AI chatbots went from novelty to expected feature in about 18 months. Now every business thinks they need one, and every software vendor claims their chatbot is powered by advanced AI.

Most business chatbots are still terrible. They frustrate customers, fail to answer basic questions, and create more work than they save. But some actually work, and the technology is improving rapidly.

What Business Chatbots Actually Do

The basic function is handling repetitive customer questions without human involvement. “What are your hours?” “Where’s my order?” “How do I reset my password?”

Good chatbots handle these questions accurately and route complex issues to humans smoothly. Bad chatbots frustrate customers with irrelevant responses and make it hard to reach a real person.

The AI part - natural language understanding, context awareness, learning from interactions - varies enormously between platforms. Some “AI chatbots” are just decision trees with fancier marketing.

Platform Categories

Rule-based chatbots follow programmed conversation flows. You define the questions and answers. No actual AI, just conditional logic. These work fine for narrow use cases where questions are predictable.

AI-powered chatbots use natural language processing to understand questions they weren’t explicitly programmed for. Quality varies wildly based on the underlying AI model and how well it’s trained on your specific domain.

Hybrid approaches combine rule-based logic for common questions with AI for handling variations and unexpected queries. This is usually the most reliable approach for business use.

Major Platforms Worth Considering

Intercom is comprehensive customer messaging with AI chatbot capabilities. The chatbot - called Fin - is trained on your help documentation and can answer questions with citations to your knowledge base.

Pricing starts at $39/month for basic messaging, but the AI features require higher tiers at $99/month and up. It’s expensive but works well if you need the full customer communication platform.

Drift focuses on sales and marketing chatbots. It’s designed to qualify leads and route them to sales rather than customer support.

The conversational marketing approach works if your goal is lead generation. For customer service, there are better options. Pricing starts around $2,500/month, which prices out most small businesses.

Zendesk Answer Bot integrates with Zendesk’s customer service platform. It suggests help articles based on customer questions and can deflect tickets before they reach your support team.

This makes sense if you’re already using Zendesk. Pricing is $50/agent/month on top of base Zendesk costs. The AI is decent but not exceptional.

Ada is a standalone AI chatbot platform focused on customer service automation. It’s more sophisticated than most competitors and includes analytics on where your chatbot is failing.

Pricing isn’t public but starts around $1,000/month based on conversation volume. It’s positioned for mid-market and enterprise rather than small business.

Tidio offers affordable chatbots for small businesses, starting at $29/month. The AI is less sophisticated than enterprise options, but it’s accessible for companies that can’t spend thousands monthly.

The visual bot builder is easier to use than coding-based platforms. You’ll hit limitations with complex use cases, but for basic customer service automation it works.

Custom-Built Options

If you have development resources, you can build chatbots using platforms like:

• OpenAI’s GPT API for the natural language understanding
• Microsoft Bot Framework for the bot infrastructure
• Dialogflow (Google) for conversation management
• Rasa (open source) for complete control

Custom development makes sense if you have unique requirements, complex integrations, or want to avoid per-conversation pricing from SaaS vendors.

Working with an AI consultancy can help evaluate whether custom development or commercial platforms make more sense for your specific situation. The build vs buy decision depends on your technical capabilities and long-term plans.

What Actually Matters

Accuracy - A chatbot that gives wrong answers is worse than no chatbot. Test extensively with real customer questions before deploying.

Escalation - How easily can customers reach a human when the bot can’t help? If this is hard or hidden, customers will hate your chatbot regardless of how good the AI is.

Training - How much work is required to set up and maintain the bot? Some platforms need extensive training data. Others work reasonably well with just your help documentation.

Analytics - You need to see where the bot fails so you can improve it. Good platforms show you unanswered questions, low-confidence responses, and conversation abandonment rates.

Integration - Can it access your order system, CRM, knowledge base, and other tools to provide useful answers? A chatbot that can only share generic information has limited value.

Common Implementation Mistakes

Deploying a chatbot without proper training and expecting it to learn on the job. AI improves with data, but starting with a poorly trained bot alienates customers.

Making it difficult to bypass the bot and reach a human. Customers should always have an easy exit to human support.

Not monitoring chatbot performance after deployment. Set up alerts for high failure rates and review conversations regularly to identify improvement opportunities.

Using chatbots for inappropriate situations. Some questions require human judgment, empathy, or complex problem-solving. Don’t force everything through the bot.

Chatbot Pricing Models

Most platforms charge based on conversation volume, user seats, or monthly subscription with usage limits.

Conversation-based pricing can scale unexpectedly if your traffic grows. Per-seat pricing is predictable but expensive for large teams. Flat monthly fees with usage caps are easier to budget.

Watch for hidden costs - setup fees, training assistance, premium integrations, and analytics features often cost extra.

When Chatbots Make Sense

You have high volumes of repetitive questions that are clearly documented in your help content. The chatbot can surface those answers faster than customers searching themselves.

You need 24/7 availability but can’t staff support around the clock. The chatbot handles overnight and weekend questions.

You want to qualify leads before they reach sales. The chatbot can gather basic information and route qualified prospects appropriately.

When to Skip Chatbots

Your customer questions are highly variable and require human judgment. AI isn’t magic - complex, nuanced questions need human support.

You have low support volume. If you only get a few questions per day, the chatbot implementation effort exceeds the time savings.

Your team is already responsive and customers are satisfied with current support. Don’t add technology just because competitors have it.

Testing Before Committing

Most platforms offer free trials. Test with real customer questions, not the vendor’s demo scenarios.

Involve your support team in evaluation. They’ll use the tool daily and know what actually matters.

Start with a limited deployment - one website section, one product line, one customer segment. Prove value before rolling out broadly.

The Reality Check

Chatbots can reduce support costs and improve response times when implemented well. They can also frustrate customers and damage your brand when implemented poorly.

The technology is genuinely improving. GPT-4 based chatbots are substantially better than earlier AI models. But the platform matters less than the implementation quality.

Focus on solving specific problems rather than implementing chatbots because everyone else has them. Know what questions you want automated, measure success clearly, and be willing to intervene when the bot fails.

A mediocre chatbot with good escalation to humans is better than a sophisticated chatbot that traps customers in conversation loops. Keep that in mind when choosing platforms.