Hiring a support person costs $3,500-$5,000 per month in the US (Glassdoor, 2025). For a 10-person startup burning $40K/month, that is a significant line item. And if you hire before building systems, your new hire will spend most of their time answering the same 20 questions over and over.
There is a better sequence. Build the support infrastructure first. Document your answers. Automate the repetitive stuff. Then hire when you actually need a human for complex, relationship-driven conversations.
Here is exactly how to do it.
The Support Stack You Need (Total: $49-99/month)
Three components replace 60-80% of what a full-time support person does:
- Knowledge base. A public, searchable collection of help articles on your website. Customers find answers themselves. Google indexes the content. AI assistants like ChatGPT and Perplexity cite it.
- AI chatbot. Connected to your knowledge base, it answers questions in natural language 24/7. No scripted flows. No decision trees. It reads your articles and responds conversationally.
- Live chat widget. For the questions AI cannot answer, customers reach you directly. This replaces the back-and-forth of email with real-time conversation.
All three work together. A customer opens your widget, asks a question, gets an AI answer. If the answer is not good enough, they click "talk to a human" and land in your inbox. One flow, one tool, one monthly bill.
Step 1: Audit Your Questions (1-2 Hours)
Before you write anything, you need data. Go through your last 100 support emails, chat messages, or DMs. Create a spreadsheet with two columns: the question and how many times it appeared.
You will notice a pattern immediately. A small number of questions account for most of your volume. Research from Zendesk (2024) shows that the top 10 questions typically represent 40-50% of all support requests.
Sort by frequency. Your top 20 questions are your first 20 knowledge base articles.
Common categories across most businesses:
- Getting started. How do I sign up, set up, connect, configure?
- Billing. How do I upgrade, cancel, get a refund, update my card?
- Features. How does X work, where do I find Y, can I do Z?
- Troubleshooting. Why is X not working, how do I fix Y?
- Account. How do I change my email, reset my password, delete my account?
Step 2: Write Your First 10-20 Articles (4-6 Hours)
Open your email sent folder. Find your best reply for each of the top 20 questions. Copy it into a document. Clean it up. Add a clear headline. Done.
Writing rules that make articles effective:
One question per article. Do not combine "How do I upgrade?" and "How do I cancel?" into one article. Separate articles rank better in search and give AI chatbots clearer answers to work with.
Start with the answer. The first sentence should solve the problem. Details, context, and edge cases come after. Customers scanning for answers will leave if the first paragraph is background information.
Use screenshots for complex steps. If the process involves clicking through multiple screens, add annotated screenshots. A 5-step process with images takes 2 minutes to follow. The same process described in text takes 10 minutes and usually results in a follow-up question.
Keep paragraphs short. Three sentences maximum. Online readers scan, they do not read line by line.
You do not need to write all 20 in one sitting. Start with 10. Publish them. Add more as you see new questions come in.
Step 3: Connect the AI Chatbot (30 Minutes)
Once your articles are published, connect an AI chatbot to your knowledge base. The setup is straightforward with most modern tools.
With Helpable, the AI (called Calli) automatically reads every published article. There is no separate training step. Publish an article about your return policy, and the chatbot can answer return questions immediately.
Configure these settings:
- Welcome message. Something simple: "Hi! Ask me anything or browse our help articles."
- Fallback behavior. When the AI does not know the answer, it should offer to connect the customer with a human.
- Language. If you serve multiple markets, enable auto-detection. Helpable supports 50+ languages and detects the customer's language from their first message.
Test it yourself before going live. Ask the 20 questions from your list. If the AI answers 15+ correctly, you are ready.
Step 4: Add the Widget to Your Site and Product (15 Minutes)
The chat widget is a small button (usually bottom-right of the screen) that opens the chatbot and live chat interface. Two places where it matters most:
Your marketing website. Visitors with pre-purchase questions get instant answers. This directly impacts conversion. According to Forrester (2023), customers who use live chat are 2.8x more likely to convert than those who do not.
Inside your product. Users hit problems while working. If they have to leave your app to find help, many will not bother. An in-app widget reduces friction to near zero.
Add the widget by pasting a small JavaScript snippet into your site's HTML. Most tools provide a single line of code that loads asynchronously and does not affect page speed.
Step 5: Set Up Your Inbox (15 Minutes)
When AI cannot resolve a question, it lands in your inbox. Configure a few basics:
- Email notifications. Get notified on new conversations so you do not miss urgent questions.
- Slack integration. If your team lives in Slack, forward new conversations there. Reply from Slack or from the dashboard.
- Business hours. Set expectations. If you respond 9-5 on weekdays, say so. Customers tolerate longer waits when they know the timeline.
You do not need complex routing, SLA rules, or priority queues yet. A single shared inbox works until you hit 20+ human tickets per day.
Step 6: Track and Fill Gaps (Ongoing, 30 min/week)
Your knowledge base is not finished after the initial 20 articles. New questions appear as your product evolves and your customer base grows.
Track these metrics weekly:
- Zero-results queries. What did customers search for that returned no articles? Each zero-result is a new article waiting to be written.
- AI deflection rate. What percentage of conversations were resolved by AI alone? If it is below 40%, your knowledge base has gaps.
- Most-viewed articles. Which topics get the most traffic? These might need more detail, related articles, or video walkthroughs.
Helpable's analytics dashboard tracks all three. Write one new article per week based on the data. After three months, your deflection rate will be significantly higher than on day one.
When to Hire Your First Support Person
The system above handles 60-80% of support volume. But at some point, you will need a human. Here are the signals:
AI deflection rate plateaus below 50%. If you have 40+ articles and the AI still cannot handle half the questions, your product might require more hands-on support than documentation can provide.
CSAT drops below 70%. Use post-conversation surveys to track satisfaction. If customers rate their experience poorly, the AI might be giving incomplete answers or the wait time for human help is too long.
Response time consistently exceeds 4 hours. When you cannot keep up with the human tickets alongside your other work, quality suffers. This is the clearest hiring signal.
You have budget for $3,500+/month. A support hire only makes sense when you can sustain the cost. Until then, $49-99/month for an AI-powered support stack is a better use of limited funds.
The Cost Comparison
| DIY support stack | First support hire | |
|---|---|---|
| Monthly cost | $49-99 | $3,500-5,000 |
| Availability | 24/7 (AI) | 8 hours/day |
| Languages | 50+ automatic | 1-2 fluent |
| Scales with traffic | Yes (AI handles spikes) | No (one person = one person) |
| Setup time | One weekend | 2-4 weeks (hiring + training) |
The smart sequence: build the stack first, hire second. Your future support person will be more effective with documentation and AI already in place. They will handle complex cases instead of typing the same password-reset instructions for the hundredth time.
FAQ
Can I really handle support without any team members?
Yes, up to a point. Solo founders running B2B SaaS products with up to 500 customers regularly manage support with a knowledge base and AI chatbot. The key is having good documentation. If your articles are thorough, AI handles 50-70% of questions without human input.
How many articles do I need before launching the AI chatbot?
Start with 10-15 articles covering your most frequent questions. That is enough for the AI to handle the basics. You do not need 100 articles on day one. Add more weekly based on what customers ask.
What happens when someone asks a question outside business hours?
The AI chatbot responds instantly, 24/7. If the AI cannot answer, the question sits in your inbox until you are online. Set a business-hours message so customers know when to expect a human reply.
Should I use email or live chat for support?
Both. Live chat (with AI) handles quick questions and reduces email volume. Email (or a contact form) catches longer, more detailed requests. The widget combines both: AI chat first, human chat or email fallback.
What tools do I need beyond the support platform?
For most small teams: Slack (for notifications) and your support platform. That is it. You do not need a separate ticketing system, CRM integration, or phone setup until you have a dedicated support team.