The Problem with AI ROI Conversations
“AI will transform your business.” “AI will save you thousands of hours.” “AI will 10x your productivity.”
These claims are useless for decision-making. They might be true. They might not. You can’t budget against them, and you can’t hold anyone accountable to them.
Real AI ROI calculation is straightforward if you’re willing to be specific. Here’s the framework.
The Three Categories of AI Return
AI generates return in three ways. Most calculations only count the first one.
Category 1: Cost Reduction
The easiest to measure. A human doing a task costs X per hour. The AI does that task in Y minutes. Savings = (X × hours_saved) per week.
Category 2: Revenue Acceleration
Harder but often larger. AI that converts 2% more leads, closes deals 30% faster, or enables your team to handle 40% more clients is directly generating revenue. Calculate: (additional_deals × average_deal_value) or (additional_capacity × revenue_per_capacity_unit).
Category 3: Error Reduction
Frequently ignored. Manual processes have error rates. Errors cost money — in rework time, in customer churn, in compliance issues. AI with lower error rates than humans generates return proportional to the cost of those errors.
The Full ROI Formula
Annual ROI = (Category 1 + Category 2 + Category 3 savings) − (AI system cost + maintenance cost + training time cost)
Divide by total investment for a percentage.
Anything above 200% annual ROI is exceptional. 100–200% is strong. 50–100% is acceptable for strategic investments.
A Real Example
A marketing agency with 8 people automates their client reporting:
Before: Each report takes 4 hours. 15 clients. Monthly = 60 hours. At $75/hour blended rate = $4,500/month in reporting labor.
After AI: Each report takes 45 minutes. Monthly = 11.25 hours. Labor cost = $844/month.
Category 1 savings: $3,656/month = $43,872/year
Category 2: With reporting done faster, team has capacity for 3 additional clients at $3,000/month each = $108,000 additional annual revenue.
Category 3: Manual reports had errors on average 1 per client per quarter. Each error required 2 hours to fix. 60 errors/year × 2 hours × $75 = $9,000 saved.
Total return: $43,872 + $108,000 + $9,000 = $160,872/year
AI system cost: $8,000 setup + $500/month = $14,000/year
ROI: ($160,872 − $14,000) / $14,000 = 1,049%
This is not a hypothetical. These are real numbers from a real deployment.
What to Measure Before You Build
Before starting any AI project, document:
- Current time spent on the target process (hours/week, who, at what rate)
- Current error rate and cost per error
- Current capacity constraints (what could you do with more time?)
- Realistic adoption assumption (will 70% of the team use this? 50%? 30%?)
These four numbers give you a realistic pre-build ROI estimate. If the math doesn’t work before you build, it won’t work after.
The Most Common ROI Mistake
Assuming 100% adoption. Real AI adoption in most organizations runs 40–70% in the first year, even with excellent implementation.
Run your ROI calculation at 50% adoption. If it still makes sense, the project is worth doing. If it only works at 100% adoption, you need a better implementation plan before you start.
