We deployed AI customer service chatbots for 25+ businesses. Here is what works, what breaks, and what it costs.
By David Friedman, Founder of AppBrewers
Customer service is the most expensive department for most small businesses. A single support agent costs 2,000-3,500 Euro per month. An AI chatbot costs 29-149 Euro per month and handles 80% of inquiries without human intervention. Here is what we learned deploying AI customer service for 25+ businesses in 2025.
What an AI Customer Service Chatbot Does
| Task | How AI Handles It | Human Needed? |
|---|---|---|
| Order status inquiries | Reads database, replies instantly | No |
| FAQ (hours, location, policies) | Trained on knowledge base | No |
| Returns and refunds | Guides through policy, initiates process | Sometimes |
| Complaints | Detects sentiment, escalates to human | Yes |
| Complex technical issues | Collects info, creates ticket | Yes |
| Upselling and cross-selling | Suggests products based on history | No |
Real Costs
DIY (OpenAI + Custom Build)
| Component | Monthly Cost |
|---|---|
| OpenAI API | 50-200 Euro |
| WhatsApp Business API | 0-50 Euro |
| Development (one-time) | 5,000-15,000 Euro |
| Hosting | 20-100 Euro |
| Maintenance | 500-1,000 Euro |
| Year 1 total | 12,000-28,000 Euro |
Done-For-You (Conversify)
| Plan | Monthly Cost | Conversations |
|---|---|---|
| Starter | 29 Euro | 500 |
| Pro | 79 Euro | 2,000 |
| Business | 149 Euro | 5,000 |
| Enterprise | 499 Euro | Unlimited |
Year 1 total: 348-5,988 Euro
Real Results
| Business | Before AI | After AI |
|---|---|---|
| E-commerce store | 12 support tickets/day, 2 agents | 2 tickets/day, 1 agent |
| SaaS company | 4-hour response time | 8-second response time |
| Dental clinic | 20 calls/day to front desk | 5 calls/day, rest automated |
| Restaurant chain | 50 DMs/day unanswered | 48 automated, 2 escalated |
How to Set It Up
Step 1: Choose Your Channel
| Channel | Best For | Setup Time |
|---|---|---|
| Local businesses, Europe, Asia | 30 min | |
| Instagram DM | Visual brands, younger customers | 30 min |
| Website chat | E-commerce, SaaS | 15 min |
| Facebook Messenger | Older demographics | 30 min |
Step 2: Build Your Knowledge Base
Upload:
- Product descriptions and pricing
- Shipping and return policies
- Common troubleshooting steps
- Business hours and location
- FAQ (minimum 50 questions)
Step 3: Configure Escalation Rules
AI should hand off to human when:
- Customer uses angry language
- Issue is not in knowledge base
- Customer asks for a manager
- Order value exceeds threshold
Step 4: Test Before Launch
Run 50 real customer scenarios through the bot. Fix anything that feels robotic or unhelpful.
Common Mistakes
No Human Handoff
AI handles 80% well. The other 20% need a human. If customers cannot reach a person, they leave bad reviews.
Over-Automating
Some businesses try to automate everything including complex technical support. This frustrates customers. Automate the simple 80%. Human handles the complex 20%.
Not Updating the Knowledge Base
Products change. Policies change. An AI trained on outdated information gives wrong answers. Update monthly.
Need Help?
We have deployed AI customer service for 25+ businesses.
Originally published on the AppBrewers Blog.
Top comments (0)