Video summary

5 BEST Ways to Use AI in Excel for Finance (2026)

Main summary

Key takeaways

Technology

Summary of Technological Concepts & AI-in-Excel Use Cases (Finance Focus)

The video argues that AI can substantially reduce time spent on common Excel finance workflows—such as cleaning data, building models, and preparing presentations—by using agent features and AI-assisted formula/model generation. It highlights five practical ways to apply AI in Excel, with a strong emphasis on speed and the need to audit outputs.

1) Excel AI Agent Mode for Data Cleanup, Consolidation, and Analysis

Scenario: Weekly Excel files with many tabs (e.g., credit card transaction segments) require manual cleanup (remove unwanted header rows/columns) and then consolidation into one dataset.

How it’s used:

  • Prompt the Excel AI agent to consolidate all tabs into a single table.
  • Specify which columns to keep and which rows to delete.
  • Request an analysis output (e.g., spend by card holder name and by day), plus formulas for reuse and auditing.

Example result:

  • AI completes a first draft in ~69 seconds.
  • Produces a consolidated tab with expected fields (date, description, amount, card holder name, company name) and a graph breakdown.

Key caveat: The presenter stresses that you must review/audit, because graphs or formulas can be wrong sometimes—though they claim it’s often 80–90% correct initially.

2) ChatGPT (with “thinking”) to Generate an Excel Model with Dynamic Assumptions

Scenario: A CFO/finance manager needs an Excel financial model (e.g., revenue impact from changing prices and growth) that would otherwise take hours to build.

How it’s used:

  • Provide role/context (e.g., CFO), product structure, pricing, and assumptions.
  • Use dummy data if you don’t want to share confidential information.
  • Use ChatGPT with a “thinking model” (and “extended thinking”) to improve correctness—better formulas, graphs, and fewer mistakes.

Output and verification:

  • ChatGPT generates an Excel file in about ~4 minutes.
  • The model includes:
    • multiple months and product lines (e.g., consumer vs. enterprise),
    • linked formulas so changes in an assumptions sheet update charts dynamically.
  • The presenter emphasizes auditing; they claim their demo worked “from the first go.”

Key caveat: Even when AI outputs work, treat the result like a junior’s work—review it.

3) ChatGPT + Canvas to Turn Models into a Dynamic “Boss-Ready” Presentation

Scenario: After building a model, you need to present it (slides/story), which often requires extra time to make it visually compelling.

How it’s used:

  • Prompt ChatGPT for a “super modern,” “dynamic,” visually impressive” presentation model.
  • Use Canvas to build interactive tools (described as a less-known feature).

Example capability:

  • The preview shows an interface where the viewer can adjust key assumptions (e.g., monthly churn, price, starting subscribers).
  • Graphs update instantly.
  • Includes breakdowns (e.g., enterprise vs. consumer and totals).

Auditing approach: The underlying code is available; the presenter suggests using AI to help audit the code and confirm logic.

4) ChatGPT Agent Mode for Automating Messy “Consolidate and Clean” Departmental Data

Scenario: License usage/costs across many departments come in inconsistent formats (language differences, missing totals, different fields such as unit price vs. total).

Key warning (what not to do)

  • Don’t ask generically “consolidate this,” because the output may be consolidated but not cleaned, making pivot tables impossible.

Better workflow

  • Provide detailed context: you’re auditing licenses, describe the problem, and define a required output format.
  • Upload data using a ChatGPT Business/Enterprise license (the presenter notes compliance considerations and recommends checking company policy).

Agent mode approach

  • Use an agent mode workflow that can open/use a terminal process (described as the AI “looking at the file” and making sense of it).
  • Demo claim: about ~50 seconds to generate a properly structured spreadsheet.

Output includes:

  • normalized rows with department, vendor, license type, usage, unit price, total cost, and utilization ratio.

Spot-checking:

  • The presenter verifies calculations—for example, cases where the original department document omitted total cost, checking AI’s computed totals against expected results.

Generalization: The same approach can apply to PDF → Excel extraction and transformation via agent mode.

5) Claude (Code-Based Excel Integration) for Complex Models (DCF) + Sensitivity/Graphs

Scenario: Build a discounted cash flow (DCF) model for real estate—typically involves starting from scratch, checking formulas, and building assumptions/graphs.

How it’s used:

  • Prompt Claude to create a DCF model with dynamic assumptions and include graphs.
  • The presenter mentions Excel support for code inside Excel (noted as not free at the time of recording, but possibly later).

Demo features produced:

  • assumptions and calculated inputs,
  • property cash flows by year and sale proceeds,
  • key investment metrics,
  • sensitivity analysis,
  • graphs and visualizations.

Iterate + audit:

  • Claude may detect and attempt to fix mistakes.
  • If graphs are wrong, request improvements; the presenter claims graph updates can take 1–2 minutes.
  • The pattern is iterative: first draft → review → comment → iterate.

Main Speakers / Sources

  • Primary source: The video presenter/host (name not provided in the subtitles).
  • AI tools referenced:
    • Microsoft Excel “AI agent mode”
    • ChatGPT (including “thinking model” and Canvas)
    • Claude 4XL (Excel-integrated code / code-in-Excel workflow)

Original video