> For the complete documentation index, see [llms.txt](https://docs.gooey.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.gooey.ai/ai-agent/copilot/what-to-do-before-you-build-a-bot.md).

# What to do BEFORE You Build an AI Agent

**Identify your use case and value hypothesis**. Where do you believe an AI agent may be useful? To which customers or people?

**How will you measure the agent's value?** Usually, that is:

1. \# of messages sent
2. D1, D3, D7, D30 Retention of users
3. Positive vs negative feedback from users
4. % of questions asked by the user that were answered by the agent

**Curate the knowledge base dataset** e.g. documents, videos, PDFs, websites, and soon real-time data sources.

**Create "Golden" Questions and Answers**

Let’s assume a human expert understood all of the documents in the above knowledge base dataset and then one of your users asked the human expert a question; what ideal answer would you expect the human expert to provide? This is the *Golden Answer.*

1. Collect 10 questions and the ideal **Golden Answer** for each
2. The question should be captured in whatever form we believe the user is likely to ask it - e.g. in Bhojpuri as a WhatsApp audio note, mixed-Hinglish text, English text, and/or Tamil text.
3. If non-English, provide the English text translation of the question
4. The ideal “golden” answer in English text
5. Collect the above as rows in the Excel or Google sheet.