Altitude 7
ModellingWorkflow

Update a Model from a Transcript

Last updated April 27, 2026

What it does

Open your model. Point Cowork at the earnings transcript. Cowork updates the revenue build assumptions for next quarter using management's guided ranges, highlights cells you changed in yellow, and leaves your prior assumptions in a comment so you have a full audit trail.

What you need first

  • ·Your model open in Excel with the Claude Excel add-in installed
  • ·The transcript saved as a text or PDF file in a folder Cowork can see
  • ·A clear convention for which tabs hold the assumptions (e.g., Assumptions and Revenue Build)

The prompt

I just dropped the {TICKER} {QUARTER} earnings transcript in the folder.

Update the open workbook as follows:

1. On the Assumptions tab, update the next-quarter revenue growth, gross margin, and opex assumptions to reflect management's guidance ranges from the call.

2. Use the midpoint of management's guided range. If they gave a directional comment ("modestly higher", "in line with prior") without numbers, leave the assumption alone and flag it in a comment.

3. For every cell you change:
   - Highlight the cell in yellow (RGB 255,255,0)
   - Add a cell comment containing: (a) the prior value, (b) the quote from the transcript that justifies the change, (c) the page or timestamp.

4. On the Revenue Build tab, propagate the new assumptions and confirm formulas still reference the right cells.

5. At the end, summarize in chat: which 5 assumptions changed, by how much, and which 3 you flagged for me to decide on manually.

Don't touch any cell outside Assumptions and Revenue Build. Don't change formulas — only inputs.

What good output looks like

  • Workbook updated in place — formulas intact, inputs changed
  • Yellow cells you can scan visually
  • Cell comments with prior value, justification quote, and source location
  • Short summary in chat that you can paste into your update note

Make it a Skill

Save as model-update-from-transcript. The instructions on highlighting, commenting, and not touching formulas are the discipline that protects model integrity — encode them once, get them every time.

Common pitfalls

  • Always work on a copy first. The first three runs especially. Once you trust the discipline, work in place.
  • Define “guidance.” Be explicit about what counts as guidance you'll act on (numbers, ranges) versus directional commentary you'll only flag.
  • Don't let it touch formulas. The prompt above is explicit about this — keep it that way.