Search in shoppers' words
Requests as people write them: Arabic, Arabizi, English, misspelled, mixed.
Details
"7aleeb" finds milk.
About 30 minutes, on your own catalogue.

Updated
An ecommerce AI agent is software that shops a store's catalogue for a customer: it searches, checks what it found, adds to the basket and prepares the checkout.
An ecommerce AI agent is software that shops a store's catalogue for a customer: it searches, checks what it found, adds to the basket and prepares the checkout. bisky is an ecommerce agent for supermarkets, grocery delivery and quick commerce.
It works inside the retailer's own app and website and on its WhatsApp Business number, typed or by voice, in Arabic and English, and it never places an order without the shopper's confirmation.
An ecommerce agent is the difference between being told where the milk is and having it put in your basket.
Three kinds of AI for ecommerce, by what they do.
| Chatbot | Site search | Ecommerce AI agent | |
|---|---|---|---|
| What it does | Answers questions | Ranks results for a typed query | Searches, checks, adds to the basket, builds the checkout |
| What it reads | Help-centre articles, scripted flows | The catalogue index | The live catalogue, the shopper's branch, the basket |
| Input it handles best | Common questions | Short keywords | Sentences: lists, budgets, recipes, follow-ups |
| Changes the basket | Rarely | No | Yes, and only orders after the shopper confirms |
An ecommerce agent is the difference between being told where the milk is and having it put in your basket. Three kinds of software get called "AI for ecommerce", and they do different jobs.
A chatbot answers questions. It is usually trained on help-centre pages and scripted flows: opening hours, delivery fees, return policy. It talks, it does not shop.
Site search ranks products for what a shopper types in the search bar. It is fast and essential, and it is built for short keywords. "Snacks for six people watching the game under 20 dollars" is not a keyword.
An ecommerce AI agent uses tools. It searches the live catalogue with the shopper's branch prices and stock, checks that what it found matches the request, adds products to the basket, changes quantities and builds a checkout. This is what people mean by agentic shopping: the agent does the legwork, and the shopper approves the result.
In grocery, start with the requests that are slow for shoppers and that a search bar handles badly.
Requests as people write them: Arabic, Arabizi, English, misspelled, mixed.
"7aleeb" finds milk.
"Milk, bread and eggs" returns a row per item, so the shopper picks each without asking three times.
A dish and a number of people becomes ingredient lines, the right packs and one tap to add them all.
"Under $20" and "cheapest first" are understood and held for follow-ups.
Product cards, a basket and a checkout the shopper confirms with a tap on your own number.
A shopper orders on a WhatsApp call or a voice call inside your app.
The basket is read back and the keypad confirms.
Grocery baskets are long, repeat often and depend on details a search bar ignores: pack sizes, the branch that will deliver, budgets, and what a family is cooking tonight. Those are the requests where an agent saves the most time, because each one would otherwise take several searches and taps.
Leave to people what needs judgement or authority: complaints, a late or wrong delivery, refunds, a large catering order, anything to do with an account or a payment dispute. The assistant can recognise these and hand the conversation to your team, who see it in the same inbox with everything said so far. A shopper can ask for a person at any time.
An agent that can change a basket needs harder limits than a chatbot that only talks.
An agent that can change a basket needs harder limits than a chatbot that only talks. bisky puts them in the server, not only in the instructions to the model, so no wording from a shopper can talk its way past them.
No tool the model can call places an order. The agent fills the basket, in sync with your app's cart, and builds a checkout. The order is placed only by the shopper's explicit confirmation: a tap on WhatsApp, a signed checkout page, or pressing 1 on the keypad after the basket is read back on a call. A typed or spoken "yes" alone never places an order. The order then goes to your platform through its connector, to your orders API, or to your team as an order ticket.
A retailer can hide alcohol, pork and tobacco, or all three for a halal store. Those products are removed by the server before the agent sees the catalogue, on every channel. Each product is judged by an AI model, not a word list, and a product that has not been checked yet stays hidden. Out-of-stock products cannot be added, and prices and stock come from the shopper's own branch.
Before launch, and before any change to the prompt, the model or search goes live, a test suite generated from your own catalogue runs: every department covered, retrieval of real products, basket changes, the language of each reply, safety, and WhatsApp flows. Generic benchmarks say little about whether an agent finds your 1 L fresh milk when a shopper writes "7aleeb".
Every search's top results are checked against the shopper's actual message before they are shown. A request for snacks for a basketball night shows snacks, not basketballs.
Most retailers need more than one. This is what each kind of tool is built for, stated generally.
Categories of tool compared by their core job. Many tools now add AI features beyond it, so check the current features of any you consider.
| Support agents | Shopping agents (such as bisky) | Search engines | |
|---|---|---|---|
| Main job | Resolve customer service questions and tickets | Turn a shopper's request into a basket | Rank products in the search bar and category pages |
| Typical questions | Where is my order? Can I return this? | Kabsa for 6. Cheapest milk. Snacks under $20. | A typed keyword: milk, rice, shampoo |
| Works from | Help centre, policies, order data, tickets | Live catalogue, branch prices and stock, basket | Catalogue index, behaviour data, merchandising rules |
| Acts on the basket | Usually not | Yes: adds, changes quantities, builds checkout | No, the shopper clicks |
| Where it lives | Help widget, email, social inboxes | Your app, your website, WhatsApp, typed or by voice | Your site and app search |
| Where it is stronger | Ticketing, agent workflows, support analytics | Multi-item requests, recipes, budgets, dialects, voice | Speed at scale, merchandising control, search analytics |
| Use it alongside the others? | Yes, for after the order | Yes, for shopping in conversation | Yes, for the search bar |
Support agents are strong at what happens after the order: tracking questions, returns, policies, routing tickets to the right person, and reporting on support quality. If most of your chat volume is "where is my order", start there.
Search engines power the search bar and category pages at large scale, with merchandising controls and analytics a team can tune. They are the right tool for the many short queries a big store gets.
A shopping agent sits in the conversation: it takes a sentence, a voice note or a call and turns it into a basket. bisky is built for that, for grocery in particular, and for shoppers who write and speak Saudi, Gulf or Lebanese Arabic, Arabizi and English.
Shoppers should not have to go somewhere new. The same agent, with one basket, works in the places they already use:
Shoppers should not have to go somewhere new. The same agent, with one basket, works in the places they already use:
Want the answer on your own products? Ask us and we will run it.
Write to hello@bisky.aiAn ecommerce AI agent is software that shops a store's catalogue for a customer. It reads a request in plain language, searches the live catalogue, checks the results, adds products to the basket and prepares the checkout. A chatbot only answers. An agent acts, within rules the store sets.
A chatbot answers questions, usually from help-centre content. An ecommerce agent uses tools: it searches the catalogue with real prices and stock, changes the basket and builds a checkout. bisky is an agent, and it still never places an order without the shopper's confirmation.
Not with bisky. It fills the basket and builds the checkout, but the order is placed only after the shopper confirms: a tap, a signed checkout page, or the keypad after the basket is read back on a call. A typed or spoken yes alone never places an order.
No. Your search bar and category pages keep their search engine. The agent answers requests people make in sentences, such as a budget, a recipe or a list, in your app and on WhatsApp, typed or by voice.
No. It is built for shopping, not for returns, refunds or delivery complaints. Every conversation is in a shared inbox where a person can take over.
Salla, Zid, Shopify, WooCommerce, product feeds, and our REST API with webhooks for retailers with their own platforms. Each branch can keep its own prices and stock.
Agentic shopping means an AI agent does the work of finding and choosing products and filling the basket, while the shopper approves the result. In grocery that means lists, recipes, budgets and repeat orders handled in a sentence.
Pricing is a proposal per retailer and depends on your stores and the channels you use. Book a demo and we will send one.
Reading helps. Seeing your own products in the answers helps more.