Section 7: HR, finance-accounting, legal-compliance, procurement and operations

Scenario 69: Categorizing spending and listing variances

Level: Intermediate · Tools: ChatGPT Data Analysis; the Microsoft Copilot for Finance scenario and Copilot Chat, if the relevant license or preview access is available [71] [130].

Who and in what situation

Ayşe is a procurement analyst at a retail company in İstanbul. She maps monthly card and procurement exports to the category hierarchy and standardizes different names for the same supplier.

How it is done today

Descriptions are cleaned manually, the category dictionary is searched, and managers are asked about large variances. Different people may classify ambiguous rows differently.

How to do it with AI

  1. Consolidate category codes and exception rules into a single approved file; keep only one transaction per row.
  2. Attach the CSV and category dictionary using Add photos & files; request fields for the category code, matching rule, uncertainty, and human review [71] [79].
  3. Have category totals and period variances calculated; review the code showing the columns and assumptions used [71].
  4. If the organization has Copilot for Finance, compare the results with the spending variance and supplier table in Microsoft’s “procurement insight acceleration” scenario; do not treat the preview as a production decision [130].

Example prompt

Match each transaction in the {spending export} file to the {approved category hierarchy}. Write the category code, matching rule, reason for uncertainty, and circumstances requiring human review. Do not guess if the description is insufficient. Show period totals with the columns and assumptions used. Do not grant budget or procurement approval.

Evidence and real-world example

Microsoft’s scenario library provides examples of reviewing spending variances with Copilot for Finance and tabulating suppliers by service; it states that the content is for demonstration purposes [130].

Points to consider

Data Analysis calls for reviewing methods and assumptions [71]. The Microsoft scenario shows the relevant access and some preview capabilities [130]. NIST recommends addressing trustworthiness considerations throughout use and evaluation [131]. Financial decisions must not be automated without human approval [123].

Sources for this scenario

  1. Data analysis with ChatGPT (help.openai.com)
  2. Uploading files and audio to ChatGPT (help.openai.com)
  3. OpenAI Usage policies effective October 29 2025 (openai.com)
  4. Finance scenario procurement insight acceleration (adoption.microsoft.com)
  5. Artificial Intelligence Risk Management Framework Generative Artificial Intelligence Profile (nist.gov)