Calculating AI ROI
By John Martines · March 20, 2025 · 8 min read
A practical framework for measuring return on AI investment — time saved, errors reduced, revenue gained.
You've deployed an AI tool. Now your CEO is asking: "What's the return?" If you don't have a clean answer, you're going to struggle to justify the next tool or get budget for scaling.
This guide gives you a straightforward framework for calculating AI ROI that actually makes sense for SMBs.
AI ROI has four components:
- Time Saved: Reduced hours for staff to do the same work
- Error Reduction: Fewer mistakes, rework, and quality issues
- Revenue Impact: Faster decisions, more deals, higher margins
- Total Costs: Tool cost + implementation + training + opportunity cost
Not all four apply to every project. Most projects have strong time or error savings. Some also have revenue upside.
The Calculation
Annual Time Savings = Hours Saved Per Task × Frequency Per Year × Loaded Cost Per Hour
Example: Invoice Processing
Your accounting team currently spends 2 hours processing each invoice. You deploy an AI invoice processing tool.
- Before: 2 hours per invoice (includes email receipt, data entry, categorization, routing)
- After: 15 minutes per invoice (AI does extraction, categorization, and routing; human reviews and approves)
- Time saved per invoice: 2 hours – 15 minutes = 1 hour 45 minutes = 1.75 hours
- Invoices per year: 8,000
- Annual hours saved: 1.75 × 8,000 = 14,000 hours
- Loaded cost per hour: $35 (fully burdened accounting staff cost)
- Annual time savings: 14,000 hours × $35 = $490,000
That's before any other benefits. Time savings alone justify the investment.
What Counts as "Loaded Cost"?
Don't just use salary. Include:
- Hourly wage or salary / 2,080 hours per year
- Plus payroll taxes (7.65% for employers)
- Plus benefits (health insurance, retirement, PTO) — typically 20–30% of salary
- Plus overhead allocation (desk space, software, equipment) — typically 10–20%
For a $50K/year employee, loaded cost is typically $50K × 1.40 = $70K/year = $33.65/hour.
Be Conservative
Most companies overestimate the amount of time truly saved because:
- Employees don't immediately fill freed time with billable or value-add work
- Some time is transition, training, and ramp-up
- You may not be able to reduce headcount immediately
A good rule of thumb: take your calculated time savings and apply a 60–70% realization factor. You'll realize 60–70% of the theoretical savings.
In the invoice example: $490,000 × 65% = $318,500 conservatively realized benefit.
The Calculation
Error Reduction Value = (Errors Per Year × Cost Per Error) × Error Reduction % × Error Detection Rate
Example: Expense Categorization
Your finance team manually categorizes 10,000 expenses per year. Current error rate is 8% (800 errors).
Each miscategorization costs:
- $15 in rework time (supervisor review and correction)
- $10 in audit risk (potential audit discovery and remediation)
- Total cost per error: $25
- Current annual error cost: 800 errors × $25 = $20,000/year
- After AI categorization: Error rate drops to 1% (100 errors)
- Error reduction: 700 errors avoided
- Annual error reduction value: 700 × $25 = $17,500/year
The Tricky Part: What Counts as an "Error"?
Be specific about what errors you're measuring:
- Categorization errors: Wrong GL account assigned (causes variance reports to be wrong)
- Duplicate detection: Same invoice processed twice (causes overpayment)
- Compliance errors: Missing documentation or approvals (audit risk)
- Data entry errors: Wrong amount, vendor, or date
The cost of an error varies dramatically by type. A duplicate invoice caught before payment costs nothing. A duplicate invoice paid and then recovered costs hundreds.
Common Error Reduction Values by Use Case
- Invoice processing: 3–8% error reduction (good data extraction + human review)
- Expense categorization: 7–12% error reduction (AI learns your rules)
- Customer data cleanup: 5–15% error reduction (AI finds duplicates and inconsistencies)
- Clinical documentation: 2–5% error reduction in completeness (AI fills gaps)
Time and error reduction are tangible. Revenue impact is harder to measure but often bigger.
The Mechanisms
- Faster decisions: If AI gets you faster data, you can make decisions quicker
- Higher close rates: If AI handles first-level customer service, more leads become customers
- Bigger deals: If AI frees up time, salespeople can pursue bigger opportunities
- Margin improvement: If estimates are faster and more accurate, you win more deals with better margins
- Retention: If AI improves customer experience, fewer customers churn
Example: Estimating AI for a Contractor
A contractor currently writes 8 estimates per week, 3 hours each. Current win rate: 30% (2.4 wins/week).
A free account opens the rest of this article and the whole library.