A client called on a Thursday afternoon to tell me their chatbot had insulted a customer.
Not maliciously. The chatbot had been asked about their refund policy, offered an answer that was technically correct but wrong in the specific way that mattered, and the customer read the gap as dismissal. Nobody was happy. It took five minutes to understand and thirty minutes to explain to a business owner who had been showing off their new AI widget to everyone they knew.
We'd deployed the chatbot six weeks earlier. I was managing it across four client sites — retail, professional services, two hospitality businesses — and the Thursday call was the moment I started keeping notes.
Three months later, here's what actually happened when we put AI chatbots on client websites.
The knowledge base is the product
The expectation going in, from almost every client: the chatbot would answer questions accurately, and the main task was making it look right. After three months, I'd reverse that priority. The visual integration takes an afternoon. The knowledge base takes weeks, and it never stops needing attention.
When a hospitality client changed their check-in policy in March, the chatbot kept stating the old policy for eleven days because nobody had updated it. Visitors asking about early check-in got information that was no longer true. Not because the AI failed. Because the humans maintaining it did.
The chatbot is only as accurate as what it knows. That sounds obvious in print. In practice, it means committing to a knowledge management habit most small businesses don't have and didn't expect to need when they signed up for a chatbot.
Visitors don't ask what you expect
Every site has a FAQ page. FAQs represent the questions businesses think visitors have. Chatbots surface what visitors actually ask, at the moment they're asking it, and the gap is usually wider than business owners expect.
The professional services client had prepared answers about their credentials, turnaround times, and what the work costs. The top three questions from their chatbot in the first month: "Are you taking on new clients?", "Can I email documents directly to you?", "Do you work on weekends?" None appeared on their FAQ. All three were different forms of the same question: will you actually be available when I need you?
That changed how they wrote their next proposal. The chatbot didn't generate the insight. The data did.
The volume ceiling cuts both ways
The case for AI chatbots is usually framed around volume: handle more conversations without adding headcount. That works for high-traffic sites. Most South African SMBs aren't operating at that level, and the argument at the low end looks different.
One retail client gets roughly 600 unique visitors a month. Their chatbot handles 20 to 30 conversations a week — a fraction of what most chatbot case studies would call meaningful volume. But those conversations happen at 9pm, on Sunday mornings, during times when nobody is available to answer. Conversion on those conversations runs higher than the site average because the person asking is motivated: they're there at 9pm on a Sunday for a reason.
The chatbot didn't replace their customer service. It covered a gap that exists because customer service has business hours and customers don't.
What this means if you're thinking about adding one
Three months, four sites, one genuinely upset customer, and several hundred conversations that wouldn't have happened otherwise. The main finding isn't about the AI.
The quality of the chatbot is determined before it launches, by how well the knowledge base reflects how the business actually works. The useful information comes from the data it collects, not from the conversations themselves. And for most SA small businesses, the case isn't about handling volume — it's about availability.
Budget more time for knowledge maintenance than for setup. The AI part works. The operational discipline around it is what makes the difference.
Frequently Asked Questions
How long does it take to set up an AI chatbot on a small business website?
The technical integration is usually an afternoon. The knowledge base — which determines how well the chatbot actually performs — takes weeks to build properly and needs ongoing maintenance every time the business changes its policies, products, or hours.
What questions do customers ask AI chatbots on small business websites?
Mostly availability questions: Are you taking new clients? Do you work on weekends? Can I reach someone directly? Businesses prepare FAQs around their credentials and services. Visitors want to know if you'll actually be there when they need you.
Is an AI chatbot worth it for a small business with low website traffic?
It depends on when your visitors engage. If people browse your site in the evenings or on weekends when you're not available, a chatbot can handle those conversations in real time. One client with 600 monthly visitors sees 20–30 chatbot conversations per week, mostly outside business hours — with above-average conversion because those visitors are motivated enough to seek out an answer at 9pm.
What's the biggest mistake businesses make when deploying a chatbot?
Treating the knowledge base as a one-time setup. When business details change — policies, hours, pricing, staff — the chatbot keeps answering with the old information until someone updates it. The AI doesn't know what it doesn't know.
Does Kern manage the chatbot knowledge base on an ongoing basis?
Yes. Ongoing knowledge base maintenance is part of the managed website service. We update it when business details change and audit it quarterly to catch outdated information.