An AI sales advisor for Shopify is a conversational system built into the store that asks the buyer about their needs and taste, recommends suitable products from the catalogue and guides them to purchase. It pays off for products where buyers take longer to decide and, in a physical shop, would ask a sales assistant: perfume, cosmetics, electronics, equipment with many variants. In Slovenia, PIKA. Digital develops them. For MINE Perfume Lab we built Fragrance Concierge across three Shopify markets.
How does an AI sales advisor work?
The advisor has three layers: product data, the conversation flow and the connection to the store.
- Product data. It takes descriptions, attributes, prices and stock from the catalogue. Shopify lets you add attributes that the standard fields don't cover (fragrance notes, material, suitability for a skin type) as metafields.
- Search by meaning. The buyer doesn't type a product name but "something fresh for summer". The system finds products that match that description by meaning, not just by keyword.
- Conversation flow. The advisor asks a few questions, narrows the choice and explains why it recommends this particular product.
- Connection to the store. The recommendation leads to the product or the basket, with the price and terms for the buyer's market.
How is it different from a traditional chatbot?
| Traditional chatbot | AI sales advisor | |
|---|---|---|
| What it can do | Answers pre-written FAQs | Recommends products based on the conversation |
| Data | A list of questions and answers | Catalogue, product attributes, stock, prices |
| The buyer needs to know | Exactly what they are looking for | Only what they like or need |
| Goal | Fewer support questions | A purchase decision |
When does an AI sales advisor pay off?
Checklist:
- Buyers take a while to decide and often ask what to choose.
- The catalogue has enough products or variants that the choice isn't obvious.
- The product can't be tried online, and a description and photo aren't enough.
- You sell in several markets or several languages.
- Your product data is in order, or you will put it in order.
It doesn't pay off with a small catalogue where the buyer chooses in a few seconds, with low-priced impulse purchases, or when the product data is not in order. The advisor only recommends as well as the data it works from.
Example: MINE Perfume Lab
MINE Perfume Lab sells Italian, hand-made niche perfume in Greece, Cyprus and Bulgaria. Without a sample, a buyer of niche perfume finds it hard to decide from a photo and a list of notes. The standard Shopify chatbots could only answer FAQs.
We built Fragrance Concierge, an AI advisor embedded in their Shopify store:
- It detects the market and shows local prices and delivery and returns terms.
- It speaks English, Greek and Bulgarian, and the buyer can switch language mid-conversation.
- It recommends by preference: it doesn't ask for perfume names, but for favourite notes, mood and occasion.
- It guides buyers from a sample set to a full bottle with a built-in conversion flow.
The system is built with no vendor lock-in, and the code stays owned by Mine. The example shows what we built. We don't publish sales results for it. The full description is in the MINE Perfume Lab case study.
How does the setup work?
- Review of the buying journey. Where buyers stall, what they ask support, which question a sales assistant in a physical shop asks first.
- Putting the data in order. We write the attributes buyers choose by into the catalogue so the advisor can read them. This is usually the biggest part of the work.
- Conversation flow. We decide which questions the advisor starts with, how many it asks and when it suggests a product. Fewer questions are usually better than more.
- Testing on real questions. Before launch we test the advisor on questions buyers have actually asked in the past, and correct answers that aren't accurate.
- Launch and measurement. After launch we monitor the conversations and the orders that went through the advisor, and keep refining the flow.
This example is made up. A running gear store has a few hundred products in different sizes and for different uses. A buyer writes that they run twice a week on gravel and have knee problems. The advisor also asks about distance and surface, then recommends three models with an explanation of why they fit, and offers a size chart. If the buyer asks for medical advice, the advisor replies that it cannot give it and suggests seeing a specialist.
What do you need before you start?
- A catalogue in order. The attributes buyers choose by must be written in the data, not just shown in photos.
- Rules for what the advisor must not say. No health claims, no made-up attributes, no discounts that don't exist.
- Handover to a person. Questions about an order, a complaint or special requests go to a person.
- Measurement. Number of conversations, share of conversations with a click on a product, add-to-baskets and orders that went through a conversation. Without this you don't know whether the advisor sells or just answers.
What are the risks?
- False claims. A language model can attribute a property to a product that it doesn't have. The advisor should answer from catalogue data, and a person should review the answers in the first weeks.
- Transparency. The buyer must know they are talking to AI. On the AI Act, the European Commission lists this as a transparency obligation for chatbots.
- Personal data. Conversations may contain personal data. Store only what you need and cover it in your privacy policy.
- Vendor lock-in. Check who owns the code and what happens to the advisor if you end the collaboration.
Where to start?
AI systems for stores and sales are described on the Systems page. How we connect the advisor with ads and creative so that the right traffic reaches it is on the performance marketing for online stores page.
Sources
- Shopify Help Center, Metafields: https://help.shopify.com/en/manual/custom-data/metafields (accessed October 2026)
- Shopify Help Center, International sales tools: https://help.shopify.com/en/manual/international/markets (accessed October 2026)
- European Commission, AI Act: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai (accessed October 2026)