What does a WhatsApp chatbot cost for a Kenyan business?

Most of your customers are already on WhatsApp, asking the same handful of questions: price, availability, location, and whether they can pay on delivery. A bot answers those instantly, at any hour, and pulls in a human only when one is genuinely needed. Here is what it costs to set one up in Kenya, including the running cost that rarely appears in the first quote.

The realistic ranges

Add one to two weeks at the start for WhatsApp Business API approval through Meta. That step is normal and it is not your developer stalling. Plan around it rather than being surprised by it.

The running cost, and the 24-hour window

A real bot runs on the WhatsApp Business API, not the free Business app, and Meta charges for messaging on top of whatever you paid to build it. The exact rates change, and they vary by conversation category and country, so treat any specific figure you read anywhere as something to check against Meta's current pricing before you budget.

What does not change is the shape of the rule, and it is the thing to design around. When a customer messages you first, a service window opens for 24 hours. Inside that window you can reply freely and cheaply. Outside it, you can only restart the conversation with a template Meta has pre-approved, and that costs more. This single rule decides whether a bot is affordable.

• Answer inside the window and your costs stay proportional to genuine demand.
• Blast templates at a list and your bill scales with your ambition rather than your sales.
• Design flows so the customer's own message opens the window. A "message us" button beats an outbound campaign on both cost and reception.
• Batch anything non-urgent into the next inbound conversation instead of paying to start a new one.

A worked budget

Suppose you run a salon that fields around 400 customer conversations a month, mostly about price and availability, and roughly 90 of those turn into a booking worth KES 2,000 on average.

The number to focus on is not the build cost, it is the staff hour. If answering those 400 conversations by hand takes two hours a day, the bot is buying back roughly 40 hours a month. Price those hours at whatever you pay for them and compare that against the monthly bill rather than against the build. A bot that saves less time than it costs to run is a bad bot, and no amount of clever conversation design fixes that.

Getting approved, and what holds it up

Before a single message goes out, the account has to exist properly. This is the step that surprises people, because it is administrative rather than technical and it does not move faster if you pay more.

• A Meta Business account, with your business verified. Have your registration documents to hand, and make sure the business name on them matches the name you are trying to use.
• A phone number that is not already on a personal or Business-app WhatsApp. This is the most common blocker we see, and freeing up an existing number means losing its chat history.
• A display name that follows Meta's rules. Names that read as generic categories tend to get rejected.
• An approved template for anything you want to send outside the 24-hour window.

Start this on day one of the project rather than the week before launch. The build can happen in parallel, and approval is the item most likely to sit waiting on somebody else's queue.

What to automate first

You do not have to automate everything, and the businesses that try usually end up with a bot nobody trusts. Take your last two weeks of WhatsApp messages and sort them into four groups: how often each question arrives, and how hard it is to answer correctly.

Almost all the value sits in the top-left box, and it is usually four to six questions. Build those, get them right, and add more once staff trust the thing. A bot that answers six questions perfectly earns its keep. One that attempts forty and fumbles a third of them gets switched off, and then the build was wasted.

What a good bot does

• Answers the predictable questions instantly: hours, prices, stock, directions, at any hour.
• Takes orders and triggers an M-Pesa STK push so the customer pays inside the chat.
• Books appointments and sends reminders, which is usually where it earns its cost back fastest.
• Hands off cleanly to a human, with the whole conversation attached, for anything it cannot handle.

That last point is the one to insist on. The rigid "Press 1 for sales" bots get abandoned because the customer can tell there is no way through to a person. The value is in handling natural questions well and knowing exactly when to stop trying.

When not to build one

If you get five enquiries a week, a bot is an expensive way to answer them. If your questions are genuinely varied, and every customer needs a real conversation, a bot will frustrate people and you will turn it off within a month. The businesses that get the most from this have high volume and low variety, where the same six questions arrive a hundred times. Look at your last week of messages and count how many were the same question before you spend anything.