Amazon generates 35% of its revenue from recommendations. You generate 0%.
The number is not a myth. Amazon has publicly confirmed that recommendations like "Customers also bought..." and "Inspired by your browsing" drive over a third of all revenue. The logic is simple: the customer is already a buyer. The persuasion is done. All that's left is to suggest what they actually need.
In Bulgaria the picture is different. Most small and mid-size online stores have no recommendation system. Neither automated nor manual. The customer buys, gets a confirmation, and that's it. The money left on the table is real.
That's how much of Amazon's total revenue comes from AI product recommendations. For a Bulgarian store with BGN 20,000 in monthly turnover, even a 10% lift means BGN 2,000 more every month.
How the recommendations AI agent works
The classic recommendation system shows "Similar products" on the product page. That's fine. But the AI agent does far more.
The agent analyzes three things at the same time:
- Order history: what the customer has bought so far and when.
- Real-time behavior: which products they are currently browsing, how long they stay on a given page.
- Patterns of other buyers: what people with a similar profile buy together.
From these three streams the agent builds a personalized recommendation. Not generic. Not random. Specific to this customer, at this moment.
The recommendation appears in two places with the highest impact: at checkout and after the purchase via an automated Viber or email message.
A real example from the Bulgarian market
Picture a pet supplies store. A customer buys a 15 kg bag of dry dog food. The AI agent sees that the same customer bought treats and a toy three months ago. It also sees that 68% of people who bought this food also add a joint-health supplement.
At checkout the agent offers the supplement in one click. Price: BGN 14.90. The conversion rate on such checkout recommendations is between 8 and 15%.
Then, 30 days later, the agent sends a Viber message: "Max's food is running low. Order again and get 10% off." No manual work from your side. No need to remember who bought what when.
- The customer buys the core product.
- The AI agent offers an add-on at checkout.
- After a set period the agent sends an automatic reminder.
- Average order value grows. The customer returns more often.
How realistic is the 15 to 25% lift
The number depends on three factors: store type, average number of items per order, and purchase frequency. Stores with repeat purchases, like cosmetics, pet supplies and supplements, see a stronger effect. Stores with one-off purchases, like electronics, see a smaller but still meaningful lift.
The practical math is as follows. A store with 500 orders per month and an average order of BGN 65 generates BGN 32,500 in revenue. With a 20% lift in average order value the revenue becomes BGN 39,000. The difference is BGN 6,500 per month. Without new customers. Without more ads.
That's exactly why AI recommendations are not a luxury for big platforms. They are a tool with measurable return for any store with repeat customers.