What can be scripted, must be scripted.
AI models work on predictive completion: they produce the most likely answer, which means it's probably right, but it could be wrong. "Probably" doesn't hold up in financial reporting or business metrics. Wrong numbers lead to wrong decisions. The calculations and the rules belong in code, where the same input gives the same answer every single time.
That does not make AI any less useful. It means using the right tool for the job. AI must still be leveraged, especially for:
- analysis and commentary
- exception handling
- reconciliation reports
- building the deterministic scripts in the first place
Let the model write the script, and let the script do the arithmetic.