# Quality and Error Effects — Estimating AI Automation ROI

Source: https://www.skillbyai.com/en/ai-automation-roi/e-errors

> AI can reduce errors, or add them.

## Model the error rate in each scenario

Error costs can dominate the case. AI with human review often lowers error rates (consistent extraction, flagged anomalies), but AI **without** review can raise them through confident mistakes. Estimate error rates per scenario from a pilot, multiply by volume and cost per error, and include the review design in the case. This also shows why removing review to save time can destroy value.

## Error costs in three scenarios, run

I ran this with plain Python 3. All figures belong to one worked example (an invoice-processing team) with invented but internally consistent numbers; prices are placeholders. At 12,000 invoices and $25 per error, manual work at 4% costs $12,000 a month, AI with review at 1.5% costs $4,500, and AI without review at 6% would cost $18,000, worse than today.

```python
volume = 12000
scenarios = {  # error rate, cost per error
    "manual today": (0.040, 25.0),
    "AI + human review": (0.015, 25.0),
    "AI without review": (0.060, 25.0),
}
for name, (rate, cost) in scenarios.items():
    print(f"{name:<18} errors/month {volume * rate:>5.0f}  cost ${volume * rate * cost:>7,.0f}")
```

Output:

```
manual today       errors/month   480  cost $ 12,000
AI + human review  errors/month   180  cost $  4,500
AI without review  errors/month   720  cost $ 18,000
```

## Measure error rates in the pilot

Error rates vary widely by document type and model; use pilot data, not vendor claims.

**Quiz:** What can happen if human review is removed to save time?

- [ ] Costs disappear
- [ ] Errors always fall
- [x] Error costs can rise enough to wipe out the savings
- [ ] Volume doubles

*Answer:* Error costs can rise enough to wipe out the savings. Review protects value.