5 Signs Your Business Is Ready for an AI Chatbot

You don't need a chatbot because you're busy. You need one because you're busy with the same ten questions, over and over.

Wanjiru runs a mid-sized clinic off Ngong Road. Every evening at 7pm, after the last patient leaves, she sits with her phone and starts replying. Sixty, sometimes ninety WhatsApp messages. Do you have Monday slots. What's the cost of a dental cleaning. Are you open on Sunday. Same questions, different names.

She used to think this was just the job. Answering people is the job. But somewhere around message forty, most nights, she starts typing the same sentence she typed at message twelve. And that's when it stopped feeling like service and started feeling like a leak.

Here's the reframe: a chatbot is not for businesses that get a lot of messages. It's for businesses that get a lot of the same message. Those are different problems, and only one of them has an easy fix.

What's actually happening right now

Before you decide whether you're ready, look honestly at how the day runs. For most Kenyan SMEs fielding customer questions on WhatsApp, it looks like this.

• A question comes in. Someone stops what they're doing to answer it.
• The answer usually already exists somewhere: a price list, a menu, a service sheet, a booking calendar.
• The customer waited anywhere from a few minutes to a few hours for a reply that took ten seconds to type once you found it.
• The busiest hours for questions are often the hours nobody is at a desk: evenings, weekends, late at night.
• Whoever answers the phone is doing it instead of something else. Restocking. Following up a lead. Closing a sale.

None of this is a technology problem. It's a repetition problem. And repetition is exactly what software is good at.

Here's the one-line reframe worth sitting with: you don't need a chatbot because you're busy. You need one because you're busy with the same ten questions, over and over, and every one of those answers is a small tax on your time that a machine can pay instead.

The core idea, with a small example

A good AI chatbot does not try to be your business. It does one narrow job, well, and it knows the edge of that job.

Take a bakery in Kilimani. Customers message on WhatsApp asking about cake prices, delivery areas, and whether a specific flavour is available this week. A chatbot for that bakery does not need to understand baking, staffing, or supplier issues. It needs three things: the current price list, the delivery zones, and this week's flavour sheet. Feed it those three documents, tell it its job is "answer pricing, delivery and availability questions, nothing else," and it can handle most of the evening traffic Wanjiru-style businesses lose sleep over.

When someone asks something outside that job, like a complaint about a late delivery, the bot should say so plainly and hand the conversation to a person. Not guess. Not improvise. Hand off.

That's the whole design principle. Narrow job. Clear documents behind it. A clean exit when it's out of its depth.

Why this matters now, not eventually

Read that number again. Your customers are not waiting to be convinced that talking to AI is normal. They already do it, more than almost anyone else on earth. The question in front of you is not "should my customers talk to a bot." They already expect fast, correct answers delivered through chat, because that is now the baseline experience they get from everything else in their digital life.

The gap most Kenyan SMEs face isn't adoption. It's that the business hasn't caught up to where the customer already is.

"But won't it give wrong answers?"

This is the objection that stops most owners at the demo stage, and it's a fair one. A chatbot that answers confidently and wrongly is worse than no chatbot at all. It can quote a price you no longer charge, promise a service you don't offer, or invent an answer to a question it should have refused.

The fix is not a smarter model. It's a smaller job. A chatbot scoped to answer only what's in your documents, and instructed to say "I don't know, let me get someone" when a question falls outside that scope, will almost never embarrass you. The failure mode isn't the AI being wrong. It's the AI being given a job too big for what it was actually taught.

Design against confident wrong answers from day one. Tell it explicitly what it does not know, not just what it knows.

The five signs you're ready

You don't need a chatbot because everyone else has one. You need one when these five things are true.

1. You answer the same questions all day. Price, hours, location, availability, delivery zones. If you could write the top ten questions from memory right now, that's your signal.
2. Your first reply is slow. Not because you're ignoring people, but because you're one person (or three) and messages arrive faster than you can type. Every minute a customer waits is a minute they might message a competitor instead.
3. Customers message after hours. In Kenya this is almost universal on WhatsApp: people message at 9pm, 11pm, on Sunday mornings, whenever the thought crosses their mind. If your business hours and your message hours don't match, you have an always-on customer and a nine-to-five reply habit.
4. Your answers already exist in documents. A price sheet. A service menu. A booking calendar. A frequently-asked-questions page you wrote once and forgot about. If the knowledge already lives somewhere written down, a chatbot can read it. If the knowledge lives only in your head, you're not ready yet, you need to write it down first.
5. You can name its one job. Not "handle customer service." That's not a job, that's a department. The job is "answer pricing and availability questions for the bakery" or "book appointment slots for the clinic" or "tell customers which branch has stock." One sentence. If you can't finish that sentence, stop and finish it before you build anything.

If you cannot name the job, you are not ready. That's not a failure, it's useful information. Go write the job description for the bot before you write a single line of code or sign a single contract.

The 90/10 rule for chatbots

Once the five signs are true, the build itself follows a simple rule: automate the obvious 90%, route the messy 10% to a human.

Most incoming questions are routine. Price, hours, location, "do you have this in stock." A well-scoped bot handles these cleanly, instantly, at any hour. The remaining slice, complaints, unusual requests, anything emotionally charged or genuinely ambiguous, needs a person. Trying to automate that 10% is where chatbots earn their bad reputation: forced, generic replies to a customer who is upset or has an unusual case.

The rule isn't "automate everything." It's "automate the boring, repetitive part, and make the handoff to a human smooth when it isn't."

WhatsApp first, not a website widget

This is where a lot of chatbot advice imported from other markets goes wrong for Kenyan SMEs. The default assumption elsewhere is a chat widget bolted onto a website. That's backwards for most businesses here.

Your customers are already on WhatsApp. That's where they message you today, at whatever hour they think of you. A website widget asks the customer to change their behaviour, open a browser, find your site, click the bubble. A WhatsApp bot meets them exactly where they already are, in the same thread they'd message a friend or a fundi.

If you only build one thing, build the WhatsApp version first. The website widget can come later, if it comes at all.

How to actually start, step by step

1. Write down your top ten repeated questions from the last month. Pull them straight from your WhatsApp history if you have to scroll back and count.
2. Gather the documents that already answer them: price list, service menu, hours, delivery zones, booking calendar.
3. Write the one-sentence job description for the bot. "Answer pricing and booking questions for [business]." Nothing broader.
4. Decide the handoff rule. What exact phrase or situation should make the bot say "let me connect you with someone" instead of guessing.
5. Launch on WhatsApp first, to a small slice of your traffic if you're nervous, and watch the first week of transcripts closely.
6. Fix what it gets wrong. Add missing documents. Tighten the job description. This is a living tool, not a one-time setup.

Notice what's not on that list: buying the most advanced model, or finding the flashiest chatbot platform. The work is in steps one through four. The tool is the easy part.

What you have after this

A week in, the obvious repeated questions stop landing on a person. Someone on your team gets their evenings back, or at least gets them back from message forty onward.

A month in, you can see the pattern in what actually comes through to a human: the genuinely unusual requests, the complaints, the things that needed a person all along. That list is worth reading closely, because it tells you what your business is actually dealing with once the noise is gone.

Six months in, the bot is not a novelty anymore, it's infrastructure, the same way your till or your booking calendar is infrastructure. Nobody talks about it because it's just working.

You don't need a chatbot because AI is trendy. You need one the day you can finish this sentence: "customers keep asking me the same thing, and I already know the answer." That day, the job is already written. You just haven't built it yet.