Building custom finance assistants

1. Building custom finance assistants

Hello again!

2. So far, so good

So far, you have learned how to structure prompts, set custom instructions, and schedule recurring queries. Now we'll combine those ideas to build a dedicated AI assistant - a focused Copilot setup that helps you research, analyze, and communicate finance insights consistently.

3. Too much context, too little time

Picture this: your CFO wants an update that ties together earnings results, market news, and ESG actions. You could start a new chat, re-enter your instructions, and upload files each time, or you could create an assistant that already knows what it's looking for. A structured prompt is a one-time request with a clear question, data, and format. Custom instructions are persistent guidelines that shape how AI answers every prompt - including tone, sourcing, and structure. Both improve quality, but they still require you to start the conversation manually.

4. AI assistants

A Copilot agent builds on those tools. It acts like an enhanced assistant that remembers your task setup and can reuse your preferred tone, format, and reference files. Instead of re-explaining your goals each time, you launch a ready-made collaborator that understands your workflow. For finance professionals, these assistants save time and improve reliability. They can summarize company filings, extract KPIs, draft commentary, or compare peer ESG data using uploaded documents. Because they remember your format and tone, outputs stay consistent across meetings and reports.

5. Setup

In Copilot, you can create an agent-like assistant using plain language. Type a prompt such as: "You are a finance assistant that prepares executive summaries from company filings and market data. Use a neutral tone and cite sources." and let Copliot get to work. Then upload your knowledge files - like 10-Ks or earnings decks - to give the assistant organization-specific context. Other tools, like ChatGPT with custom GPTs or Gemini and Claude projects, provide similar features. The experience differs, but the concept is the same: define a role, add knowledge, and reuse it whenever you need.

6. Example

Follow these steps to create an agent. 1. Open a new Copilot chat. 2. Write a natural language prompt describing the assistant's role, tone, and task. 3. Add instructions for tone, structure, and sourcing. 4. Upload relevant reference files or datasets. 5. Save the setup as your finance assistant for reuse.

7. Working with your assistant

Once it's ready, you can prompt your assistant right away. Often the newly configured assistant will offer suggest conversation starters, or guided prompts to get you started. The assistant can process your files and return summaries, analyses, evaluations - anything you require tailored your own specifications.

8. Knowledge files and integrations

Uploading files transforms your assistant into a specialized Copilot agent. Instead of searching the web, it draws from your internal documents - reports, forecasts, or policy manuals - ensuring that every output is accurate, relevant, and compliant. App and service integrations further improve your ability to continue working within your organization's ecosystem or preferred applications and service providers.

9. Building practices

Start small with one clear task, like summarizing a filing. Keep your instructions concise and measurable. Use trusted, up-to-date files only and beware of what is shared with the AI model. Re-train or refresh context when data changes. As always, verify all outputs before sharing.

10. Summary

Assistants combine structured prompts, custom instructions, and knowledge files into one reusable workspace. As you add integrations, your Copilot assistant can become more powerful - eventually evolving into an agent that can support more complex, multi-step workflows.

11. Let's practice!

Let's build another example together using Copilot's Agent Builder.

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.