Learn how you can apply data science and big data analytics methodologies to digital marketing to improve marketing performance, predict campaign success, customer churn, and content popularity.
19. https://flic.kr/p/iDXwPx
Customer segmentation & journey analysis
Marketing campaign performance
Content marketing optimization
Predictive attribution
Churn & loyalty
What are some of the key use cases…
Hello everyone I am Sameer khan I am big data evangelist and I lead IBM customer analystics. I have has opportunity to lead many digital marketing teams in my past life so today I am going to talk on how to bridge the gap between data science and digital marketing.
I started my marketing career as a campaign manager and I had no insight on anything other than digital analytics. Then my friend who was Hadoop developer introduced me to the world of data science and from there is no looking back. In this session I will show you how you can cross the chasm and use data science for digital marketing success.
When I talk to people who are not associated with data science or big data industry they think data science and big data analytics are two different industry segments competing with each other. One is iron man with all the tech wiz and other has the power of Gods. This is far from truth. Data science is practice a methodology that can be applied to any type of data.
Where as big data analytics is focused only on big data.
For example- in my past life I was involved in a customer churn reduction project. Initially we were building a model with limited data (12 months) and we were 60% accurate in predicting churn for next quarter. We decided to use all the data available and improved our prediction to 83% accuracy. A significant jump in accuracy. Why did this happened? It was mainly because we increased the volume of data. So it's proven that by simple taking all data you have you can increase accuracy and infact you can find new insights that you didn't has a clue existed.
Just like iron man uses Arc reactor to power his suit a marketer needs sophisticated tools to run advanced analytics. Big data tool space is massive and is filled with tools of all shapes and size. For the purpose of this session I use three simple criteria 1. The tool has to be free or atleast have a free tier 2. It has to have a wider acceptance and 3. it should be easy to use.
Raise your hand if you know the popular game show called Jeopardy. Great. So IBM originally created a super computer called Watson that was able to beat the leading Jeopardy scorers. Watsonanalytics.com is built on some of the text analytics capabilities of jeopardy combined with predictive analytics. You can create a free account.
The second one is R studio. You can either download it on your on your mac or PC or run it on amazon elastic compute engine which is Amazons big data analysis infrastructure. You definitely have to learn R programming and there are tons of resources on Edx or Coursera to help you learn R in few months.
The last one is Bigml which is another free predictive analysis tool which is widely used, has free tier and a very popular.
Now let's move on and discuss how digital marketing can further benefit from advanced analytics and data science methodologies. When we think of marketing analytics we generally refer to digital, social or marketing automation analytics. I think this is great although time has come for marketers to up their game and take a non traditional approah to data analytics. In the rest of the post we will discuss ten different ways on how you can take advantage of data science. The best part is we will be using completely free tools to perform the analysis.
Marketers generally think about the data explosion as we think about analysis but we are also experiencing marketing technology explosion. Raise your hand if you have seen this chart. Its from a website called Chiefmartech and it has logos from all types of marketing technology offering. Infact, Gartner did a survey on the usage of data across industries' sand found that Business cannot consume 90% of collected data due to organizational silos.
Why should we combine your basic digital marketing analytics with big data analytics or data science? First of all like they say in India its very tasty. Secondly you can solve some really big problems that you cannot by basic analysis.
Finally, when it comes to digital marketing its all about timing. If you are not able to deliver you message on time then you end up in a situation like this.
There is a very popular India snack mix called its called Chivda and there are more than a thousand possible combinations of this snack mix available in market today. Each combination has a very unique tastes and all are very tasty. Digital marketing also has different types of data sources and each combination can be use to drive different campaigns. When you combine it with big data it becomes really interesting and highly ROI centric.
So lets talk about where you can see the biggest benefits from the applications of big data and digital marketing combination.
Customer segmentation and journey analysis: cluster analysis can allow you to slice your customer data for laser targeted retargeting
Marketing performance optimization: maximizing marketing performance because now you have better customer visibility and improved
So you don’t end up target the customers with the same product they have already purchased from you.
Content marketing: this is my favorite that's why it's number one. Raise your hand if you have struggled to get your content asset to go viral?trust me you are are not alone. With significant growth in content production across all fronts it is becoming harder and harder to come up with a winning piece or for that a series of winning content pieces. I have had similar experience but I promise you by the time I am done you will have a proven method of predicting content popularity.
Predictive attribution - we all love marketing attribution don't we? With a little bit of regression analysis we can start predicting outcomes without the need of expensive tools.
(Funny) Churn prediction: churn prediction is another good use case as I mentioned earlier where you can use your past churn and existing customer data.
Fiat combined Digital Marketing with statistical analysis and improved marketing performance by 20% and customer loyalty by 7%.
So Since we don't have time to show all of these use cases i would like to show how to predict content popularity in 6 steps.
Step1: Prepare data: in order to predict something we need to have historical data. To predict which content will be mots popular you need historical content data. There are multiple free tools in the market to get content marketing data.
Buzzsumo is my favorite. You can get a free 14 day trial of the pro account. I simply scrapped all the articles on the topic of big data analytics 30 days.
Or you can use Google chrome scraper plugin to scrape all the content.
Then I structured the data into columns (headline, title, category, abstract, wordcount) with proper labels and the last column is whether or not the content went viral (1) and not viral (0). I used the criteria of shares > 1000.
Step2: Create a free account at IBM Watsonanalytics: it's easy it's free and it's open to anyone.
Step3: Uploading your data on IBM WatsonAnalytics.com and analyzing your dataset.
Load the csv file into Watsonanalytics.com.
Since Watson is capable of handling plan English langue you can ask questions such as “what is the most popular content publication data”
and Watson will give you answers in tables and charts.
Step4: Next we select your prediction criteria. Since our focus is to predict the popularity of content the column “popular” will be our taget.
Once you select the prediction criteria you will see the prediction screen with multiple charts. It shows which variables or combination of variables drive the popularity of content. You can toggle between single and multiple dependent variables.
Then we move on the doing some detailed analysis if our prediction model.
You can drill down to individual prediction chart. Each chart has a prediction strength for example the one highlighted shows newsdesk and sectionname drives content popularity by 90% which is pretty significant.
This will help you in identifying key patterns. In this table we see a blog can go viral if the wordcount is higher than 829 and targets health and crosswords games category.
You can also look at the decision trees if you like charts better than tables.
Finally step 6 is to take action. You now have the power of predicting the success of your next blog article.
Put the data to action and create content pieces like these.
Lets talk. If you have any questions I can be reached on Twitter or through my website datacrackle.com.