Hello, everybody! We hope this article finds you well.
Have you ever wondered whether the money you put in advertising actually leads to an increase in sales? Or which is the advertising channel that leads to the highest returns so you can put more money there? After all, there are many different channels through which an organization can promote itself (TV, brochures, newspapers, social media, etc.) and advertising and promotional effort are cornerstones of success. Therefore, the allocation of advertising budget should not be viewed as a trivial task.
In today’s part of our Data Science Series we will be delving into the challenging realm of advertising budget distribution. We will leverage different set of tools, dive into historical sales data, find important dependencies, estimate the contribution of each advertising channel and allocate a given advertising budget in a way that maximizes return on investment (ROI) and as a consequence – profit.
The analytical process is an iterative one as shown in the picture on top. To achieve our goal, we go through several different stages. Let’s review them in more detail:
1) Business Case/ Pain Point
Signing off advertising budgets, without knowing what the end result of the campaigns will be, is a difficult thing to justify. Adopting data-driven approach when deciding on the distribution of the advertising budget can lead to significantly better end result.
2) Data Requirements
There are several data requirements which should be met when constructing such budget allocation tool. First, data regarding the end result (outcome) of advertising should be available – sales, revenue, etc. Also, data on the advertising effort in each channel is needed. It can be GRPs (gross rating points), advertising expenditure or any other measurement of advertising effort. Other data, which may be deemed important, is also welcomed – industry specific data, macroeconomic data, etc.
3) Data Exploration
As we have already said in previous parts of our Data Science Series (e.g. Part 3 & Part 4), this stage of the analytical solution should be approached with great attention. On this stage, one should familiarize himself/herself with the data for the task that lies ahead. Look at the historical sales, identify past trends, investigate historical spending on advertising as shown on the picture. This is also the time when expectations and hypothesis about the data should be formulated so that they can be put to test later on.
4) Measure Channel Contribution
If we are to discover which channel yields the best results, we need to first break down sales into baseline and incremental sales. The baseline shows the sales that can be expected had no advertising and promotional effort been applied. Incremental sales are generated as a result of the advertising and promotional effort. They can be split further, so that the individual impact of each channel is estimated.
Modelling advertising mix is a tricky business and there are many possible approaches to tackle it. There are different competing strategies and methods and there still isn’t a consensus among researchers in this field.
Our approach is the following:
- To better capture the prolonged effect of advertising on consumer behavior we use adstock modelling to transform the advertising channels.
- We rely both on statistical analysis and domain knowledge to choose the best parameters for the adstock transformations.
- We then apply multivariate regression analysis to find the contributions of each advertising channel and estimate the baseline sales.
- With the newly acquired insights, we are able to determine the ROI of each channel given the current spending.
The previous picture shows the contribution of each advertising channel and the baseline sales over the studied period.
Adstock transformation also allows us to account for the fact that advertising can have diminishing returns to scale and saturation level. This means that with the increase of advertising effort, marginal returns do not linearly increase to infinity as well, but they are rather increasing with a diminishing magnitude and after a certain point (saturation level), they start to approach zero. Using the previously estimated parameters we can create a chart (as the one shown on the right) showing the impact of each advertising channel at different level of advertising activity.
5) Budget Allocation
We have almost reached the end of our journey. All of the necessary ingredients to formulate an optimal allocation of the advertising budget are available.
Our budget allocating tool works by looking into the marginal return at different level of spending and selecting the channel which yields the highest return. This process is repeated until the budget is exhausted. This framework also allows for the specification of different constraints on the advertising channels if they are deemed relevant enough.
Having estimated the optimal spending in each advertising channel, a comparison with the current expenditure levels can be made, such as the one on the chart, and also, the expected increase in sales can be calculated.
Looking back, we can now summarize the work done:
- We looked at the data, identified dependencies, trends, patterns, formulated hypotheses.
- Then we started with the advertising mix modelling by carrying out an adstock transformation of the channels.
- We used multivariate regression analysis to estimate the impact of each channel.
- At the end, a budget allocation tool was created to allocate advertising budget with the goal of maximizing the sales revenue.
We are more than enthusiastic to have shared our Advertising Budget Optimization Tool approach with you. That being said, we know other alternative approaches to this topic exist and will be more than happy to hear about them, their pros and cons and your experience with the topic.
We do hope that this topic was of interest to you and that we have captured your attention.
With this series our aim is to increase data science coverage and to make data-driven decisions an integral part of more companies around the Globe.