As a modern marketer, your team is likely rich with data. You have a marketing automation platform that contains all of your leads and the programs you ran against them. Hopefully, it also has the outcome of those marketing programs in terms of MQLs, and perhaps SQL and Wins. The problem many of us face is not knowing what to do with all the data we have. How do we use it to improve our marketing and drive more revenue? This blog post identifies an advanced analytics technique that will help you determine what combinations of marketing programs lead to superior returns as a best-in-class data company.
Complements and Substitutes
Think back to your first economics class, do you remember complements and substitutes? Complements are things that go together like peanut and butter and jelly, diapers and beer, or a tape deck and Thriller cassette. Substitutes are things that are similar to each other, so you can use one in the place of the other. Coffee and tea are substitutes, just like butter and margarine, and all fantasy football kickers (sorry Gostkowski).
Your marketing programs are no different. There are some programs that complement one another and others that don’t. Perhaps when you run a webinar it is best if you complement it with a drip email program. When you run these two programs they produce higher returns compared to when you run them separately. Other marketing programs, however, are substitutes for each other. When you run one, you do not need to run the other. In some cases, running both programs at the same time produces diminishing returns.
As you peer into the depths of your marketing automation platform you may be left wondering, how would anyone ever figure this out? Certainly, you have a gut feeling about which programs are complements and which are substitutes, but how do you back up those feelings with data? We like to use a data mining technique called association analysis to help answer this question.
Association analysis is a technique large retailers use to uncover patterns in their transaction data. Each transaction contains all the items that a customer purchases in a single transaction. From it, you can create a set of IF/THEN rules. For example, IF people purchase peanut butter and jelly THEN they are likely to buy bread.
Marketing Database Transactions
Similar to the supermarket example, you can use association analysis to examine the programs in your marketing database and create a set of rules. To do this, think of each of your leads as a transaction at the grocery store. Every marketing program they were part of can be thought of as the items they purchased (antecedent). The resulting MQL, SQL, or Win is the outcome of those marketing programs (consequence). For example, IF “Super Cool Webinar” and “Awesome Drip Email” programs are run on a lead THEN the outcome is a Win.
IF “Super Cool Webinar” AND “Awesome Drip Email” THEN Win
Measure the Outcome
Your marketing database contains many combinations of rules, so how do you tell which rules are good and which are flukes? Association analysis examines all possible IF/THEN rules and helps you select those that most likely indicate a strong relationship between the program combinations and the desired outcome. There are myriad metrics that data scientists use to evaluate the validity of each rule, but we will focus on the big three.
The first metric is support. Support is simply the number of times the rule happens in your dataset. Because “Super Cool Webinar” and “Awesome Drip Email” resulted in a Win 25 times out of a database of 1,000 leads, the support of this rule is 2.5%. A higher support number indicates that the rule happens more frequently.
Support = [(25 / 1000) * 100%] = 2.5%
The second metric to consider is confidence. Confidence helps you understand the uncertainty of a rule. It compares the number of times a combination of marketing programs resulted in the expected outcome (MQL, SQL, or Win) to the number of times the outcome did not occur. While “Super Cool Webinar” and “Awesome Drip Email” resulted in 25 Wins, there were 5 leads with the same programs that did not go on to Win. Confidence in this rule is 83.3%. A high confidence indicates a strong association rule.
Confidence = [(25 / (25 + 5)) * 100%] = 83.3%
The third metric to consider is the lift ratio. The lift ratio is a better way to judge the strength of an association rule. It is a ratio between the rule’s confidence and a benchmark confidence. The benchmark confidence is the number of times the expected outcome occurred versus all transactions. In our example, there are 100 wins in our database of 1,000 leads. The benchmark confidence is 10%. The lift ratio is 8.33. A higher lift ratio implies higher confidence in an association rule.
Benchmark Confidence = [(100 / 1000) * 100%] = 10%
Lift Ratio = [83.3% / 10%] = 8.33
Your marketing database is full of data. It is time to uncover new insight to help propel your business to the next level. You can do this with association analysis. It will quickly identify which programs work well together (complements) and which do not (substitutes), so you can adjust your future programs to take advantage of this multiplier effect.
At ORM Technologies we are experts in advanced analytics. We focus on helping our customers make sense of their sales and marketing data. If you have questions on using association analysis or would like to know how we can help you please shoot us a note at firstname.lastname@example.org.