Suppose you are a manufacturer, you plan to sell your products to retail stores, how would you decide which retail stores to connect first given the huge amount of retail stores in US?
A good solution is utilizing machine learning techniques to build up a prioritization model and generate a score for each retail store.
The end goal of this solution is to maximize your potential revenue/profits by saling your products to the best retail stores.
To generate a prioritization score for each retail store, you need following things：
Dun & Bradstreet (D&B) data, Dashmote data, and Google public data.
Dun & Bradstreet is a global company that provides business intelligence products to clients through its database and analytics software. The products are used in improving business profits, marketing, and risk management.
Dashmote is a leading data democratization platform for food, beverage, manufacture, sales & marketing.
2. Feasible model to use - Classification
Classification: The score is the likelihood of being an attractive prospect/customer/retail store.
Training data: Two types of retail stores - Customers and Non-customers.
Customers: Retail stores who have done business with you.
Non-Customers: Those never done business with you and existed for a long time.
Testing data: Those new retail stores.
Output: You don’t know weather the new retail stores would do business with you, but the prioritization model will calculate the likelihood.
Strategy: You just need to connect with those new retail stores with high score.
2.1 The most influential features in an exmaple are as follows:
2.2 The F1 score for classifiation model in an example is 95.4%.
You can attempt various classifiers like Xgboost, Randomforests, Catboost, Adboost, etc and ensemble them in a stacked structure.
3. Feasible model to use - Regression
Regression: The score is the predicted transaction amount (dollor) or prodcut volume.
Output: The prioritization model will anticipate the transaction amount (dollor) or prodcut volume of a new retail store.
Strategy: You just need to connect with those new retail stores with high predicted transaction amount (dollor) or prodcut volume.
Advantage: The advantage of regression is that it will prvide the transaction and revenue forecasting.
3.1 The most influential features in an exmaple are as follows:
3.2 The MAPE score for Regression model in an example is 5%.
You can use various metrics to measure the regression accuracy like R-square, MAPE, RMSE, etc.
4. Ultimate Conlusion
You should run both classification models and regression models and prioritize those new retail stores with high likelihood of being attrctive customers from classification and high transaction or products volume predicted by regression.