AI chatbots are everywhere. Most of them are terrible. The ones that actually help customers share a few specific qualities that have nothing to do with the AI model and everything to do with how they are built.
This guide is for business owners who are considering adding an AI chatbot to their website or product. No code. No jargon. Just what you need to know to make a good decision.
The Three Types of Business Chatbots
Not all chatbots are created equal. Understanding the types helps you figure out what you actually need.
Type 1: FAQ Bots
These answer common questions based on a pre-defined knowledge base. "What are your hours?" "How do I reset my password?" "What is your return policy?" They are the simplest to build and the cheapest to run.
Best for: Businesses that get 20+ identical questions per day. Retail, hospitality, professional services.
Cost: $2,000-5,000 to build. $50-200/month to run.
Type 2: Workflow Bots
These do things, not just answer questions. They can book appointments, process orders, update account information, or route requests to the right department. They connect to your existing systems and take action on behalf of the customer.
Best for: Businesses where customer interactions follow predictable workflows. Healthcare scheduling, restaurant reservations, service bookings.
Cost: $5,000-15,000 to build. $100-500/month to run.
Type 3: Conversational AI
These handle open-ended conversations, understand context, and can assist with complex problems. Think of a knowledgeable sales associate who never sleeps. They can discuss your products, recommend solutions, handle objections, and guide customers through complex decisions.
Best for: Businesses with complex products or high-value sales. Technology companies, financial services, B2B sales.
Cost: $10,000-40,000 to build. $200-1,000/month to run.
What Makes a Chatbot Good
We have built chatbots that customers love and chatbots that customers tolerate. The difference comes down to five factors:
1. It knows when it does not know. The worst chatbots confidently make up answers. A good chatbot says "I don't have that information, but let me connect you with someone who does." This requires explicit guardrails, not hope.
2. It remembers context within a conversation. If a customer says "I bought the blue widget last week and it broke," the chatbot should not ask "which product are you referring to?" three messages later. Context retention is table stakes.
3. It escalates gracefully. When the chatbot hits its limits, the handoff to a human should be seamless. The human should see the entire conversation history and the customer should not have to repeat themselves.
4. It matches your brand voice. A law firm's chatbot should not sound like a tech startup's chatbot. The tone, vocabulary, and level of formality should match what customers expect from your business.
5. It gets smarter over time. Every conversation is training data. A well-built chatbot system includes a feedback loop where human agents flag incorrect responses, and those corrections improve the bot's future performance.
The Build vs Buy Decision
You have two paths: use an off-the-shelf chatbot platform or build a custom one.
Off-the-Shelf Platforms
- Intercom Fin: Best for SaaS companies with existing Intercom accounts. $0.99 per resolved conversation.
- Zendesk AI: Best for businesses already on Zendesk. Integrated with their ticketing system.
- Drift: Best for B2B lead qualification and sales conversations.
- Tidio: Best for small e-commerce stores. Affordable and easy to set up.
These platforms get you up and running in days. The tradeoff is customization — you are limited to what the platform supports.
Custom-Built
A custom chatbot makes sense when:
- Your workflows are unique enough that off-the-shelf tools cannot handle them
- You need deep integration with proprietary systems
- You want full control over the AI model, prompts, and behavior
- Data privacy requires keeping everything on your own infrastructure
Custom bots take longer to build (4-8 weeks) but do exactly what you need. They also give you ownership — no monthly platform fees eating into your margins as you scale.
The Implementation Timeline
Here is what a realistic chatbot project looks like, week by week:
Week 1: Discovery. Map your most common customer interactions. Identify the 20 questions that make up 80% of inquiries. Define what "success" looks like.
Week 2: Knowledge base. Gather and organize the information your chatbot needs. FAQs, product details, policies, processes. This is usually the hardest part — most businesses do not have their knowledge well-organized.
Weeks 3-4: Build and train. Configure the chatbot, set up the AI prompts, connect integrations, and test with real scenarios. Fine-tune the tone and response quality.
Week 5: Soft launch. Deploy to a subset of your traffic. Monitor every conversation. Fix issues fast. This is where most problems surface.
Weeks 6-8: Iterate and expand. Based on real conversation data, improve responses, add missing knowledge, and gradually increase the percentage of traffic the bot handles.
What to Watch Out For
- The "set it and forget it" trap: A chatbot that is not monitored and improved will slowly degrade. Budget for ongoing maintenance.
- Over-promising the AI: Do not tell customers they are talking to something that can handle everything. Set expectations. "I can help with most questions about our products and services" is better than "I can help with anything."
- Ignoring the training data: Your chatbot is only as good as the information you give it. If your FAQ page is outdated, your chatbot will give outdated answers.
- Skipping the human fallback: Every chatbot needs an escape hatch to a real person. The businesses that skip this to "save money" end up losing customers instead.
The ROI Question
Here is the math. If your support team handles 500 customer interactions per month, and the chatbot resolves 40% of them without human involvement:
- 200 interactions resolved automatically per month
- At an average handling time of 8 minutes per interaction: 26.7 hours saved
- At $25/hour blended cost: $667/month in labor savings
- Annual savings: roughly $8,000
A $5,000-10,000 chatbot pays for itself in 6-12 months. After that, it is pure margin. And the savings grow as your business grows — the bot scales without adding headcount.
The real value, though, is not the cost savings. It is the 24/7 availability. Customers get answers at 2 AM on a Saturday. They get answers during your busiest periods when your team is overwhelmed. They get answers in seconds instead of waiting in a queue.
That is what converts browsers into buyers.
