How to build a CapEx Classification AI Agent

TL;DR

This agent automates and verifies CapEx (Capital Expenditure) classification by analyzing project documents against your organization’s CapEx guidelines, prompting for clarifications, and writing the verified analysis to Google Sheets.

What it Does

• Ingests project documents (RFCs) and CapEx guidelines.

• Uses AI to analyze if the project meets CapEx criteria.

• Prompts the user for clarifications if any criteria are unclear or unmet.

• Runs a second AI check for completeness, accuracy, and bias.

• Writes the final, verified analysis to a Google Sheet for record-keeping.

Who It’s for

• Finance teams

• Project managers

• CapEx/OpEx analysts

• Anyone responsible for CapEx compliance and documentation

Output

• A clear, AI-generated CapEx classification analysis.

• Follow-up questions if the RFC/project description is unclear or incomplete.

• A verified summary written directly to your specified Google Sheet.

Common Pain Points of Classifying CapEx

• Unclear or incomplete project descriptions.

• Manual, time-consuming cross-referencing of guidelines.

• Risk of missing compliance criteria.

• Inconsistent documentation and record-keeping.

• Lack of audit trail for decisions and clarifications.

What This Agent Delivers

• Automated, consistent CapEx analysis.

• AI-driven follow-up questions to ensure completeness.

• Double-checks for accuracy and bias.

• Centralized, auditable record in Google Sheets.

• Saves time and reduces manual errors.

Want to See More? Check Out This Quick Tutorial:

Step-by-Step Build (StackAI Nodes)

1) CapEx Classification Guidelines (Files Node)

What it does:

• Lets you upload your CapEx guidelines document.

• Extracts and processes the text for use by the AI.

Goal:

• Ensure the AI has the latest, official CapEx criteria for analysis.

2) RFC (Files Node)

What it does:

• Lets you upload the project’s RFC (Request for Capital) or description.

• Extracts and processes the text for analysis.

Goal:

• Provide the AI with the project details to be classified.

3) User Response (Input Node)

What it does:

• Allows the user to answer follow-up questions from the AI.

Goal:

• Clarify any uncertainties or missing information in the RFC.

4) OpenAI (LLM Node)

What it does:

• Analyzes the RFC against the CapEx guidelines.

• Prompts the user for clarifications if any criteria are not fully met.

Goal:

• Ensure the project meets 100% of CapEx criteria, or gather more info if not.

Instructions

You are a capitalization analyst. 

Your job is to analyze project descriptions and perform capitalization analysis. 

You will be provided with project descriptions, files, and more. Please cross-reference all of the information with the provided Capitalization Criteria to ensure that the project description meets 100% of the criteria. 



Criteria to capitalize:

<criteria> 

{doc-1.documents} 

</criteria> 

Critical: if you have any uncertainty or if there is any vagueness in the project, ask the user follow-up questions

Prompt

<RFC>

{doc-0.documents} 

</RFC>



<instruction>

In order to write a capitalization analysis for all files, prompt the user to follow up on ALL points where the RFC has not completely met the criteria. If there is any uncertainty, continue to prompt the user until you are 100% certain about if their proposal meets the criteria.

</instruction>



<user_response>

{in-0} 

</user_response>

5) OpenAI Verification (LLM Node)

What it does:

• Reviews the initial analysis for completeness, accuracy, and bias.

• Flags any issues or confirms the analysis is sound.

Goal:

• Provide a second layer of review for quality assurance.

Instructions

You are a senior reviewer. Your job is to check the previous analysis for completeness, accuracy, and bias. 
  If you find any issues, point them out clearly. Otherwise, confirm the analysis is sound

Prompt

Review the following capitalization analysis for completeness, accuracy, and bias. If you find any issues, point them out clearly. Otherwise, confirm the analysis is sound.



<analysis>

{llm-0}

</analysis>

6) Write to Google Sheets (Action Node)

What it does:

• Writes the verified analysis to a specified Google Sheet and worksheet.

Goal:

• Create a permanent, auditable record of the analysis.

7) Output (Output Node)

• What it does:

  • Presents the final result to the user.

• Goal:

  • Display the outcome and ensure the process is complete.