The number one reason AI projects get killed isn't technical failure. It's that nobody built a credible business case before starting. A clear ROI calculation protects you from two mistakes: investing in a project that can't pay for itself, and not investing in one that clearly can.
Here's the framework we use with every client.
The Basic ROI Formula
AI ROI comes down to a simple equation:
ROI = (Annual Value Created - Total Cost) / Total Cost x 100
The challenge isn't the math. It's accurately estimating the "Annual Value Created" and the "Total Cost." Let's break both down.
Calculating Annual Value Created
Value from AI comes in four categories. Most projects deliver 1-2 of these:
1. Labor Cost Savings
This is the most straightforward to calculate and the most common source of AI ROI.
Formula: Hours saved per week x Loaded hourly rate x 52 weeks
Example: An AI document processor saves your team 25 hours per week. Your average loaded hourly rate (salary + benefits + overhead) is $45/hour.
25 hours x $45/hour x 52 weeks = $58,500/year in savings
Important: "Loaded hourly rate" includes benefits, taxes, and overhead -- not just salary. For most SMBs, loaded rate is 1.3-1.5x the base hourly salary.
2. Revenue Increase
AI can increase revenue through faster response times, better lead scoring, or improved customer experience.
Formula: This varies by use case, but a conservative approach is:
Additional revenue = Current metric x Expected improvement percentage
Example: Your sales team closes 10% of leads. An AI lead scoring system helps them focus on the best prospects, improving close rate to 13%.
If your average deal is $15,000 and you get 200 leads/year:
- Current revenue: 200 x 10% x $15,000 = $300,000
- Projected revenue: 200 x 13% x $15,000 = $390,000
- Revenue increase: $90,000/year
Important: Be conservative with revenue projections. Use the low end of expected improvement ranges.
3. Error Reduction
Manual errors cost real money in rework, refunds, and customer churn.
Formula: Error rate x Volume x Cost per error
Example: Your team processes 5,000 orders per month with a 2% error rate. Each error costs $75 in rework and customer recovery.
Current annual error cost: 5,000 x 2% x $75 x 12 = $90,000 After AI (0.3% error rate): 5,000 x 0.3% x $75 x 12 = $13,500 Savings: $76,500/year
4. Speed Premium
Faster delivery often commands premium pricing or prevents revenue loss.
Formula: Deals lost or discounted due to slow delivery x Average value
Example: You estimate you lose 2 deals per quarter because competitors respond to RFPs faster. Average deal value is $20,000.
Speed premium: 8 x $20,000 = $160,000/year in recovered revenue
Important: This is the hardest to quantify accurately. Only include it if you have real data showing lost deals due to speed.
Calculating Total Cost
Total cost of an AI project includes more than just the development fee:
| Cost Category | Typical Range | Notes |
|---|---|---|
| Development / Build | $10,000 - $100,000 | Depends on complexity |
| Infrastructure (Year 1) | $1,200 - $12,000 | Cloud hosting, API costs |
| API / Model Costs | $600 - $6,000/year | OpenAI, Anthropic, etc. |
| Maintenance | $2,400 - $12,000/year | Updates, monitoring, fixes |
| Training / Onboarding | $1,000 - $5,000 | Getting your team up to speed |
| Total Year 1 | $15,200 - $135,000 | |
| Ongoing Annual | $4,200 - $30,000 |
For a typical SMB AI project, Year 1 total cost is usually $20,000-$60,000.
Putting It All Together: An Example
Let's say you're a professional services firm evaluating an AI system to automate client report generation.
Annual value created:
- Labor savings: 15 hrs/week x $50/hr x 52 = $39,000
- Error reduction: $8,000/year
- Speed premium (faster client delivery): $15,000/year
- Total annual value: $62,000
Total costs:
- Development: $30,000
- Year 1 infrastructure + API: $4,800
- Training: $2,000
- Annual maintenance: $6,000
- Year 1 total: $42,800
- Ongoing annual: $10,800
Year 1 ROI: ($62,000 - $42,800) / $42,800 = 44.9%
Year 2+ ROI: ($62,000 - $10,800) / $10,800 = 474%
The project pays for itself in about 9 months and delivers nearly 5x return from Year 2 onward. That's a strong business case.
Common ROI Pitfalls
Being too optimistic about savings
Use conservative estimates. If you think AI will save 30 hours/week, model it at 20. Under-promise and over-deliver.
Ignoring ongoing costs
AI systems need maintenance, monitoring, and occasional model updates. Budget 15-25% of the build cost annually for maintenance.
Forgetting the "do nothing" cost
Compare AI ROI not just against zero, but against the escalating cost of manual work as your business grows. If you're growing 20% year-over-year, your manual processing costs grow too.
Not accounting for implementation time
There's a ramp-up period (usually 2-4 weeks after launch) where the AI system is running alongside human processes. Factor this into your timeline and payback calculations.
The Decision Framework
| Year 1 ROI | Recommendation |
|---|---|
| > 100% | Strong investment. Move forward with confidence. |
| 50-100% | Good investment. Ensure you have the organizational readiness. |
| 20-50% | Marginal. Consider starting with a smaller scope. |
| < 20% | Weak case. Look for a higher-impact process to automate first. |
What Investors and Boards Want to See
If you need to present an AI business case to stakeholders, include:
- The problem: What manual process you're targeting and its current cost
- The solution: What the AI system will do (one paragraph, not technical)
- The numbers: ROI calculation using this framework
- The timeline: Implementation time and payback period
- The risk: What could go wrong and how you'll mitigate it
- The ask: Specific budget and timeline request
Keep it to one page. Decision-makers don't need a 40-slide deck.
Need help building a business case for AI at your company? Schedule a free strategy session and we'll help you run the numbers.