Incremental Response Model Sales Modeling - SAS Enterprise Miner

[eddieray01]

General Questions re Sales Incremental Response: What is the formula for calculating incremental response when you have treatment and control when you are given summary variables such as the overall sample(N), # responders out of sample in both Treatment and Control groups, discounted Sales revenue (discount is offer to treatment group where purchased/responded) vs. Sales revenue with no offer discount (did not get message) to Control group but still purchased, cost of the marketing campaign, and cost of producing product? I was trying to learn these concepts outside SAS Enterprise Miner to help understand more thoroughly before it does predictions but have not found any documentation on calculating this by hand. 

• Assuming cost of marketing campaign applies to Overall sample(N) not just those in Treatment group since control group also got message without treatment (in other words still cost ie..mailing postage..etc)? Or does marketing campaign cost only apply to Treatment group?  Assuming control group has to get some message to respond since they were part of sample?
• Assuming Cost of producing product would be based on Overall sample(N) * Cost? Or is it based only those that were the responders in the Treatment and Control groups?

Appreciate if anyone has context on this since I'm trying to understand it conceptually before it goes into Enterprise Miner for predictions..etc

[Cynthia_sas]

Hi:
Is this a general question for the community or do you have a question about the work in the Data Science Academy classes? If you want to ask the instructors a question about the class, we need to know the module you're working in, the title of the class and the Lesson, Demo or Exercise which you have a question about.

Thanks,
Cynthia

[eddieray01]

I am thinking general question for community but I am studying it academy for data science module Experimentation. Struggling with how one would calculate by hand given scenario. 

[TWoodfield]

Incremental response modeling predicts the incremental response per experimental unit (customer). The calculation of the overall incremental response is very simple:

Total test response – Total control response.

You could also calculate average test response, but you must average in the zeroes. What you want to know: Did we make more money with the test case versus the control?

If you are just looking at the response rate, the formula is:

Response rate test – response rate control.

If you have an unbalanced design, there may be special considerations for sample size. For equal sample size N, the incremental response rate is

(# responders test/N) – (# responders control/N)

The incremental response is

(Total sales test) – (Total sales control)

For unequal sample sizes, you would want average sales,

(Total sales test/# test cases) – (Total sales control/# control cases)

Example:

	Control	Test
# Prospects	10,000	10,000
# Responders	800	1,000
Total revenue (or profit)	9,900	10,000

Incremental response rate: (1000/10000)-(800/10000)=0.1-0.08=0.02.

Incremental response: 10,000-9,900=$100.

Incremental response (averages): (10000/10000)-(9900/10000)=$1.00-$0.99=$0.01.

Average revenue per responder:

Test: 10000/1000=$10.

Control: 9900/800=$12.38.

Conclusion: You had more responders, but each responder on average paid less, for a small incremental gain.

While you might conclude that the test was a failure, you could build an incremental response model to predict those customers who would be more likely to be influenced by the test campaign, and then apply the campaign only to those customers with a high incremental response score.

If the costs of the two campaigns are different, then you should subtract costs and work with profit.

Hope this helps.

Terry Woodfield

[Cynthia_sas · Accepted solution]

Hi:
One of our instructors sent these links which might also be useful to you:
• Incremental Response Modeling Using SAS® Enterprise Miner
https://support.sas.com/resources/papers/proceedings13/096-2013.pdf
• Incremental Response Modeling in SAS Enterprise Miner
https://www.youtube.com/watch?v=zabWaSS_BDI

Cynthia

[eddieray01]

Thank you so much.  I needed to understand  from scenario before running it through.   I appreciate the help!  

[eddieray01]

Sorry follow up. Re incremental response. If marketing cost is $2.00 and product cost let's say is $10. Then treatment group would have $12 cost per customer. Would control group also be associated with $12 cost per customer. So control group is still getting marketed with a cost from the campaign as well but no promotional offer.

[TWoodfield]

A common scenario for incremental response modeling is that the control group (champion) model is to do nothing. The existing branding and marketing effort will bring in customers. The treatment group (challenger) is to do something above and beyond the do-nothing strategy, like initiate a direct-mail campaign. If the campaign costs $2, and the product brings in $10 when a purchase is made, then the profit for the treatment is $8, whereas the profit for the control group is $10 when a purchase is made. On the other hand, suppose your control group action is to use a traditional direct-mail campaign, whereas the treatment group action is to enhance the direct-mail in some way, for example, include a calendar with the promotional material. Whatever the two actions may be, they will cost something, and I think it is unlikely that the cost will be identical. Furthermore, the campaign may be to persuade customers to login to a website and buy stuff, so the total income could be variable. Whatever the scenario, you want to model profit=total income minus costs.

I do not understand why you are adding the cost of the campaign to the income. Perhaps I do not understand your question. Are you saying the cost to produce the product is $10? If so, then if you make a sale, you get price minus cost=price-$12. If you sell the item for $20, you earn price minus cost of product minus cost of campaign=$20-$10-$2=$8. If you do not sell to customer X, then your loss is $2, the cost of the campaign. You typically do not lose the cost of the product because you can sell it to someone else. Furthermore, with production on demand models, you do not make the product and incur the $10 cost until the order comes in. It would be rare to count the cost of the product as part of the loss when no sale is made unless it is a perishable product that can only be sold during the life of the campaign.

I hope this helps.

[eddieray01]

Thank you. I was tripping up on production on demand and was thinking there was some inventory component. But the production on demand explanation is exactly what I was looking for.