Posted on 2021-08-03
Ryax Technologies Ryax Technologies

Predict which clients among your CRM will purchase again in a defined period

+ More details


Acquiring new clients might be easy, retaining them is way more difficult. Stop putting efforts and clienteling on non-revenue driving clients and focus on those whom probability to come back is the highest. This is what this analysis will offer - why don't you give it a try?

You can choose to export this customer list to SalesForce or any other CRM tool to automate your marketing campaigns, by focusing on the high value customers only. You can also use this analysis to prepare client listings for future events or dedicated CRM actions.

Analytics status

  • Available

Business benefit

Running this analysis has a positive incidence on the following metrics:
▪ total turnover
▪ AOV (Average Order Value)
▪ items per basket
▪ client acquisition and retention

Data inputs (mandatory)

We'll be using historical sales data containing the following information :
▪ InvoiceNo : ID of the actual purchase
▪ StockCode : SKU number (or SKU ID)
▪ Description : written description of the item/SKU
▪ Quantity : number of the same SKU being sold to the same client (if applicable, otherwise it will be "1")
▪ InvoiceDate : actual purchase date
▪ UnitPrice ; SKU/Item unit price
▪ CustomerID : ID number of the client
▪ Country : Country where the purchase has been made

Data Output

You are delivered the list of your customer IDs along with the prediction results :
▪ "1" : the customer will come back within the defined period
▪ "0" : the customer will not come back within the defined period

Technical description

This algorithm mixes two different techniques :
▪ first one is called RFM (Recency, Frequency and Money), that is being used in our regular customer segmentation analysis. It does correlate recency (last time your purchased from observation date), frequency (number of purchases over the last 12 months) and Money (average basket value)
▪ second one is an XGBoost model giving a binary answer to a specific question. In our case, it is answering the following : "Is my client going to purchase in the next 30 days?". Answers are probabilities converted to binary class values rounded to 0 or 1.


Clients must have purchased at least once in the last 12 months from observation date.