How does an AI social inbox agent know what to say?
It answers from a knowledge base you build from your own documents. Not from the internet. Not from generic training data. From a Pinecone vector database populated with your actual business information: your services, prices, hours, location, policies, team bios, FAQs, and anything else a customer might ask about. When a customer asks "what are your prices?" the agent queries your knowledge base and answers with your real prices. If the question falls outside what you've told it, it doesn't guess. It tells the customer the team will follow up and flags the conversation for review.
The distinction matters because most businesses that have tried chatbots have been burned by generic ones. They ask a basic question and get a generic, unhelpful answer that damages trust rather than building it. A knowledge-base-powered agent is fundamentally different. It's only as good as the information you give it, which means it's entirely within your control, and it's specifically about your business rather than any business.
Setting up the knowledge base is a one-time task. You document your business in a Google Drive document - services, pricing, team, policies, hours, location, FAQs, any recurring customer questions and their correct answers. That document syncs to Pinecone and becomes the agent's entire world. When you update the document, the knowledge base updates. The agent always answers from your latest information.
What's the difference between an AI sales agent and a customer support agent?
A sales agent is built to drive transactions. It helps customers browse products, answer questions about specific items, suggest alternatives, handle sizing, and guide them toward a purchase decision. A customer support agent is built to handle any inbound question: hours, pricing, services, booking, policies, location, team. Sales agents are the right fit for retail, e-commerce, and product-forward businesses. Customer support agents are the right fit for any business that receives high volumes of repetitive questions across social channels, regardless of industry.
The Kaizora Social Inbox Agent is distinct from the Social Media Sales Agent for this reason. SMSA was designed for Instagram-based retail - browsing, ordering, stock questions. The Social Inbox Agent is designed for service businesses, clinics, restaurants, wellness centres, professional practices, and any organisation where customers reach out to ask questions before deciding to visit or book. Different goals, different conversation design, different outcomes.
A useful mental model: if a customer message starts with "I want to buy," that's a sales agent conversation. If it starts with "What are your hours?" or "Do you do X?" or "How do I book?" that's a support agent conversation. Most service businesses receive far more of the second type than the first. And most service businesses are currently handling those conversations manually, one by one, every day.
Which social platforms does the AI social inbox agent support?
WhatsApp Business, Instagram DMs, and Facebook Messenger - simultaneously, through one unified agent. A customer reaching out on WhatsApp gets an instant response there. A customer who found you through Instagram and DMs you gets a response there. A customer messaging through Facebook gets a response through Facebook. The same knowledge base powers all three channels. Your team monitors one conversation log rather than managing three separate inboxes with three separate workflows.
For most businesses, WhatsApp is the dominant customer channel across many global markets LATAM, MENA, India, Southeast Asia, and increasingly the EU and UK. But Instagram is increasingly used for customer questions, particularly from customers who discovered the business through a post or a tagged story. Facebook Messenger carries lower volume but is still the default channel for some customer demographics. Having all three connected means no message falls through a crack because it came through the "wrong" channel.
The unified inbox also means your team's time is spent on conversations that actually need a human - escalations, complaints that require judgement, requests that fall outside the knowledge base. Everything else is handled automatically, instantly, at any hour. Response time drops from hours to seconds. Message volume your team has to personally handle drops by 70-80% in typical implementations.
Can the AI hand off a booking mid-conversation?
Yes, seamlessly. When a customer's question naturally leads toward a booking - "can I come in for a consultation?", "how do I make an appointment?", "do you have availability this week?" - the agent transitions into a booking flow within the same conversation. It collects name, service, date, and time preference, checks your live availability, and confirms the appointment. The customer never leaves the chat. Support and booking are one continuous experience, not two separate systems with a hand-off moment that causes friction.
This integration is one of the most compelling aspects of a well-built inbox agent. The customer journey often looks like this: they ask a question about a service, the agent answers it well, they decide they want to book, the agent handles the booking. Zero friction. Zero clicks to a different URL. No switching context. The natural progression from curious to booked happens in one conversation thread.
For businesses that currently send a separate "click here to book" link mid-conversation, the drop-off at that link is significant. Most customers who get a booking link in a chat don't follow through. Keeping the entire interaction in the same thread is a material conversion improvement.
The economics: AI versus a full-time customer service hire
A full-time customer service rep in the US costs $30,000 to $70,000 a year; in the UK, £22,000 to £40,000 plus benefits, training, leave cover, and turnover risk. They work 8 hours a day, 5 days a week, in one language (or maybe two). They're not available at 11pm when a customer is planning a visit for tomorrow. They call in sick. They leave.
An AI social inbox agent is available 24 hours a day, 7 days a week, handles unlimited concurrent conversations, costs a fixed monthly amount substantially lower than a human hire, never takes a day off, and doesn't need retraining when your service menu changes - you update the document. In English, and with a non-English-language flag for messages the agent shouldn't attempt automatically.
The right framing isn't "replace your team." It's "stop having your team answer the same 12 questions every day." A good social inbox agent handles the repetitive 80% of inbound messages automatically, freeing your team for the 20% that actually requires human judgement. The team becomes more valuable, not redundant. And the total cost of customer communication drops while response quality and speed improve.
What happens when the agent can't answer a question?
When a customer asks something outside the knowledge base - an unusual request, a complaint that needs a specific resolution, a question about something you haven't documented - the agent doesn't fabricate an answer. It tells the customer that it's flagging the question for the team and that someone will follow up. The message is logged with a specific flag your team can filter for. The customer gets an acknowledgement. Your team gets a notification. No message goes unanswered, and no message gets a wrong answer.
This is the critical difference between a knowledge-base-powered agent and a generic chatbot. Generic chatbots are trained to always produce an answer, which means they produce wrong answers when they're uncertain. A properly constrained agent that knows to say "I'm flagging this for the team" when it's uncertain is genuinely more useful, and more trustworthy, than one that confidently makes things up.
Over time, the knowledge base improves. Every question that gets flagged is a data point about a gap in your documentation. You add the answer. The agent now handles that question. The flagging rate drops. The system gets better the longer you run it.