AI vendors love big ROI numbers — here is how to verify them
Every AI vendor has a case study claiming 300% ROI. Those numbers are not necessarily fake — but they are almost certainly cherry-picked from the most favorable deployment scenario. Your business is not the case study.
The ROI formula that actually works
Step 1: Calculate the current cost of the process
Labor cost: Hours spent multiplied by fully loaded hourly rate. Error cost: Average cost per error times errors per month times 12. Delay cost: Revenue lost or penalty incurred due to process delays. Opportunity cost: What could the team accomplish if they were not doing this manual work?
Annual baseline cost equals labor plus errors plus delays plus opportunity.
Step 2: Estimate the AI solution cost (all-in)
- Build or license cost
- Integration cost
- Data preparation cost
- Training and change management
- Ongoing costs: model monitoring, retraining, infrastructure — annual
Step 3: Model realistic improvement — not the vendor’s best case
Use conservative assumptions. Automation rate: 60% is realistic for most workflows — not 100%. Error reduction: assume 50–70%. Adoption rate: assume 70% in year one. Time to value: assume 50% of target improvement in months 1–3, 80% in months 4–6.
Step 4: Calculate three scenarios
Pessimistic: Everything takes longer, costs more, delivers less. Realistic: Middle-ground assumptions. Optimistic: Best case grounded in data. If the pessimistic scenario still shows positive ROI within 18 months, the investment is low-risk.
What most ROI calculations miss
Intangible benefits
- Employee satisfaction: Reducing tedious manual work improves retention
- Customer experience: Faster response times and fewer errors make customers happier
- Scalability: AI handles volume without linear headcount growth
- Competitive positioning: Being more efficient than competitors wins deals
Intangible costs
- Technical debt: Budget 15–20% of initial build cost annually for maintenance
- Vendor lock-in: Factor in the switching cost
- Regulatory risk: AI regulations are evolving — budget for compliance adjustments