How it works
From your purchases to the products to offer, in three steps. Here you will see what you need, how a suggestion is calculated and what the limits are.
Three steps
Export your history
A text file with no header. Each row is a purchase: the customer, a semicolon, the product (if you get stuck, we will help you).
C0042;coffee-beans-1kg C0042;descaler C0117;coffee-beans-1kg C0117;cups-set-6Load it
From the web app, or automatically from your own system. The import runs in the background, in blocks of 5,000 rows, and you can follow it under "Recent imports". Reloading an archive with the same name replaces it.
Get the suggestions
Your online shop or ERP asks RecommenderPro for the products to offer a customer or a cart, and shows them wherever you want. In the web app you can also look them up by hand.
History, cart and email
The same archive can be queried in three ways, each with its own menu in the web app.
- For Customer (history mode): when you know the customer. It starts from everything they have already bought. You set the minimum score (default 60) and the maximum number of suggestions.
- For Cart: when you only know what is in the cart right now. You pass the list of products, even just one, and you do not need to know who the customer is.
- E-Mailing (custom export): generates a CSV file with customer and product, holding the suggestions for every customer in the archive, ready to import into your email marketing tool. You choose the minimum score and the number of suggestions per customer, up to five. Sending the emails stays in your tool.
Where you show the suggestions, and with which thresholds, is up to you: the engine returns a list, and your online shop, ERP or agents’ app uses it as you prefer.
The score is recalculated every time from the purchases in the archive. If you load the updated history, or one archive per season (for example one for summer and one for Christmas), the pairings follow your recent purchases. There are no rules to maintain by hand.
How a suggestion is calculated
For a customer, RecommenderPro does four things.
- It takes the products the customer has already bought.
- It finds the other customers who bought at least one of those products.
- For every other product it counts how many of those customers bought it, and divides by the total number of those customers.
- It removes the products the customer already has, drops those below the minimum score and sorts the rest from highest to lowest.
A score of 72 means that, among customers with purchases in common, 72 out of 100 also bought that product. It is an observed frequency, not the probability that the customer will buy.
For a cart the same reasoning applies, starting from the products it contains. An empty cart falls back to the 30 most frequently bought products in the archive. You set the minimum score (default 60) and the maximum number of suggestions.
Where it runs
In the cloud, or on your own servers. On your servers the engine is a 64-bit Windows application and uses PostgreSQL. The web app is available in Italian, English, German, Spanish and French.
The limits, stated up front
- There is one engine: it counts how often products appear in the same purchases. It does not look at prices, margins, categories or star ratings, and it has no seasonal model: seasonality comes in through the purchases you choose to load, for example one archive for summer and one for Christmas.
- Loading is always complete. There is no way to append rows to an existing archive: you reload the updated file.
- The score describes what happened in your history. If your catalogue changes a lot, the suggestions realign when you load new data.
- With few customers, or with purchases not tied to a customer, the comparison has little to work with.
Want to see what it would suggest to your customers?
We show you the web app and how it connects to your online shop or ERP. If you decide to start, we work with you on the integration.