of AI Training

There are two separate costs: the one-time cost of embedding your training data, and the ongoing cost of chatbot conversations that use that data.

Embedding Cost (One-Time)

When you upload documents, the platform breaks them into chunks and creates vector embeddings for each chunk. This costs 3 credits per chunk (about $0.003). The number of chunks depends on the length of your content:

• A 1-page FAQ document might produce 3 to 5 chunks
• A 10-page employee handbook typically produces 20 to 40 chunks
• A full product catalog might produce 50 to 150 chunks
• A comprehensive knowledge base with multiple document types usually falls between 100 and 500 chunks

At 3 credits per chunk, here are some real-world examples:

• Small FAQ (15 chunks): 45 credits = $0.045
• Employee handbook (35 chunks): 105 credits = $0.105
• Full knowledge base (200 chunks): 600 credits = $0.60
• Large documentation set (500 chunks): 1,500 credits = $1.50

This is a one-time cost per chunk. You only pay again if you delete and re-upload content. See How to Delete or Update Specific Training Data.

Conversation Cost (Ongoing)

Once your data is embedded, the ongoing cost is the chatbot conversations themselves. Each time someone asks a question, the platform retrieves relevant chunks and sends them to the AI model along with the question. The cost depends on which model you use:

• GPT-5-nano: 1 to 2 credits per message, good for simple factual lookups
• GPT-4.1-mini: 2 to 4 credits per message, best balance of quality and cost for most use cases
• Claude Sonnet: 4 to 8 credits per message, excellent for nuanced and detailed responses
• GPT-4.1 / Claude Opus: 8 to 15 credits per message, best accuracy for complex questions
• Reasoning models (GPT-5.2): 10 to 30+ credits per message, for tasks requiring deep analysis

See How Much Does an AI Chatbot Cost for a detailed breakdown of conversation costs by model.

Real-World Monthly Cost Examples

Small Business FAQ Chatbot

15 chunks of FAQ content, GPT-4.1-mini model, 300 conversations per month averaging 3 messages each. Training: 45 credits ($0.045 one-time). Monthly conversations: 900 messages at roughly 3 credits each = 2,700 credits ($2.70/month).

Customer Support Knowledge Base

150 chunks from support docs and product guides, GPT-4.1-mini, 2,000 conversations per month averaging 4 messages. Training: 450 credits ($0.45 one-time). Monthly conversations: 8,000 messages at roughly 3 credits = 24,000 credits ($24/month).

Internal Expert System

400 chunks of detailed technical documentation, Claude Sonnet for better accuracy, 500 conversations per month averaging 5 messages. Training: 1,200 credits ($1.20 one-time). Monthly conversations: 2,500 messages at roughly 6 credits = 15,000 credits ($15/month).

Reducing Costs

• Choose the right model for the task. Most FAQ and customer support chatbots work great on GPT-4.1-mini at 2 to 4 credits per message. Reserve expensive models for use cases where accuracy on complex questions justifies the cost.
• Write efficient training data. Well-organized, non-redundant content means fewer chunks and lower embedding costs. See How to Organize Training Data for Best Results.
• Use your own API keys. If you bring your own OpenAI or Anthropic API key, the platform markup is lower. You pay the AI provider directly at their published rates.

Compared to Alternatives

For context, most enterprise chatbot platforms charge $50 to $500+ per month as a flat subscription fee before you even start using them. Per-message pricing platforms typically charge $0.01 to $0.05 per message (10 to 50 credits equivalent). Our credit-based model means you pay only for what you use, with no minimum monthly commitment, and the actual cost per message is typically lower than dedicated chatbot platforms.