A Bedones agent recognises your products because it reads your catalogue before its first conversation: every name, every price, every variant, every photo. When a customer sends a screenshot with no reference and no product name, the agent matches the image against that catalogue, identifies the item and answers with its price, availability and options.
Why a standard chatbot does not recognise your products
A standard chatbot does not recognise your products because it has never read them. Its engine runs on keywords and pre-written scenarios: if a message contains the word “price”, it fires the “price” reply. The catalogue itself sits outside the system, in a shop or a file that the bot never opens.
The consequence shows up in the very first exchange. A customer writes “I want the red dress” and gets back “Thanks for your message, an agent will reply shortly.” The sentence is polite and moves nothing forward: no price, no available size, no way to order. Meanwhile that customer has already opened a competitor’s chat.
A chatbot that ignores your catalogue therefore pushes all the work back onto you. Every question about price, size or availability lands on your phone, usually in the evening, often twice over. That is exactly where the difference between a chatbot and an AI agent is decided, and it shows up in lost sales rather than in a feature list.
What your Bedones agent reads in your catalogue
Your Bedones agent reads your catalogue item by item, not keyword by keyword. For each reference it records the name, the description, the price, the declared variants and the availability status. None of that is copied into canned replies: it becomes the raw material the agent draws on at the moment a real question arrives.
The agent also treats your catalogue images as content rather than decoration. It knows what each product looks like: its shape, its colour, the details visible in the photo. That visual reading is what makes it possible to identify an item from a screenshot, at the point where the customer’s message contains no usable text at all.
This reading happens in three layers, and the agent combines them in every reply instead of favouring one. A customer can ask through text, through an image, or through a reference to a running promotion, and still get the same answer about the same item:
- the item record: name, description, price, variants, availability
- the attached visuals, which make a product findable without its name
- the commercial context: running promotions, similar items, complementary products
The screenshot test
The screenshot is the test that separates an agent from an auto-reply. A customer browses your page, spots an item they like, photographs their screen and sends the image as a message. No name, no reference, no link: just a screenshot, often cropped, sometimes crooked, with “how much?” underneath it.
Faced with that image, a standard chatbot has nothing to compare it against: it holds no reference visuals and falls back on its generic reply, the one that hands the customer over to a human. The Bedones agent holds every visual in your catalogue, and it runs four operations without any input from you:
- it analyses the image received
- it matches it against the visuals in your catalogue
- it identifies the exact item
- it answers with the price, the availability and the options
For the customer, the exchange feels like talking to someone who knows the shop by heart. For you, nothing happened at all: the screenshot came in, the answer went out. The same mechanism applies when the image arrives in a Facebook comment or an Instagram direct message rather than on WhatsApp.
Salesperson answers, not robot answers
Because the agent genuinely knows your products, its answers are about the item rather than about the procedure. A question about sizing gets the sizes still available. A question about fabric gets the fabric. A customer torn between two models gets a comparison of the two, with their respective prices and what actually separates them.
Out-of-stock situations change character too. Instead of a “no longer available” that closes the conversation, the agent flags the shortage and offers the items closest to the one that is missing. Promotions follow the same logic: the price quoted is the price of the day, not one frozen in a script written three weeks earlier.
Those answers stay yours. When a wording does not suit you, you correct the agent in plain language, the way you would correct a salesperson: that is the principle described in giving feedback to your agent. The correction applies to the conversations that follow, with no reconfiguration and no rule to rewrite.
What your catalogue needs for recognition to work
Product recognition depends directly on the quality of what your catalogue contains. An agent can only identify what it has seen: an item with no photo will not be recognised in a screenshot, and a variant that appears nowhere will never be offered to a customer. Filling in the catalogue is the real setup work.
Three things make the difference in practice. Clear photos, taken from the angle your customers actually see the product from. Variants listed explicitly, rather than buried in the middle of a description. And availability kept current, because false availability produces a false answer: the agent is precise, it is not clairvoyant.
Once that catalogue is in place, it serves every platform at the same time. The same item feeds the reply to a Facebook comment, to an Instagram direct message and to a WhatsApp conversation, as shown by the agent that connects comments, messages and catalogue. You fix a record once, and every conversation takes the change into account.