We learnt Correlation in Statistics 101. A quick refresher on Correlation is below:
- What does Correlation mean?
- Positive correlation means that there is a strong linear relationship between two variables indicating they move in the same direction.
- Negative correlation means that there is a strong linear relationship between two variables, but they move in the opposite direction.
- Zero correlation means that there is little or no relationship between two variables.
- Correlation Coefficient (CC) Range
- Correlation Coefficient ranges between -1 (perfect negative correlation) to +1 (perfect positive correlation).
- Correlation Coefficient Significance
- 95% Confidence means that the result is at least 95% accurate and there is 5% chance of error. In general, 95% level is an industry standard.
2. Industry usage
How can we use correlation to do Cross Channel Analysis?
As ads on all screens ( TV, Desktop, Mobile, Tablets) and Print increase we expect to see more website traffic, more sales etc and vice versa. Both media and sales should move in one direction so we expect to see strong positive correlation between dependent variable (online visits/sales) and independent variables (Digital Ad Impression & Traditional GRPs).
What data we need?
Website Visits and Sales are dependent variables depending on increase or decrease in media/advertising activity. TV GRP, PRINT GRP, Display Impressions, Paid Search Impressions are independent variables being controlled by advertising brand manager.
- Independent Variables: Traditional GRP: Both TV and Print, Display Impressions, Paid Search Impressions.
- Dependent Variables: Sales and Different kinds of online visits numbers.
Both data sources should have the same time period such as daily or weekly etc.. As we have continuous data we should use Pearson’s correlation to determine significance. This can be done using EXCEL Stat, SPSS, SAS or any other Statistics tool.
How does Cross Channel Correlation Analysis help?
Correlation Analysis can help understand synergies between various media channels (Display, Search, TV & Print) and online visits behavior or Sales.
Evaluating correlation and trending for various media channels with sales data can help identify most effective media channels or a combination of them.
What kind of visuals can be produced?
What kind of insights can be produced?
- TV and Print media are Impactful in Driving Both Paid and Organic Search Visitors.
- TV and Print show highest positive correlation with Direct Visits.
- TV commercial showed a strong impact on all kind of visits (Organic/Paid Search Visits, Direct Visits) to Website and final sales.
- Print Showed Strongest positive Correlation with Sales in a specific time period Compared to Other Media.
- Display Media Had Lower positive Correlation with Sales Compared to TV and Print.
How is Cross Channel Correlation Analysis Actionable?
- Insight : TV GRP show positive Correlation with all sorts of online visits.
- Action: It is imperative to have Website url in your TV commercial.
- Insight: TV commercial do not show any correlation with sales.
- Action: With tight budgets, brand manager can be more comfortable with media budget allocation and introducing/cutting a particular media channel.
- Insight: “Paid Search impressions did not show positive correlations with Paid Search Visits. However, increased TV GRPs show an increase in Paid Search Visits.”
- Action: Launch Paid Search Campaign with TV campaign rather than Paid Search or TV Campaign alone.
We have identified our independent and dependent variables. Is it possible to develop a model to predict sales using this set of variables? Many companies claim to predict future sales based on advertising mix. But note that there is no way to determine customer journey in Online + Offline as Non- line (as some Marketers call it!) world. Several times, it is not possible to connect consumer data where consumer is touching the various channels. In absence of connected data actionable Cross Channel Correlation Analysis is a powerful tool.
Now is your turn. How can we make this more powerful? Do you have other ideas? Do you like the idea above? Are you already using such analysis? Please share your thoughts via comments.