Demo scenario

Shoe store

Shoe Store: a sneaker photo in the chat and the order is on its way

Customers send the store photos of shoes from Instagram or the street. The bot recognises the model, picks the size from foot length, clears doubts and places the order.

Niche: Shoe store, Kyiv
Channel: WhatsApp Business
Launch: 3 weeks
Volume: ~1,500 chats/month

I want a bot like this How it works

More about WhatsApp Instagram stores

12:09

Shoe Store

This business uses a secure service from Meta to manage this chat. Tap to learn more.

Today

Hi12:03

do you have these in a 9?12:03

Hi Andriy! Yes, that's our Court White, size 9 is in stock 👟12:03

Quick tip: they run a bit small. What's your foot length?12:04

around 27 cm12:04

Then a 9.5 will fit you perfectly. We have it too12:04

and if it still doesn't fit?12:05

No problem: exchange or return within 14 days, and we cover the exchange shipping12:05

ok, I'll take the 9.512:05

Order #4821
Court White, size 9.5
Cash on delivery
Total: $6212:06

Shoe StoreOrder #4821Confirm12:06

Thanks! Please send your city and post office branch12:06

Kyiv, branch 2412:07

Got it! A manager will confirm the order and it ships today before 6 PM. I'll send tracking here 📦12:07

DonePlaced the order and passed it on for shipping

Results

4 minfrom photo to an order with delivery details

1photo for the bot to find the model

9.5size instead of 9: the model runs small

14 daysto exchange: terms given before the order

* Demo case: names, quotes and numbers are placeholders and will be replaced with real ones.

Challenge

Managers searched for the model by photo by hand, asked for the size again and often got it wrong with small-fitting models. Returns ate into the margin.

Finding a model by photo took up to 10 minutes
Sizes were advised by eye
Orders were copied into the CRM manually

Solution

The AI assistant recognises the model from the photo and catalogue, advises a size that accounts for the fit, explains exchange terms and creates the order in KeyCRM. Once a manager confirms it, the bot creates the Nova Poshta waybill and sends the number in WhatsApp.

Model recognition from the customer's photo
Size from foot length adjusted for the model
The order without a manager, the waybill once they confirm

Architecture

How the bot is built

The customer writes in a messenger, the AI assistant understands the request and works with the business's services itself. No copying between tabs.

TriggerWhatsApp

Has a photo?

yes

Recognise model

Advice & order

AI assistantAdvice & order

ModelPrompt: model fitCatalogue & sizesMemory

Order

Waybill once confirmed

Waybill to customer

Trigger: where the customer comes fromAI assistantThe agent's model, prompt, memory and knowledge← swipe the diagram →

Behind the scenes

What happens while the customer types

Checks the size chart

The bot knows Court White fits small and suggests a 9.5 instead of the usual 9 for a 27 cm foot.

CatalogueCourt White · sizes

	Size	Foot, cm	Stock
2	8	26.0	3
3	9	26.5	5
4	10	28.0	2
5	9.5 ✓	27.0	4

Size matched by the AI assistant · just now

Creates the order

The order appears in KeyCRM with the photo, size, payment method and WhatsApp as the source. A manager checks it and confirms with one tap.

KeyCRMOrder #4821

NewConfirmedShippedReceived

ItemCourt White, 9.5

Total$62

PaymentCash on delivery

SourceWhatsApp

Created by the AI assistant · just now

Ships with Nova Poshta

Once confirmed, the bot creates the waybill and sends the number in WhatsApp. Within 24 hours of the customer's last message this is a regular non-template reply: the first 1,000 a month are free, then Meta's rates apply.

Nova Poshta API

Waybill number20 4500 9912 3417

Created

Shipped

At branch

RecipientAndriy Bondar

CityKyiv, branch 24

PaymentCash on delivery

Stack

Tools and integrations

The bot's brain

AI assistant

Recognises products in photos
Knows how models fit
Places the order

Agent parts

ModelPrompt: model fitCatalogue & sizesMemory

ChannelWhatsApp Cloud API

Dialogue, buttons, photos

Cloud API

CatalogueCatalogue

Photos and size charts

Sync

CRMKeyCRM

Orders

REST API

DeliveryNova Poshta API

Waybills and statuses

API 2.0

Honestly

Pros and limits

What the business got

The model is found in seconds
Size from foot length and the model's fit, not by eye
Customers get the waybill without asking
Managers handle wholesale and tricky questions

Limits worth knowing

Since 1 October 2026 only the first 1,000 replies a month per number are free, and WhatsApp templates (reminders, statuses, broadcasts) are paid at Meta's rates
Recognition is weaker on blurry photos, then the bot asks again
The catalogue and stock must be kept up to date

Rollout