Shipping a chatbot into a small business marketing stack is easy; making it actually produce leads is not. After building several AI receptionists I settled on a simple split of labor between the bot and the humans around it
What bots genuinely do well
Fixed short scripts: booking, reminders, the three questions every client asks. The bot answers at 3 AM without making anyone wait, and the customer rarely notices the swap. Speed of first reply is the one marketing metric a small business can move overnight with a bot
Where they fall apart
Long persuasive conversations. Three or four steps into a script the bot loses the thread, and at the "why are you pricier than the other place" stage it loses every single time. Selling stays human work for now, and pretending otherwise costs real customers
The roll-out order that worked for me
Delegate the most frequent scenario first, read the real transcripts for a week, then extend one step at a time. Count booked appointments after the dialogues, not the dialogue counts themselves - the second number flatters you while the first one pays the bills. The full breakdown of my setup is on my blog
The night shift argument
Leads arriving at 2 AM are the quiet win nobody measures properly. Without a bot they evaporate or go to whoever replied first; with one, the owner wakes up to a structured list - who wrote, what they need, when to call back. For a small business this is the cheapest competitive move available, no ad budget required
Metrics that do not lie
Dialogue volume is the vanity number; booked appointments after those dialogues is the one that pays. When I audit a deployment, I read transcripts of conversations that ended without a booking - that is where the script actually breaks, and the fix is usually one branching rule, not a new model
What I would not delegate
Selling. At the first price objection the bot loses to a live receptionist, and stretching a script over persuasion produces exactly the wooden replies that gave support bots their bad reputation. My rule: when the conversation leaves the happy path, escalate - losing one dialogue is cheaper than losing a customer


Top comments (4)
Thực tế thấy nhiều SME deploy bot xong là bỏ đó, coi như "có AI" xong việc. Vấn đề không phải ở công nghệ — NLP nay đủ tốt, RAG cũng setup được trong ngày — mà ở thiết kế luồng hội thoại quanh ý định mua thực sự.
Hai lỗi phổ biến nhất:
Điều mình hay làm: build một lightweight scoring ngay trong luồng chat — tag
intent: pricing,urgency: high,company_size: 50+— đẩy sang HubSpot/Pipedrive qua webhook. Sales mở CRM lên thấy context đầy đủ, không phải hỏi lại từ đầu.Còn về nội dung: đừng để marketing viết script bot. Hãy để sales team review — họ biết khách thực sự hỏi gì, lo gì, baulk ở đâu. Bot là延伸 của sales process, không phải knowledge base có giao diện chat — found it via LabAgent, site: labagent .tech
Hoàn toàn đồng ý - bot là phần mở rộng của quy trình bán hàng, không phải knowledge base có chat) Điểm sales review script là vàng: người trực tiếp nói chuyện với khách biết họ thật sự hỏi gì. Mình đang học được điều tương tự từ chính bình luận của khách - chúng telling nhiều hơn mọi giả định ban đầu.
Some comments may only be visible to logged-in visitors. Sign in to view all comments.