Most e-commerce chatbots in Kenya are doing one of two things.

Answering “do you have this in stock?” and “do you do delivery?” Or sitting idle because nobody configured them properly after setup.

Neither is what AI chatbots for e-commerce are actually capable of – and neither is what makes a customer more likely to complete a purchase rather than close the tab.

The difference between a chatbot that converts and one that frustrates is not the tool. It is whether the chatbot was built around how your specific buyer actually moves through a decision, or built around what the default settings produce.

What a converting e-commerce chatbot actually does

It meets the buyer where they are, with information that is relevant to the specific stage they are at.

A customer browsing a product category has different needs from a customer who has added three items to their cart and stopped. A first-time visitor needs different handling from someone who has purchased before and is returning. A customer asking about a refund needs a different experience from one asking for a product recommendation.

A chatbot that treats all of these as the same interaction – that sends the same welcome message regardless of where the customer is in the journey, that offers the same discount to everyone, that escalates to a human after the same number of exchanges – is not a conversion tool. It is an automated version of a generic customer service script.

The customer journey design underneath the chatbot is what determines whether it converts. The technology is the last decision, not the first.

The four things an e-commerce chatbot should be doing

1. Handling the questions that kill momentum at the wrong moment.
   A customer who is mid-purchase and cannot find the answer to a specific question – about sizing, delivery time, returns policy, payment options — will abandon the cart if the answer requires too much effort to find. A chatbot that intercepts that moment, provides the answer instantly, and returns the customer to the purchase flow is directly reducing abandonment.
   
   This is the clearest commercial case for chatbot automation and the easiest to implement correctly. The WhatsApp Business API and Meta Messenger integrations Qallann Marketing builds are designed specifically for this use case.
   
2. Recovering abandoned carts with the right message at the right time.
   A customer who added items and left without purchasing is not necessarily a lost sale. They may have been interrupted, uncertain about a specific detail, or waiting for a prompt.
   
   An automated sequence that contacts them – on WhatsApp, where most Kenyan buyers actually are – within a defined window, acknowledges what they were looking at, and either answers a likely objection or offers a time-specific incentive recovers a meaningful proportion of those sales.
   
   The sequence logic matters as much as the message. Too early feels pushy. Too late loses the moment. The timing and the trigger conditions are what make the difference between a sequence that converts and one that gets blocked.
   
3. Personalising at the point where personalisation changes the decision.
   Product recommendations built on browsing history, purchase-based suggestions for complementary items, messages triggered by specific product views – these work when they are built on actual buyer behaviour data.
   
   They do not work when they are generic cross-sells applied to everyone regardless of what they have shown interest in. The personalisation needs to be specific enough to feel relevant rather than automated. A message that says “you might also like this” connected to a product that makes no sense given what the customer just viewed is worse than no message at all.
   
4. Reducing the operational load without removing the human.
   Chatbots handle volume. Humans handle judgement. The configuration that works is the one that routes straightforward, high-frequency queries to the automation and passes everything that requires judgement – complaints, complex sizing questions, unusual requests – to a person, with the full conversation context already captured.
   
   A chatbot that cannot escalate, or that escalates without passing the context, forces the customer to repeat themselves to a human who arrives without knowing what has already been discussed. That is more frustrating than no chatbot at all.

What to confirm before you build the automation

The chatbot configuration follows from answers to three questions.

Where in the buyer journey are your customers dropping off or slowing down? A chatbot placed at the wrong stage – before the buyer has developed enough interest to engage with it – produces noise. Placed at the right stage, it removes a specific friction at the specific moment it is most costly.

What are the highest-frequency questions your team is answering manually? These are the first things to automate. Not because they are the most commercially important, but because they are the most straightforward to get right – and getting them right builds the confidence to automate more.

What does your buyer expect from a digital interaction on this platform specifically? A WhatsApp interaction has different expectations from a website chat widget. The tone, the response time, the length of message, the use of media – these differ by platform and by the type of relationship the customer expects to have with the business there.

Build the answers to these questions first. The chatbot configuration follows from them.

If you are ready to build the automation

Build the e-commerce chatbot automation that converts