Quick answer
Use a chatbot for the repetitive questions that have one correct answer, such as prices, opening hours, location, delivery areas and order status, because it replies instantly at any hour. Keep people for complaints, refunds, negotiation, sensitive topics and anything that needs judgement. Most small businesses do best with both: the bot answers first and hands the chat to a person with the full history.
Key takeaways
- Chatbots win on speed, availability and consistency for routine questions with one right answer.
- People win on complaints, judgement calls, negotiation, empathy and anything sensitive.
- The weak point of every bot is the moment it does not know the answer, so plan the hand-over before you switch it on.
- A hybrid set-up, where the bot answers first and a person takes over with the full history, suits most small businesses.
- Start with the bot suggesting replies for your team, then let it reply on its own only when its answers are reliable.
Every business that answers customers on WhatsApp eventually hits the same wall: the same ten questions arrive all day, the team falls behind, and the customers who really need a person wait in the queue behind people asking for your opening hours. A chatbot looks like the obvious fix, but replacing people entirely usually backfires.
This comparison is for owners and managers deciding what to hand to a bot and what to keep human. It covers what each does well, where each fails, what they cost and how a sensible split looks for a Kenyan shop, clinic or hotel.
What is the difference between a chatbot and a human agent?
A chatbot is software that replies to customer messages automatically, either from rules you set (keywords, buttons, menus) or from an AI model that reads your business information. A human agent is a person on your team who reads the message, understands the situation and decides what to say.
The difference matters most at the edges. A bot is excellent at "What time do you close on Sunday?" and poor at "You delivered the wrong size, my daughter's party is tomorrow and I paid via M-Pesa an hour ago." The second message needs someone who can check the payment, apologise properly and decide whether to send a rider tonight. If you want the deeper background on how AI bots differ from menu bots, read AI chatbot vs rule-based chatbot.
Chatbot vs human: how do they compare side by side?
Neither option is better overall. They are good at different jobs, as the table shows.
| Chatbot | Human agent | |
|---|---|---|
| Reply speed | Seconds, every time | Minutes to hours, depending on the queue |
| Availability | 24 hours, weekends and public holidays | Working hours, unless you pay for shifts |
| Consistency | Same answer every time (right or wrong) | Varies by person, mood and experience |
| Handling volume | Many chats at once without slowing down | One or a few chats at a time |
| Complaints and emotion | Weak; can sound cold or miss the point | Strong; can apologise, reassure and fix |
| Judgement and exceptions | Follows rules or its knowledge base only | Can bend a policy when it makes sense |
| Selling and negotiation | Can share prices and options | Can read the customer, upsell and close |
| Cost pattern | Set-up time plus per-message charges | Salaries, training and supervision |
| Main risk | Wrong or unhelpful answers, customers stuck in a loop | Slow replies, missed chats, burnout |
What should a chatbot handle?
Give the bot questions that come up often and have a single correct answer you can write down. In a Kenyan business that usually means:
- Prices and availability: "Bei gani?", "Do you have this in size 40?"
- Location and hours: directions to your Westlands branch, Sunday hours, parking.
- Delivery: which estates and towns you deliver to, how long it takes, the fee.
- Payment details: your Paybill or Till number and what to put as the account.
- Booking requests: collecting the service, preferred day and name before a person confirms.
- Order status: where a structured system can answer it.
A Nairobi boutique, for example, might find that most evening messages are about sizes, prices and delivery to Kitengela or Ruiru. A bot that answers those at 10pm means the customer does not message the next shop instead. With chat.or.ke's automation flows you can answer these with buttons and menus, and the AI assistant can answer open-ended versions of the same questions from your own knowledge base, in English or Kiswahili.
What should always go to a person?
Hand the chat to a person whenever the customer is upset, the stakes are high or the answer depends on judgement. Typical examples:
- Complaints, refunds, wrong deliveries and damaged goods.
- Anything involving money already paid, such as an M-Pesa payment that did not reflect.
- Medical, legal or financial questions where a wrong answer could hurt someone.
- Large or custom orders where negotiation and trust decide the sale.
- Any customer who asks for a person, in any words ("agent", "mtu", "nataka kuongea na mtu").
Be strict about the last one. Nothing damages trust faster than a customer typing "I want to talk to someone" three times and getting another menu. Our guide on how to hand over from a chatbot to a human walks through the triggers and the hand-over message.
How much does each option cost?
The cost of a chatbot is mostly set-up time plus the messages it sends; the cost of people is mostly salaries. That makes bots cheap per conversation once built, and people expensive but irreplaceable for complex work.
On chat.or.ke there is no extra charge for the AI assistant or for automation. A reply the bot sends within the 24-hour window is charged like any other customer reply, currently KES 0.75 per message from your prepaid wallet, and messages customers send you are free. The pricing page lists every rate.
People cost more but bring things software cannot: they spot a frustrated regular, offer a goodwill discount, or notice that three customers in one day reported the same faulty batch. Do not measure a bot only by how many staff hours it saves. Measure whether customers get faster, correct answers and whether your team spends its time on the conversations that need it.
What does a hybrid set-up look like in practice?
A hybrid set-up lets the bot answer first and pulls in a person at the right moment, with the full conversation visible. Here is how it might work for a clinic in Kisumu:
- A patient messages at 7pm. The bot greets them with three buttons: "Book an appointment", "Opening hours & location", "Talk to reception".
- "Opening hours & location" gets an instant answer with the address near the hospital roundabout and the hours.
- "Book an appointment" asks for the patient's name, the service and a preferred day, and saves each answer.
- At 8am, the receptionist sees the request in the shared inbox, confirms the slot and pauses automation on that chat to continue personally.
- A patient who types a symptom question is handed straight to a person, because the clinic decided the bot must never give medical advice.
The same pattern works for a Mombasa hotel (bot handles rates and directions, staff confirm bookings and special requests) or a hardware shop (bot shares price lists, staff handle bulk quotes).
Which should you choose?
Choose a hybrid for most small and growing businesses, but be honest about your situation:
- Mostly people is the better fit if you receive a handful of chats a day, sell high-value or highly custom work (interior design, events, legal services), or your customers expect a personal relationship. A bot adds little and can feel impersonal.
- Mostly bot suits very high volumes of simple, repeated questions, such as a school during admissions week answering fee and reporting-date questions, provided a person is always one tap away.
- Hybrid suits most shops, clinics, hotels, SACCOs and service businesses: routine questions answered instantly, everything else handed over with context.
If you are unsure, start with the AI in suggestion mode, where it drafts replies and a person approves or edits each one. After a couple of weeks you will know which answers it gets right and can let it reply on its own for those. When the bot and the team share one WhatsApp inbox, the split is easy to adjust as you learn.
Published 20 October 2026 Updated 20 October 2026 Plain-text version
