AI vs Excel. What Finance Teams Should Automate First.
Excel is not the problem and AI is not the replacement. The question is which parts of your spreadsheet work are actually modelling and which parts are just moving data.
6 min read
- Data collection
- Automate first
- Excel, manual
- Modelling
- Keep in Excel
- Excel
- Reporting
- Automate second
- Manual rebuild
Every few months someone announces that AI will kill the spreadsheet.
It will not, and it should not.
Excel is a genuinely good modelling environment and finance teams are fluent in it.
What Excel is bad at is being a data pipeline, an approval system, and a reporting tool at the same time.
That is the part worth automating.
§01 · The problem
Where the time actually goes.
- 01Copying data from systems into sheets by hand
- 02Rebuilding the same report every period
- 03Version conflicts across shared workbooks
- 04Broken links when a source file moves
- 05Formulas nobody remembers writing
- 06Approvals tracked in a tab called Status
Keep the modelling in Excel. Automate everything that happens before and after it.
§02 · The mechanism
The AI workflow.
- 01
Source systems connected once
Automation - 02
Data cleaned and normalised
AI Agent - 03
Sheet populated automatically
Automation - 04
Analyst models in Excel as before
Human - 05
Outputs published to the report
Automation - 06
Anomalies flagged against prior periods
AI Agent
- Automation
- Deterministic. No model involved
- AI Agent
- Model extracts, classifies, matches, or routes
- Human
- A judgement call only a person makes
§03 · The difference
Before vs after.
| Before | After |
|---|---|
| Export, paste, reformat, repeat | Data lands in the sheet automatically |
| Report rebuilt from scratch each month | Report refreshes from the same source |
| Four versions of the same workbook | One workbook, one source of truth |
| Analyst spends the day on data prep | Analyst spends the day on analysis |
| Errors found by whoever reads it last | Anomalies flagged before anyone reads it |
| Approval status lives in a colour code | Approvals tracked in a real workflow |
§04 · The measure
Time saved.
| Task | Before | With AI | Comparison |
|---|---|---|---|
| Data collection | 6 hrs/month | Automatic | |
| Report rebuild | 4 hrs/month | 15 min | |
| Version reconciliation | 2 hrs/month | None | |
| Time on actual analysis | 20% | 70% |
Continue reading
Tell us what you’re building.
Share a few details about what you need, and we’ll get back to you with a quotation.