Let's talk numbers. A single full-time customer service agent costs $2,000-4,000/month in most markets — salary, benefits, training, management overhead. For a small company doing 100-200 customer conversations per day, you need at least 2-3 agents to cover business hours. That's $6,000-12,000/month just to answer "Where's my order?" and "Can I return this?"
Now here's the alternative: an AI customer service agent running on Tencent Cloud Lighthouse costs roughly $30-60/month for the infrastructure plus $50-200/month in LLM API costs, depending on volume. That's a 90%+ cost reduction — and the AI works 24/7 without sick days, training ramp-up, or turnover.
If you're a small company watching your customer service budget eat into your margins, this is the most impactful operational change you can make.
Before we dive into the solution, let's be honest about what customer service actually costs a small company:
| Cost Category | Monthly Estimate |
|---|---|
| Agent salaries (2-3 people) | $4,000-12,000 |
| Training and onboarding | $500-1,000 (amortized) |
| Software tools (helpdesk, CRM) | $100-500 |
| Management overhead | $500-1,000 |
| Missed sales (slow response) | Hard to quantify but real |
| Total | $5,100-14,500/month |
And that's just for business hours coverage. If you want 24/7 support, double or triple the agent costs.
Here's what the AI alternative looks like:
| Cost Category | Monthly Estimate |
|---|---|
| Tencent Cloud Lighthouse instance | $30-60 |
| LLM API tokens | $50-200 |
| Setup time (one-time, amortized) | ~$0 (self-service) |
| Maintenance | ~2 hours/month |
| Total | $80-260/month |
That's not a typo. The infrastructure cost for a 24/7 AI customer service agent is less than what most companies spend on coffee for their CS team.
Visit the Tencent Cloud Lighthouse Special Offer page:
For small companies, the 2-core, 2 GB RAM tier is the entry point — it handles up to 100 daily conversations comfortably. If you're doing 100-500 conversations, step up to the 2-core, 4 GB RAM tier.
The key to keeping costs low is smart model selection and prompt engineering.
ssh ubuntu@<your-instance-ip>
# Run setup
clawdbot onboard
# Disclaimer → Yes
# Mode → QuickStart
# Config → Use existing values
# Model → DeepSeek (best cost-to-quality ratio)
# Channel → Your primary messaging channel
# Hooks → session-memory only
Model choice is your biggest variable cost. Here's a practical comparison:
| Model | Cost per 1M tokens | Quality | Best For |
|---|---|---|---|
| DeepSeek | ~$0.14-0.28 | Very good | High-volume routine CS |
| GPT-4o-mini | ~$0.15-0.60 | Good | Budget-conscious, decent quality |
| GPT-4o | ~$2.50-10.00 | Excellent | Complex queries (use sparingly) |
| Qwen | ~$0.14-0.70 | Very good | Bilingual scenarios |
For most small companies, DeepSeek offers the best balance. It handles routine customer service queries well at a fraction of the cost of premium models.
Critical: Never hardcode your API keys in configuration files or scripts. Use the Tencent Cloud console's visual panel for secure credential management.
Every token costs money. Keep system prompts under 200 tokens (cut the fluff), set response length limits (2-3 sentences for routine queries), and use structured knowledge base entries (bullet points over prose). A lean prompt costs half as much per request as a bloated one.
Small companies typically need 1-2 channels. Start with your highest-volume platform:
# Most common for small businesses: WhatsApp
clawdbot onboard
# → WhatsApp (QR link) → Scan QR → Approve pairing
openclaw pairing approve whatsapp <code>
Add more channels as you grow. Each additional channel costs nothing extra — it's the same instance, same model, same knowledge base.
Channel guides: WhatsApp | Telegram | Discord
For small companies, the most cost-effective approach isn't pure AI — it's AI handling 60-70% of conversations with a part-time human agent covering the rest.
Here's the math:
| Scenario | Monthly Cost | Coverage |
|---|---|---|
| 3 full-time agents | $6,000-12,000 | Business hours only |
| OpenClaw + 1 part-time agent | $500-800 | 24/7 AI + human for complex issues |
| Savings | $5,200-11,200/month | Better coverage |
The AI handles routine queries around the clock. The part-time human handles escalations during business hours. Customers get faster responses, and your costs drop by 80%+.
Even the cheapest Lighthouse tier gives you enterprise-grade reliability. Enable daemon mode:
loginctl enable-linger $(whoami) && export XDG_RUNTIME_DIR=/run/user/$(id -u)
clawdbot daemon install
clawdbot daemon start
clawdbot daemon status
This ensures your agent auto-restarts after crashes and reboots — no manual intervention needed.
After your first month, calculate the actual savings:
Most small companies see payback within the first week and ongoing savings of $5,000-10,000+ per month.
The beauty of this setup is that it grows with you. More conversations? The same instance handles them (just pay more in API tokens). Need more channels? Run clawdbot onboard again. Need more compute? Upgrade your Lighthouse tier with one click. You're never over-provisioned or under-provisioned.
Every month you delay is another $5,000-10,000 in unnecessary CS costs. Visit the Tencent Cloud Lighthouse Special Offer page, select OpenClaw (Clawdbot) under AI Agent, and click "Buy Now". In 30 minutes, you'll have an AI customer service agent that costs less per month than a single day of human agent salary — and it works 24/7.