Every business software company is selling AI chatbots right now. The pitch is compelling: 24/7 customer support, instant responses, reduced staffing costs. The reality is more nuanced. Some chatbots transform customer service. Most become expensive annoyances that drive customers away.
When Chatbots Work
Chatbots work when the questions are predictable and the answers are clear. Product availability, shipping times, business hours, pricing tiers, return policies — these are perfect chatbot territory. The customer gets an instant answer. Your team handles fewer repetitive queries.
They also work well as a first filter. The chatbot handles the simple questions and routes complex ones to a human with context already captured. This means your support team spends time on problems that actually need human judgment.
When Chatbots Fail
Chatbots fail when they try to replace human interaction in situations that require empathy, negotiation, or complex problem-solving. A customer with a billing dispute does not want to argue with a bot. A prospect evaluating a R500,000 purchase needs a conversation, not a menu.
They also fail when they are trained on bad data. If your documentation is incomplete or outdated, the chatbot gives wrong answers confidently. That is worse than no chatbot at all.
What Good Implementation Looks Like
- Train on your actual documentation, FAQs, and product information
- Set clear boundaries for when the bot hands off to a human
- Monitor conversations weekly and retrain on questions the bot cannot answer
- Give customers an obvious way to reach a real person
- Track resolution rate, not just conversation count
The Cost Question
A basic chatbot using your existing documentation can be set up for R5,000-R15,000. A custom-trained bot with CRM integration costs R30,000-R80,000. The ROI calculation is simple: how many hours does your team spend answering questions the bot could handle? Multiply that by hourly cost. If the bot saves more than it costs within 6 months, it is worth it.
Our Recommendation
Start small. Deploy a chatbot that handles your top 20 most frequent questions. Measure the results for three months. If it works, expand. If it does not, you have only invested a fraction of what a full implementation would cost.