A classic chatbot answers from scripts written in advance: it recognises keywords and follows a fixed decision tree. An AI agent acts. It reads your catalogue live, understands a freely worded question, composes a routine or an outfit from your real products, prepares a cart, and hands over to your team on the cases that need it. The first informs. The second helps someone decide, and buy. For an e-commerce site the difference shows up most on unexpected questions: a chatbot jams or loops the moment it leaves its script, an AI agent carries the conversation on because it reasons over your catalogue instead of following a fixed path. Choosing between them depends on your volume of questions and their complexity, not on fashion.
The classic chatbot: how it works, and where it jams
A classic chatbot runs on scripts: a decision tree where every expected question leads to a pre-written answer. It recognises keywords or buttons, not the actual meaning of what the customer typed.
As long as the customer stays on the expected path (« what are your opening hours? », « where is my order? »), it works well. The moment they step off the script — a rephrasing, an ambiguous question, a request for advice (« what works for dry, sensitive skin? ») — the chatbot loops, or pushes them to a contact form. That rigidity explains a harsh finding: in France, general-purpose chatbots rank as the least-liked channel of all, according to an Ipsos BVA study of 5,000 people in 2025. For the detail of that mechanism and its limits, see our full guide to e-commerce chatbots.
The problem is not the technology itself, it is what the technology can do: a script does not check your live stock, does not compose a selection of products, and does not understand edge cases. It closes exchanges; it doesn't open them.
The AI agent: what actually changes
An AI agent is a different kind of thing because it acts rather than recites. It reads the catalogue live, keeps the whole thread of the conversation in mind, and builds an answer for the customer's precise situation rather than for a generic scenario.
In practice that means three things. First, product knowledge comes from real data: the Ryma advisor, for example, is fed the official documentation of the major brands, which lets her answer on an ingredient, a range or a contraindication precisely, not with a rough summary. Second, she composes: a morning routine, a coordinated outfit, a basket of several references that work together — something a script cannot do. Third, she knows when to stop: on sensitive cases (pregnancy, allergy, a complaint), a fixed rule applies and the wheel goes to the retailer's team, rather than risking an unsuitable answer.
That ability to act rather than to answer is precisely what separates a conversational agent from a classic support tool.
Chatbot or AI agent: the comparison
The table below compares the two approaches on the five points that genuinely matter to an e-commerce site.
| Criterion | Classic chatbot | AI agent (advisor) |
|---|---|---|
| Product knowledge | Frozen entries, updated by hand | Catalogue read live |
| Unexpected question | Loops, or pushes to a form | Understands and builds a fitting answer |
| Preparing the cart | No: points to a product page | Composes a coherent selection or routine |
| Sensitive cases | No detection, generic answer | Detects them and hands over to the retailer's team |
| Traceability | Rarely centralised | Every conversation on the record in a dashboard |
That comparison echoes a qualitative fact about the sector: major brands such as L'Oréal, La Roche-Posay and Vichy have each built their own AI skin-assessment tool into their sites — proof that automated advice already goes well beyond a simple script for the players who can afford it.
What it looks like for real: the Beautymarket journey
On the ground, the difference between a chatbot and an AI agent is measured in conversion, not only in comfort. Beautymarket, the number one beauty and skincare store in Morocco (1,000,000 visitors a month, measured by SimilarWeb), put the Ryma advisor into production on its dermo-cosmetic catalogue (L'Oréal, La Roche-Posay, Vichy, Avène, Bioderma, CeraVe).
Before going into production, 100,000 test conversations were used to check the reliability of the answers. In testing, 70% of requests were handled end to end, with no human involvement. That is what an AI agent can bring when it has a real catalogue to interrogate and a real volume of questions to handle. The full case is on the Beautymarket case, and our beauty and skincare guide sets out how that model applies to a dermo-cosmetic catalogue.
How to choose for your store
The right choice depends on your volume of questions and their complexity, not on how new the technology is. If your store gets few questions and they are nearly always about the same things (hours, delivery, returns), a classic chatbot is enough: it costs less to set up and covers the essentials.
But as soon as your customers ask advice questions — « what suits my skin », « what goes with this piece », « what's the difference between these two products » — a chatbot hits its ceiling fast, and that is exactly where an AI agent changes the game: it turns a question into a potential sale instead of leaving it without a useful answer. That is particularly true in beauty and skincare, where repurchase depends first on whether the product kept its promise, far ahead of price, according to McKinsey's State of Beauty 2025: advising well on the first purchase weighs directly on loyalty.
A simple rule of thumb: the broader and more technical your catalogue, the wider the gap opens in your favour with an AI agent. If you are on Shopify, setup can start straight from your existing catalogue — see our guide on installing an AI on Shopify.
Frequently asked questions
›Is an AI agent more expensive than a classic chatbot?
It depends on the volume and complexity of what you need, not on a universal price tag. A simple scripted chatbot generally costs less to set up, but it hits its ceiling as soon as questions leave the expected frame. An AI agent takes an investment matched to what it has to cover: the catalogue, the advice, preparing the cart. The right calculation is made store by store.
›Does it take longer to install than a chatbot?
No, setup can be quick. On Shopify, the advisor starts from the app you authorise inside your store; elsewhere, from a catalogue export, without touching your internal systems. A direct Shopify connection is available on top, for stock or order tracking. In practice, she can be live in a few days.
›Can an AI agent get things wrong?
Like any tool it is not infallible, which is why the guardrails exist. The register of guidance is reviewed and approved by health professionals (a dermatologist, a paediatrician, a pharmacist), and the advisor never recommends a product outside your catalogue. On sensitive subjects such as pregnancy, an allergy or a complaint, a fixed rule applies: she hands over to your team rather than answering in its place.
›Does it need access to my internal systems?
No. On Shopify you authorise the app from inside your store: the access is displayed by Shopify and revocable in one click, and no key is handed to us. On another platform, it starts from a catalogue export — you can add live stock or order tracking later, but it is not a condition to begin.
›What happens to my team once the AI agent is in place?
Your team keeps the wheel at all times. A dashboard lets them follow every conversation, step in when they need to, and take over the sensitive cases the agent passes to them automatically. The aim is not to replace the team, but to free it from repetitive questions so it can concentrate on the exchanges that need a real person.
Alexandre builds Ryma, the AI-powered advisor for online stores, and writes these guides from real deployments, with every figure attributed.