AI for Ecommerce in Kenya
Built for chat orders and descriptive addresses, not for a postcode market.
What is ecommerce?
AI for ecommerce in Kenya addresses the practical differences of this market: most orders arrive through chat rather than a checkout, delivery addresses are descriptive rather than structured, and payment is frequently on delivery.
Ecommerce advice imported from elsewhere assumes a checkout, a structured address and a card payment. Kenyan ecommerce frequently has none of those.
A large share of orders arrive as WhatsApp messages. The delivery address is a description rather than a line one and a postcode, and payment is often on delivery, which makes a failed delivery a direct loss rather than a refund.
That changes what is worth building. Resolving a described address before dispatch is worth more here than a recommendation engine, because a failed delivery costs the goods, the trip and frequently the customer.
Why we start with failed deliveries
Duara works with online retailers and social commerce sellers from Nairobi, on order handling and delivery accuracy rather than on storefront design.
We start with failed deliveries rather than with conversion. Failed deliveries are the largest recoverable loss in most Kenyan online retail, and they are invisible in the platform analytics everybody looks at instead.
Who this is for
- Most orders arrive by WhatsApp or Instagram
- Failed deliveries are a regular cost
- Product information is inconsistent
- Order handling does not scale with volume
Industries we serve
- Online retailers
- Social commerce sellers
- Fashion and apparel
- Electronics retail
- Grocery and food delivery
- Marketplace sellers
Problems this solves
These are the problems of selling online in a market that does not work the way ecommerce platforms assume.
- Orders arriving as conversations. A WhatsApp thread contains the order, the address and the payment discussion, and somebody has to turn that into a dispatch.
- Addresses that are descriptions. Deliveries are directed by landmark and phone call, and an unclear address becomes a failed delivery.
- Failed deliveries costing twice. On a pay on delivery order a failed attempt costs the trip and the goods are back in stock with a customer who has cooled.
- Product data that varies. The same item is described differently across listings, which affects search, stock and returns.
What we deliver
Chat orders turned into structured records, addresses resolved and confirmed before dispatch, product information made consistent, and returns handled without manual reconstruction.
Chat orders converted into structured records
Address resolution and confirmation before dispatch
Consistent product data across listings
Automatic order status updates to customers
Returns handling without manual reconstruction
Reporting on failed deliveries and their causes
Features
- Built for chat first. Orders arriving as WhatsApp messages are the normal case here, and a system assuming a checkout misses most of the volume.
- Addresses confirmed before dispatch. A described address checked with the customer beforehand costs a message. A failed delivery costs the trip and often the sale.
- Failed deliveries treated as the priority. They are the largest recoverable loss in most Kenyan online retail and invisible in standard platform analytics.
- Product data made consistent. The same item described the same way across listings, which affects search, stock accuracy and returns.
- Status updates without asking. Customers who know where their order is stop messaging to ask, which is most of the support volume.
Benefits
- Fewer failed deliveries. Address confirmation before dispatch addresses the single largest avoidable cost in this market.
- Chat volume becomes manageable. Orders arriving as messages get turned into records without somebody reading every thread.
- Support volume drops. Proactive status updates remove the where is my order messages that dominate customer contact.
- Stock accuracy improves. Consistent product data means the same item is not counted as three different things.
Technologies we use
- Conversational order extraction. Turning chat threads into structured orders with items, quantities and delivery details.
- Address resolution. Interpreting descriptive addresses and confirming them with the customer before dispatch.
- Product data normalisation. Making descriptions and attributes consistent across listings and channels.
- Status notification. Automatic updates over the channel the customer ordered on.
- Returns processing. Handling returns against the original order without manual reconstruction.
Address resolution is the highest value and least fashionable application in Kenyan ecommerce. Everybody wants recommendations; almost nobody has measured what failed deliveries cost them.
How the process works, step by step
Ecommerce projects usually take five to nine weeks, starting with whichever costs more, failed deliveries or chat handling time.
-
Step 1
Free order review
We measure your failed delivery rate and how much time goes into handling chat orders.
-
Step 2
Address confirmation first
Where failed deliveries are the larger cost, confirmation goes first because it pays back immediately.
-
Step 3
Chat extraction added
Order details are pulled from conversations, with anything ambiguous confirmed rather than guessed.
-
Step 4
Status updates switched on
Proactive notifications reduce the support volume that dominates most sellers days.
-
Step 5
Product data cleaned
Consistency across listings follows, which improves search and stock accuracy together.
Quality assurance and how we work
The commitments attached to systems handling customer orders.
- Ambiguous orders are confirmed, not guessed. A guessed quantity or variant becomes a return, which costs more than the message it would have taken to ask.
- We start with failed deliveries. They are the largest recoverable loss in most Kenyan online retail and they do not appear in standard platform analytics.
- Nothing is dispatched on an unconfirmed address. Confirming costs a message and a failed delivery costs the trip, the goods handling and frequently the customer.
- We do not automate customer messages you have not approved. Wording that reaches customers is yours, since tone in chat commerce is part of the brand.
- We report what failed deliveries actually cost. Most sellers have never calculated it, and the figure usually reorders their priorities.
Frequently asked questions about ecommerce
Which problem should we fix first?
Usually failed deliveries. They are the largest recoverable loss in most Kenyan online retail and they are invisible in platform analytics.
Why not start with recommendations?
Because almost nobody has measured what failed deliveries cost them. Recommendations are more fashionable and worth considerably less here.
Can it handle WhatsApp orders?
Yes, and it should. A large share of orders arrive as chat, and a system assuming a checkout misses most of the volume.
How does it deal with our addresses?
By interpreting the description and confirming it with the customer before dispatch, since addresses here are directions rather than structured lines.
What if an order is ambiguous?
It asks. A guessed quantity or variant becomes a return, which costs more than the message it would have taken to confirm.
Will it message customers automatically?
Only with wording you have approved. Tone in chat commerce is part of your brand and we would not take that decision.
Does it reduce support messages?
Considerably. Proactive status updates remove most of the where is my order contact that dominates seller time.
Can it work with our online store?
Yes, alongside the chat channels, since most Kenyan sellers run both and the orders need to meet in one place.
What about pay on delivery orders?
They are treated as the higher risk case, because a failed delivery there is a direct loss rather than a refund.
How long before it is handling orders?
Five to nine weeks, starting with whichever costs you more, failed deliveries or chat handling time.
Can it clean up our product listings?
Yes, and consistency across listings improves search and stock accuracy together, which sellers often treat as separate problems.
Does it handle returns?
Yes, against the original order rather than reconstructing what was sent from memory.
Will it work for social commerce sellers?
Yes, and small sellers often gain most since order handling time is the constraint on how much they can sell.
Can it integrate with delivery services?
Yes, and passing a confirmed address to the courier is where most of the delivery improvement actually lands.
Does it handle Swahili and mixed language orders?
Yes, and chat orders here are frequently mixed, so a system handling only English loses part of the volume.
Can it detect fraudulent orders?
It flags patterns for review. Deciding an order is fraudulent stays with you, since a wrong refusal loses a real customer.
What about stock going out of sync?
Consistent product data is the precondition, and we would fix that before promising accurate stock across channels.
Do we need a big catalogue for this to help?
No. Order handling and address resolution help at any catalogue size, and small sellers feel the time saving sooner.
Who can see our customer data?
Only your own staff. Order and customer records stay in your systems and train no public AI model.
What does an ecommerce project cost?
Order volume and channel count determine it, quoted after the order review.
Glossary of terms
Ecommerce terms in plain language.
- Social commerce
- Selling through chat and social platforms rather than a website checkout.
- Pay on delivery
- Payment collected when goods arrive, which makes a failed delivery a direct loss.
- Address resolution
- Turning a described location into something a rider can find reliably.
- Failed delivery
- An attempted delivery that did not complete, costing the trip and often the sale.
- Product data
- The descriptions and attributes of items, which affect search, stock and returns.
- Order extraction
- Turning a chat conversation into a structured order record.
Related services in Industries
Talk to us about ecommerce
Work out what your failed deliveries cost last month, including the rider time. That figure usually decides where to start.
Start your project