Some time ago i was working on an idea called as Ad Attributes or Advertisement attributes. I’d like to share my thoughts on this idea with audience.
Advertisement attributes are for creating a favorable selling climate.
Today consumers are constantly targeted with product information by marketing companies. Consumers are faced with numerous advertisements with vast information on products. Thus consumers use the heuristics approach to help them in making their purchasing decisions. This approach is basically using mental shortcuts to streamline the selection process cognitively. This is to avoid being puzzled or paralyzed by the huge number of products offered in the market by numerous companies.
Advertisement Attribution is focus on the relationship between consumer involvement and hedonic and utilitarian ad attributes, and on how these are linked to overall advertising effectiveness.
- Attribution can correctly model ad sequencing and can provide recommendations to business.
- Attribution analysis builds a picture of how advertising campaigns deliver results and provides a platform for continuous improvement.
- Use the predictive analytics capability to predict the outcome of digital campaigns.
- Make an understanding of the dynamics of digital advertising in understanding how effective it can be in marketing mix.
- Run the campaign to various channels and achieve varying levels of success through models, recommendations and with a solid analytical data model.
Below will take us to data analysis approach in “Ad Attribution” with a sequence of questions.
Stating and Refining the Question:-
1. Descriptive Question:-
What are the true worth of channels effecting to any company’s success and understand the value of digital advertising and its attributes.
a. What an advertiser wants to know is how digital advertising really affects the bottom line.
b. How do you know what works and what does not?
c. Do combinations of advertising work best?
d. How does advertising work with search and social media?
e. What external factors are effecting the campaigns poor or good performance?
3. Inferential Questions when we have data source and list of data source are below:-
- Digital Adservers such as DoubleClick(tagging technique to track ad impressions, clicks and site activity)
This gives us information two kind of sale:-
Simple Online Sale
- Impressions served.
Complex Online Sale:-
- Impressions served.
- Cost per 1000 impressions
2. TV logs:- There is no definitive proof that an ad was effective, but with digital ads there is a digital trail that can provide some insight.
3. Print Media log:- There is no definitive proof that an ad was effective, but with digital ads there is a digital trail that can provide some insight.
4. Offline sale log:-
- Sale at automobile store.
- Surveys data filled by customer
- Feedback data
5. Search log(Google/Bing):-To avoid unwanted popups and security factors user generally go to search engine to get information about the product he/she wants.
6. Social media (touchpoints) logs, video logs: – Analyze third party cookies.
7. Ad location
4. Predictive Questions:-
In Predictive question we are not interested in outcome conversion that happened online. But why this conversion is happening but not others, what region/area searches and why, why conversion is more at one particular time, why conversion fluctuating at what events.
This can lead us to another data source where let say conversion happen in particular income group/ age group.
Predictive questions also contains the search results to search engine effecting to what sale in market.
This question again lead us to requirement of another datasource i.e. companies performance or sales figures.
Exploratory Data Analysis
Let us say we have data samples are ready to run our analysis. Now it’s important to examine the structure and components of our dataset.
- With Search engine results we come to know that at particular regions users are searching for a specific disease and its symptoms. In some cases we have ample time before patient reaches to doctor. This weeks or months of time we can Influence Pharmaceuticals decisions by reaching out to Doctors and Health Care Providers. There are many decision we can control like price of a medicine at the particular demographic area which would be a direct profit without sale or introduce an alternative medicine to bring an entry to a new business.
- Find relation between advertisement and activities not only campaign and conversions.
- Calculate Click Through Rate (CTR), it is simply the number of clicks on an ad divided by the number of impressions served. Calculate Fraud clicks.
- Calculate Last click only counts clicks immediately before a conversion event, which eliminates credit for accidental and repeat clicks.
- Visualize our data in graphical representation.
Using Models to Explore Your Data
At this stage, a statistical model needs to build to provide a description of how the world works and how the data were generated. The model is essentially an expectation of the relationships between various factors in the real world and in our dataset.
Comparing Model Expectations to Reality
- Describe the sampling process
- Describe a model for the population (populations is subset of our data)
Drawing a fake picture: – To begin with we can make some pictures, like a histogram of the data.
Reacting to Data: Refining Our Expectations
Let’s say model and the data don’t match very well, as was indicated by the histogram above. So what to do? Well, we can either
- Get a different model
- Get different data
Interpreting Our Results and Communication
Communication is fundamental to good data analysis. We gather data by communicating our results and the responses we receive from our audience should inform the next steps in our data analysis. The types of responses we receive include not only answers to specific questions, but also commentary and questions our audience has in response to our reports.
Most weighting schemes attempt to look at the activities of individual cookies as they progress towards a conversion (or not). Weighting schemes assign values to certain events and add them up to provide a relative measure of that item’s value. A popular weighting scheme is an extension of “last view” and is known as “U” weighting. For example each cookie that converts, it assigns a weight to the first ad that the viewer sees, a larger weight to the last ad seen and lesser weights to the ones in between. This seems like a very rational approach, and it tends to deliver answers we expect. Unfortunately, the results usually provide very little real insight because there is no analytical basis for choosing the value of the weights or even what should be weighted. Different types of purchases have very different decision patterns, and a single weighting scheme cannot possibly be right for all.
Therefore we have different weightage schema:-
Difference between positive and negative correlations.
What is the best channel to invest into the advertisement industry to get high ROI.
Regression attribution uses an algorithm to model the behavior of market performance in our data modeling. Any number of aspects may be used in the model. The advantage of this approach is that it actually attempts to determine the weights or importance of events analytically rather than by subjective weights assigned by the user. The drawback is that there are numerous dimensions to be considered and decisions on which ones to use tend to be subjective.
Both weighted and regression methods look at the activity of each individual data source and relationship — and there can be many to account for. This often leads to sampling, which may pass statistical significance measures but can miss the important key events.