I recently started working with a new client. They have been very diligent with test-and-learn in their marketing efforts. They have accumulated lots of lessons to share.
One of the lessons is that digital ads amplifies the results from their unaddressed direct mail (i.e. geotargeted direct mail).
[There have been many studies on the symbiotic relationship between direct mail and digital. For example, Canada Post has published a study showing that direct mail followed by pre-roll results in a 3% lift in motivation. Although this lift sounds small it does drive meaningful changes in behaviour.]
Use Data Science To Go Beyond The Obvious
Based on this learning, their analytics team suggest that marketing should always pair their unaddressed mail campaigns with digital ads. While the conclusion seems logical, it’s only scratching the surface of much richer insights that can be had.
Data science can help marketers get to smarter conclusions if they ask the right questions.
I always remember one technique I learned from the super inspiring Leslie Ehm—the 5 “whys”. She told us to find new ideas, ask “why” 5 times when someone answers a question. It digs into what may seem fairly obvious. That’s what I’ve adopted whenever I do analysis.
So if we know digital ads amplify the results of geotargeted mail, the next why can probe into segments within the universe—Are particular geographic areas where that works better or worse? Does timing of the ad makes a difference? Can retargeting lift the amplification effect even more?
But how? Here is how I’ve used marketing mix analysis to help a client answer some of these whys.
An example of Marketing Mix Analysis - How Data Science made targeting more efficient
Background: Why did we do the analysis?
This client doesn’t sell direct to consumers. They rely on other professionals to sell to individual consumers. As in most business-to-business (B2B) settings, they have sales reps who meet with these professionals regularly. The reps play a central role in the sales/marketing mix. Marketing supplements these reps with the usual suspects—breakfast meetings, webinars, emails, magazine ads and the likes.
There are 3 underlying problems with this sales model:
- Having reps travel around the country to meet with each professional is very costly.
- With our busy lives, it is increasingly difficult to schedule sales meetings.
- Marketing had analyzed the ROI for each tactic, but want to be more targeted with their efforts.
So similar to my new client, they knew different marketing activities can amplify the results of the more expensive reps activities. But instead of just layering on these activities, they want to be smarter of who they want to layer on what activities. This is where my team & I came in with the marketing mix analysis.
Process: What analysis did we do?
The marketing mix analysis was done using a number of regression models in R.
We hypothesized that professionals in different stages of the sales funnel may respond to marketing and sales activities differently. For example, those who are new to the industry may need more hand-holding, while those who are familiar with the products may want more self-service options. So we built a separate regression model for each stage.
We looked at how marketing activities over the past year impact the behaviour of these professionals. What we were interested in was to pair up the right groups of professionals with the most efficient mix of marketing activities.
This model was then turned into a “calculator”. The media team used it to determine whether activities targeted at certain groups are worth the investment.
While I can’t divulge all the secrets, I’m going to share a few insights we got from the analysis and how they were useful.
Insight #1: Combinations of some tactics can have exponential effects
The main output of the models was the likelihood of the professionals to act based on marketing and sales activities. Here is one of the charts we compiled. It shows the likelihood of a certain segment of professional to “convert”, i.e. moving from one stage of the sales funnel to the next, grouped by the combination of marketing tactics and sales activities that were targeted towards them. This is all compared with the base of having no activities targeted towards them.
It was surprising for us to see the big difference in the first 2 bars. The difference between the top converting tactic combination and the second one is a just set of emails. In other words, if we target a professional with a combination of tactics excluding email, they are 163 times more likely to convert to the next stage. If we also target these same professionals with the email series, they become 390 times more likely to convert! However, email on its own is one of the small contributors at the bottom of the chart. The power of combinations!
Insight #2: Non-targeted tactics can have a big impact to targeting
One reality data marketers have to deal with is that there are mass influence that cannot be targeted. For example, in B2B, we may place magazine ads that are not personalized or targeted. In B2C (business-to-consumer), we may have TV, radio, billboard, etc. that are not one-to-one.
When we look at the influence of the mass media, we can see that professionals who have higher exposure to mass media are less influenced by targeted tactics. This means if we know we are placing heavy mass media in certain geography or industry, we can dial back on more personalized marketing tactics, especially the higher cost ones such as breakfast sessions.
Insights #3: Can't go cheap with some industries
Another thing we learned was that certain industries are just not responsive to anything other than sales reps. This means prioritizing sales reps for professional in these industries can have a bigger impact. On the same note, it means we can save money by prioritizing more scalable and cheaper marketing tactics for professional in industries that are receptive to targeted marketing.
I hope this inspires you to look further at how data analysis can be so much more helpful to marketers when it pushes the boundaries of the most obvious observation. Ask the 5 “whys” next time when you look at analysis results and see if you can find a smarter question to ask your data model.