AI for Customer Service: The Good, the Bad and the "Kindly Hold"
Have you ever spent twenty minutes explaining your problem to a chatbot, only for it to reply, "Sorry, I did not understand. Kindly rephrase"? Then you rephrase. And it suggests an article about resetting your password, which has nothing to do with the delivery that never arrived. Welcome to the dark side of AI customer service in Kenya. The good news is that there is a bright side too, and the difference usually comes down to how the business sets it up.
The good: what AI customer service in Kenya does well
Used properly, AI can make customer service faster and kinder, especially for small teams who cannot afford to staff a phone line around the clock. It shines at:
Answering common questions instantly: opening hours, prices, delivery areas, payment options
Responding outside working hours, so the 11pm WhatsApp message gets a reply before morning
Collecting details before a human takes over, like order numbers and location
Routing enquiries to the right person or department
Drafting replies for staff to review, so they respond faster without sounding robotic
Summarising long complaint threads so a manager can catch up in seconds
For a business that gets the same twenty questions every day, this is a genuine relief. Your staff stop typing "We are open 8am to 6pm, Monday to Saturday" forty times a day and spend that energy on the customers who actually need help.
The bad: where it goes wrong
Things fall apart when businesses use AI to avoid customers rather than serve them. Common sins include:
No way to reach a human, or a way so hidden it might as well be in a locked drawer
Bots that loop the same three answers whatever you ask
Confident but wrong answers about refunds, prices or policies
Replies that ignore what the customer already said
Cold, formal language that makes an annoyed customer even angrier
There is also a real risk with AI that writes its own answers. If it has not been given accurate information about your business, it may invent a refund policy, promise a delivery date or quote a price that does not exist. Customers will screenshot that and hold you to it, probably in a WhatsApp group of four hundred people.
The "kindly hold" problem
Kenyan customers already have plenty of experience waiting. Banking halls, eCitizen queues, traffic on Mombasa Road. When they reach out to your business, they want the problem moving. A bot that says "kindly hold" and then disappears is worse than no bot at all. If something needs a human, the bot should say so clearly, give a realistic time, and pass the conversation on with all the details already captured, so the customer does not have to repeat their story from the beginning.
Rules for getting it right
If you are planning to add AI to your customer service, keep these principles in mind:
Start with your top questions and answer them well, rather than trying to cover everything
Feed the system your real policies and prices, and update them whenever they change
Always offer an easy, visible route to a human
Tell customers they are talking to an automated assistant
Match your tone: warm, local, clear, and comfortable with Swahili and English mixed together
Review conversations every week to spot failures and fix them
Do not let the bot make promises about money without human approval
Measure what matters
Do not judge success by how many conversations the bot handled. Judge it by whether customers got their problems solved and whether they would come back. Look at complaints, repeat contacts and reviews. A bot that closes many chats but leaves people frustrated is not saving you money; it is quietly costing you customers.
Fatuma, who runs a busy furniture shop in Eastleigh, put it best after testing a bot for three months: "The machine answers the easy questions. My staff answer the hard ones. And nobody ever has to type agent eleven times again."
Your next step: write down the ten questions your team answers most often. If a bot cannot answer those well, it is not ready for your customers. When you are ready to build one properly, our software and integration services can help.