What is an AI Layer for e-commerce

What is an AI Layer

«Layer» is the exact word: this is neither a separate application nor a replacement for the platform, but a level that sits on top of an existing store and reads what that store already knows — catalogue, prices, stock, orders, customer history — making it queryable in words.

The difference from the software that came before is exactly there. An internal search engine matches strings; a chatbot recites answers written in advance. An AI Layer interprets a whole sentence and then does something: queries the catalogue, compares two products, reads a shipment status, opens a ticket when a human is needed.

How it differs from a chatbot or an internal search

In all three cases the customer types into a box. What happens next is not the same.

The clearest example is a request such as «I'm looking for comfortable shoes to walk around town with my daughter, under €100». It contains no category, no brand, one price constraint and two hints of context. Keyword search returns nothing; a rule-based chatbot asks you to rephrase; an AI Layer turns it into a selection from the real catalogue, with products that are actually in stock.

What it is made of

Such a layer is not a single language model. It is a set of parts working together:

• An understanding of the request, extracting intent and constraints from the customer's sentence.
• A semantic representation of the catalogue, so a product can be found even when the customer's words are not the ones on the product page.
• A connection to live store data: today's price, current stock, that order's status.
• A knowledge base with the store's own answers on returns, shipping, sizes, warranties.
• A set of actions the assistant can take: show products, compare them, add to cart, open a ticket.

The visible part — the chat — is the least interesting. The value is in the data connection: without it, what remains is a model that speaks well and knows nothing about the store.

Which data it reads, and when

A well-built AI Layer reads data at the moment it is needed, not from a copy that is a day old. That is the difference between recommending an out-of-stock product and not recommending it. In Flow-e's case the reads cover catalogue, prices, stock and orders, and happen on demand, during the conversation.

Preferences the customer expresses — a brand they like, a size, a price limit — remain available for later conversations: on the second visit the assistant starts from what it already knows.

What changes for the merchant

There is an effect that almost always goes unnoticed, and it is the most useful one: conversations are a measurement. Every request that returns nothing is information about a product customers look for and the catalogue does not carry, or carries under a name nobody uses.

Traditional analytics does not produce this. A failed search leaves a row in a report; a conversation leaves the exact sentence the person used.

When it makes sense, and when it does not

It makes sense when the catalogue is large enough to make finding things hard, and when customer questions repeat. 80% of people abandon a site after a fruitless search or if the answer does not come quickly (Google Cloud / The Harris Poll, 2023 · Forrester, 2016), and around 70% of support tickets concern a small number of recurring questions (Gorgias, 2024).

It makes far less sense on a twenty-product catalogue, where the customer sees everything on one screen, or on a store with no traffic: a layer of intelligence above zero visitors is just a cost.

How it is switched on

On Shopify, activating Flow-e takes about five minutes and no code: you install the app, the store connects itself, and the widget is switched on from the theme editor. The same switch turns it off, with nothing to uninstall.