Next Best Offer (NBO) and its Application in the Retail Industry
What is NBO?
In the retail industry, NBO is often used to increase customer loyalty, increase sales, and increase customer satisfaction. For example, by analyzing a customer's past purchases and style, the most suitable products or promotions can be offered to that customer. This can increase customer satisfaction while meeting the customer's needs and increasing the company's revenue. NBO's aim is to provide the customer with the most suitable offer by personalizing the customer experience, thereby increasing customer loyalty and increasing sales.
How Does CLV Work in the Retail Industry?
Data Collection: The first step is to collect customer-related data. This may include shopping history, demographic information, interactions on the website or app, loyalty program information and other relevant data.
Data Analysis: The collected data is examined using data analysis methods. This analysis is performed to understand customers' shopping habits, preferences, purchasing history and other important characteristics.
Customer Segmentation: Based on the analysis results, customers are divided into different segments. These segments may include groups of customers who share similar behaviors, preferences, or demographic characteristics.
Personalized Recommendations: For each customer segment, the most suitable offers or products for that segment are determined. This is accomplished in a customized way based on the characteristics and behavior of the customer segment.
Offer Presentation: The determined suggestions are presented to customers through appropriate channels. This can be accomplished using a variety of channels, such as email, mobile app notifications, website pop-ups, or in-store promotions.
Feedback and Improvement: Customer reactions and sales results are monitored and analyzed. This feedback is used to improve NBO algorithms and bid strategies. It is constantly optimized to achieve better results.
Technical Implementation of the CLV Model
1.Data collecting
We need 5 data tables.
Revenues data: This data shows how much revenue was generated historically from which customer and from the product the customer used. In other words, if we have a customer and there are five products that this customer uses, it is shown how much income was generated from these products on what date.
Customer information: Here you can define customer-specific information along with the customer ID. This may be information such as gender, region, income.
Product holdings: In this data, for each customer, there are rows in the form of how many products there are, the name of those products, their id, start date, and end date if finished, equal to the number of customers multiplied by the number of products.
Balances: Similar to revenue data, all balances in the form of date, customer and product are kept here.
Additional information: Finally, if there is additional data you want to use in modeling, you can provide it in the additional information table.
2.Data Cleaning and Preparation
After all data is provided, necessary arrangements are made for modeling and analysis. Using this data, we first make a subscription prediction. Let's say you have 5 products in total in the 12-month data you provide and 3 products that a customer of yours uses. While assigning a target variable, we set a flag as 1 if the customer has that product in the relevant months, otherwise 0.
In other words, for all the products you provide, whichever product is available for all 12 months is assigned as the target. Then, based on the customer movements in that month, the customer's probability of subscribing is estimated. In this way, it is determined how likely your customer will use the products he does not use. We also enable you to make short-term or long-term predictions such as whether customers will subscribe in the next 3 months or within 6 months. After estimating the subscription, some parameters are selected to be used in cross-sell analysis.
3.Model Selection and Training
After these settings are made, you can make your B2metric automl settings and choose the algorithms you want to train. You can also give various hyperparameters to improve model performance. If there is missing data in your data, you can determine your filling strategy and model validation strategies.
4.Estimate and Evaluation
Customer base analysis: You can see various metrics of your customers. For example, your total number of customers, average account ages, average age of your customers, tier distribution of customers, etc.
![](https://framerusercontent.com/images/umlzDuh8kYYPVTPLVyI69Pgwg.png)
Product holding analysis: Depending on your products and customer ownership of these products, you can see metrics such as how many product types and types of products you have, how many distinct product types and product mixes you have, what is your highest-income product, what is your lowest-income product.
![](https://framerusercontent.com/images/1SGYoWY2nLWaXbRKWt52do7oVU.png)
![](https://framerusercontent.com/images/4s9axpCilIFVDAqsdMpoAEYee5s.png)
Machine learning model analysis: You can see the error metrics, confusion matrix and importance ranking of the features used in the subscription probability model you trained.
![](https://framerusercontent.com/images/e1cI9bs6dOU336lwsRWtKpbmGs.png)
Total Impact (Expected Impact): This concept expresses the total financial impact of adding a new product to the existing product mix. For example, if your customer's existing products are A and B and you plan to add product C to this customer, your goal is to calculate the additional revenue that this change will bring.
Current Product Mix: Products the customer currently has (A and B).
New Product Mix: Customer's products (A, B and c) when the new product is included
The average revenues of A and B are calculated among customers using the A-B mix. Next, the probability of subscribing to product C is calculated. For the new mix (A-B-C), the average revenues of products A, B and C are calculated. As a result, the probability of buying product C is multiplied by the difference in average income between the old and new mixes.
Direct Impact: This is the probability of the customer buying the new product (C) multiplied by the average revenue of this product.
Indirect Impact: It is the difference between Expected impact and direct impact. If the indirect impact is greater than the direct impact, the indirect impact of the added product on the sales of other products is greater than its direct impact. Conversely, the direct effect of the product is greater than the indirect effect.
Products with the Highest Impact Value: This metric shows the products with the highest impact value.
Average Cross-Sell Impact Per Product: Average cross-sell impact values are shown for each product. The products most suitable for cross-sell are the products with the highest total expected impact.
![](https://framerusercontent.com/images/bTc19qfTFoyooKEfRvEXG3m4g.png)
Top 10 Average Revenue Per Product Mix: Shows the product combinations with the highest average revenue.