Video summary

3 Copilot Agents Just Replaced a Full Day Of Work (Here's Proof)

Main summary

Key takeaways

Finance

Overview: Automating Monthly Finance Close-Cycle Reporting

The video demonstrates a three-Copilot-agent workflow to replace manual monthly finance reporting with an accelerated, repeatable process:

  1. Copilot Analyst extracts patterns and anomalies from internal financial and transaction data (GL + POS).
  2. Copilot Researcher validates those trends using public web research, adding industry context and citations.
  3. A custom Copilot Studio agent converts the results into a repeatable CFO “one-pager” each month, using the same workflow against refreshed data.

A running example uses an illustrative business: “F9 Finance coffee shop” with three locations, 6 months of GL actuals vs. budget, and 150,000 POS transactions.

Instruments / Assets / Tickers

  • None explicitly mentioned (no stocks/ETFs/bonds/commodities/FX tickers cited in the subtitles).
  • Benchmarking is described indirectly via coffee shop / US coffee industry dynamics.

Key Internal Findings (Copilot Analyst)

From the uploaded F9 Finance data, the Analyst agent highlights qualitative results (no specific monetary figures are provided in the subtitles):

  • “Strong Q2 acceleration after weak start”
  • Margin expansion driven by scale, not pricing
  • No evidence of higher ticket sizes
  • Cost structure improving, but COGS (Cost of Goods Sold) creep is a warning sign
  • Locations: revenue performance is similar, while margins differ
  • POS data suggests a highly predictable and optimizable business, enabling actionable improvements

Internal drivers the Analyst output includes

  • Budget vs. actual changes
  • Location-level insights
  • Customer / POS behavior trends
  • Time-of-day and day-of-week patterns

Forecasting / reforecasting stance

  • The Researcher prompt explicitly avoids reforecasting budgets and instead focuses on whether observed internal trends are market-wide.

External Validation & Benchmarks (Copilot Researcher)

The Researcher agent performs:

  • Public web scanning (not a static knowledge base)
  • Longer runtime than Analyst, producing a report with:
    • Cited sources / footnotes
    • Trend-by-trend comparison of internal vs. industry
    • Qualitative benchmark categories such as:
      • Cost of sales
      • Net profit margin
      • Peak hour capacity

External conclusions emphasized (qualitative)

  • F9 Finance coffee shop trends closely mirror broader US coffee industry dynamics
  • Major themes include:
    • Rising costs, especially rising COGS
      • Explanation: soaring coffee bean prices and ingredient inflation
    • February underperformance presented as a seasonal pattern
    • Morning throughput optimization as a key operational focus
    • Notes on competitive moves (what similar companies are doing)

Explicit Numbers, Timelines, and Performance Metrics

Runtime and workflow speed (high-level claims)

  • Analyst execution: ~2 minutes
  • Researcher execution: ~15 minutes, 88 steps
  • Custom CFO update agent output: ~5 minutes with one click

Comparison to manual work (claims)

  • Analyst reduces manual effort from about ~half a day of pivot-table work
  • Researcher reduces ~3 to 5+ hours of manual research and writing
  • CFO one-pager manual effort: ~2 to 3 (up to 4–5) hours, depending on revisions
  • Total workflow savings/avoidance per month:
    • ~8 to 12 hours saved/avoided (varies by complexity/entities)

Demo dataset sizes

  • 6 months of GL data (actuals + budget)
  • 150,000 rows of POS transactions
  • Three locations in the chain
  • Demo references June 2023 as the target reporting month

Methodology: Step-by-Step Framework

Step 1: Analyst (internal pattern extraction)

  • Provide structured data:
    • GL: actuals + budget
    • POS transactions
  • Prompt asks the model to identify:
    • Key trends
    • Things to be aware of
  • Output includes:
    • Executive summary
    • Trend list
    • Risks / opportunities
    • Suggested next steps

Step 2: Researcher (market/industry context + citations)

  • Copy Analyst risks/opportunities into the Researcher prompt
  • Ask whether trends are business-specific or industry-wide
  • Output:
    • Cited research report
    • Trend explanations, seasonality confirmation, benchmarks
  • Limitation noted:
    • No access to paywalled sources

Step 3: Custom agent in Copilot Studio (repeatability + CFO one-pager)

  • Define a repeatable workflow:
    • internal analysis
    • external benchmarking
    • generate CFO one-pager narrative format
  • Automate data pulls (e.g., SharePoint / OneDrive) so the agent runs with one click monthly
  • Include guardrails for:
    • tone/style
    • formatted outputs with citations/assumptions

Recommendations, Cautions, and Disclosures

  • Not financial advice (The video focuses on automating finance analysis and reporting workflows, not investment guidance.)

  • Credibility / verification caution

    • Presenter advises spending ~5 minutes checking sources before sharing with leadership to reduce hallucination risk.
  • Tool limitations
    • Analyst: can find patterns but may not identify root causes if causes aren’t present in GL data—requires human judgment.
    • Researcher: cannot access paywalled sources (examples: Bloomberg, Gartner, trade associations) and may have incomplete coverage due to public-only sourcing.
    • Custom agent: can generate confident narratives even if underlying data is incorrect—so human review is required (“The agent does the mechanical work, you do the judgment.”).

Presenters and Sources Mentioned

  • Presenter: Mike (sign-off: “this is Mike signing off from F9 Finance”)
  • Tooling: Microsoft Copilot (copilot.microsoft.com / Microsoft Teams agents)
  • Paywalled sources referenced as excluded/limited:
    • Bloomberg
    • Gartner
    • Industry trade association sources

Original video