AI for customer service: how to calculate ROI for your SMB
Erwin Berkouwer
An AI chatbot or agent for customer service sounds attractive, but what does it really deliver? Use this calculation model to figure out ROI in 10 minutes.
More and more SMBs are exploring AI for customer service — a chatbot on the website, an agent handling emails, or a copilot for your support desk. The big question remains: what does it really deliver? In this article you get a working calculation model based on common benchmarks for SMEs. These are indicative rules of thumb — your own numbers determine the real outcome.
The Four Cost Components of Customer Service
Before you can calculate profit, you need to know where your money is now. In an average SMB customer service team, the cost breakdown looks like this:
- Personnel costs: typically 70-80% of total budget
- Software and tooling: helpdesk system, phone service, knowledge base
- Overhead: management, training, quality control
- Turnover and recruitment: average €8,000-€15,000 per departing employee
An AI implementation mainly affects the first line item. The other components are secondarily impacted.
Step 1: Determine Your Current Volume and Costs
Gather these figures from the last 3 months:
- Number of tickets/contacts per month
- Average handling time (AHT) per ticket
- Hourly rate per employee (incl. employer taxes — estimate €40-€60 for SMBs)
- Percentage of repetitive, predictable questions
Example: 1,200 tickets per month × 8 minutes AHT × €50 hourly rate = €8,000 per month in handling costs.
Step 2: Determine the Deflection Percentage
Not every ticket is suitable for AI. As an indicative rule of thumb, in practice you often see:
- Simple FAQ questions: 70-85% fully resolved by AI
- Status and order updates: 80-95% (if well integrated with your ERP)
- Refunds, complaints, complex issues: 10-30% (AI does prep work, human decides)
- Sales-related questions: 40-60%
A healthy mix often yields 35-55% full deflection within 3 months.
Step 3: Calculate the Savings
Take the figures from step 1 and apply the deflection percentage.
Example:
- 1,200 tickets × 45% deflection = 540 tickets/month handled by AI
- 540 × 8 min × €50/hour = €3,600/month savings
- On annual basis: €43,200
Don't forget the "shadow benefit": less wait time for customers, higher NPS, and the ability to grow with the same team without additional FTE.
Step 4: Subtract the Investment
A realistic investment for an SMB implementation:
- One-time: €4,000 - €15,000 (depending on complexity and integrations)
- Monthly: €300 - €1,200 (LLM costs, hosting, maintenance)
In the above example: at €10,000 one-time + €600/month you're at payback in about 4 months.
What Most Calculators Forget
Three cost items often missing from vendor cases:
- Maintaining content: someone needs to keep the knowledge base current (budget 4-8 hours/month)
- Quality assurance: review of AI answers, especially in the first 3 months
- Escalation routing: if AI fails, it must smoothly escalate to a human — otherwise NPS scores drop
Budget at least 10% extra on your monthly spend for these.
Conclusion
AI in customer service pays for itself in SMBs typically within 4-8 months. Crucial is to calculate thoughtfully: choose the right use cases, measure from day one, and invest in both content and escalation paths. Want help with your business case? Schedule a free consultation.
