How to check the impact on marketing activities — marketing mix modelling
We often see marketing activities in almost every sector now, yet the impact generated by that particular activity is always hard to capture. Data on the other hand has power to capture the impact or the return on the investment made on any marketing activity.
Marketing Mix Model
With the variety of advertising channels through which we can arrive at new potential customers expanding each year or increase sales revenue, it’s a higher priority than any time in recent memory to have a comprehension of which channels are driving the business forward and which are definitely not.
Market Mix Modelling (MMM) is a strategy which helps in evaluating the effect of a few showcasing contributions on deals or Market Share. MMM helps in the finding out the adequacy of each advertising contribution to terms of Return on Investment. At the end of the day, a promoting channel with better yield on Investment is more powerful as a medium than a showcasing contribution with a lower ROI.
MMM utilise the multilevel regression technique where we take into consideration the sales data along with data from various marketing channels.
It is easy to capture the return on investment for the online marketing activities as we can follow its hits, likes as well as the clickstream database. But it is equally hard for the offline activities like advertisement on TV, print media or any BTL activity, in a way that a company can’t say whether the increase/decrease in sales is due any particular marketing activity or is it just any external factor which triggered the sales for them.
In order to build a marketing mix model, we use past transactions data along with the relevant marketing activity data.
Let’s discuss more about the data to be used:
We can make use of the following type of time series data:
- Number of hits, likes, views, share etc.
- Sales/transaction data
- Number of people landed on portal/website or number of leads etc.
- Marketing spends data for different channels
- Seasonality data
- Competitor dataset if possible
A very basic level equation of a marketing mix model is as follows:
Where S_t the total sales, S_b the base sales i.e. expected sales with being impacted by the marketing activity introduced in the model, S_1,S_2,S_3… are the sales activity of the marketing channels, x_1,x_2,x_3… are the coefficients associated with different channels.
Here S_t -S_b(also known as incremental sales) is the overall contribution by the marketing activities and hence individual contributions are directly proportional to the amount spent on each channel.
There are few important things to keep in mind:
Seasonality or external factors
While building a market mix model, we must always keep in mind the impact of seasonality of the particular product. So for example, we can say we see the spike in sales of the ice creams in summers, or if there is any change in tax rules for automobiles we may see the change in the sales pattern accordingly.
Marketing decay or Carry Over Effects
Most of the marketing activities won’t give the immediate impact. Say we see a billboard about a product daily and only make our mind to purchase after constantly seeing it for a month or two, so we see the actual impact of any marketing activity after a month not soon after doing the activity. In this way, there is a period of contrast between when we put any promoting into the world and when we see a visit, buy, information exchange, and so on. This time distinction is by and large alluded to as the persistent impact. And this delay in a time period will vary as per product (for small daily usable it could be a week or so and for bigger things like purchasing a car, it could be a long period).
We can directly use these carryover impacts in our model by simply multiplying by a factor(between 0 and 1) for a desired time period or maybe use the radioactive decay concept to determine the carryover impact of any activity.
In the below mentioned dummy example we can see that the carryover impact reduces for any activity as days pass on. So we must test the impact for each channel carefully and use that combination of carry over rate (adstock) accordingly.
There are multiple ways to model these relationships of diminishing returns (negative exponential function, power function, etc.), so it’s now up to us to test which approach gives the best fit to the data.
There are numerous approaches to demonstrate these connections of consistent losses, so it’s presently dependent upon the test which approach gives the best fit to the information.
Reaching the threshold for spends
As we can observe in the below chart that there is a non-linear nature in the impact vs. the marketing spend. It is because the sales will meet the saturation at some point of time as per the capacity to serve the customer or as per the actual need in the market but there is no limit to the marketing spends (one can spend as much as they want to).
Expected Result
We might get the return on investment channel accordingly. Below is an dummy example for the evaluation part
Here as we see for channel 1 and 2 we get the ROI of 2.3 and 1.8 respectively, which means for every 1 rupee we spend we get return of 2.3 and 1.8 respectively. In case of channel 3 we get only 0.3 rupee for each 1 rupee spend, which means channel 3 is underperforming.
About the Author
Harshit works as a Data Scientist at OLX where he uses data science for the organisation’s internal product to improve the customer experience on the platform and increase revenue for the company.
Check out about his journey as a Data Scientist where he also answers most asked questions about the Data Science domain: https://www.linkedin.com/posts/activity-6774233149258903552-0GH9
