What makes Microsoft Excel easy or hard to automate
Once a workbook is in OneDrive or SharePoint, Microsoft Graph treats it as a set of resources: worksheets, ranges, tables, named items and charts. A workflow can append rows to a table, read a calculated cell, or update inputs and read the result after recalculation. That last pattern is useful when a pricing or forecast model already lives in Excel and nobody wants to rebuild it in code.
Tables are the key to reliable automation. A formatted Excel table has named columns and grows as rows are added, so the workflow can append by column name rather than guessing the next empty row.
The hard parts are location and scale. Many companies still run critical workbooks from a desktop or file server, which Graph cannot reach. Large workbooks with heavy formulas are slow to open and recalculate through the API, and timeouts are common. For bulk data work, downloading the file and processing it in code is often faster than many small API calls.
The AI layer
Excel holds the numbers and the logic. AI adds the parts formulas cannot:
- Cleaning inputs. Customer names, product codes and categories from several sources rarely match. A model proposes standard values for review.
- Writing commentary. After totals are computed and checked, a model writes the variance notes finance teams usually type by hand.
- Reading unstructured inputs. A model can turn emailed requests or PDFs into the rows a workbook expects.
Keep calculations in formulas. The model fills inputs and explains outputs.
Limits to plan around
- Graph throttling. Excel API calls are limited per app per tenant (about 1,500 requests per 10 seconds) and per app overall. Batch reads and writes into ranges.
- Workbook sessions. Use a persistent session for a series of edits, or changes may be slow and inconsistent. Close sessions when done.
- Concurrent editing. If people edit while a workflow writes, conflicts and recalculation delays follow. Schedule runs when the file is quiet or write to a separate table.
- File format. The workbook API works on .xlsx. Convert .xls and handle .csv as plain files.
How Lumeor typically connects it
For most Microsoft 365 clients we move the workbook to a SharePoint library, convert the data area to a table, and run n8n or Power Automate with an Entra ID app scoped to that site. LLM steps clean inputs or write commentary, and totals are checked in code before anything is saved. For heavy processing we download, process with Python, and upload a new version. See n8n vs Power Automate for the platform choice, and the month-end close reporting guide for a full example. Teams on Google Workspace should read our Google Sheets page.