Guiding business to better decisions
1. Guiding business to better decisions
The cumulative gains and lift graph not only are great ways to visualize the performance of your model, they are also great tools to make better business decisions.2. Estimating profit
One way to make smart use of lift graphs, is to estimate the profit that you can make with a campaign. Consider for instance the lift graph given here. The population consists of 100 000 candidate donors, and 5% among these candidate donors is target. Assume that we expect targets to donate 50 Euro, and that addressing a donor, for instance by sending him a letter, costs 2 Euro.3. Estimating profit
Given this information, we can calculate the expected profit of a campaign. The profit depends on 5 elements: the percentage of targets in the selected donor group, the percentage selected donors, the population size and of course the reward of reaching a target and the cost of the campaign. Indeed, the total cost of the campaign is the cost of the campaign, 2 Euro, times the number of donors addressed, which is the percentage of selected donors times the population size. The reward of the campaign is the reward of reaching a target times the number of targets reached. This is the percentage of targets times the number of donors addressed. The final profit is then the reward minus the cost.4. Estimating profit
Assume that you address the top 20% donors with the highest probability to donate according to the model. From the lift curve, you can see that the lift at 20% is 2.5, meaning that the top 20% contains 10% targets. Using the profit function, you can observe that this results in a profit of 60 000 euro, which is of course a good result. If you would address all candidate donors, you can calculate that you expect to make 50 000 Euro loss.5. Campaign selection
The cumulative gains graph can be used to decide how many donors one should target if one wants to make a certain profit. Assume that you want to send out a campaign via e-mail. You have a pool of 1 000 000 candidate donors with 2 percent targets, but you don't want to send an e-mail to all donors as you don't want to bother candidate donors that are not interested in donating for this campaign. You built a model to predict which candidate donors are most likely to donate, the cumulative gains graph is given here.6. Campaign Selection
Assume that you want to reach 16 000 targets, which is 80% of all targets. You can read from the cumulative gains curve, that in order to reach 80% of the targets, you need to address the top 60% of the candidate donors, which is 600 000 donors. Without the model, you would have need to send an e-mail to 800 000 donors in order to reach the same number of targets.7. Let's practice!
Let's put this into practice and consider some other business cases in the exercises!Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.