Big data provides opportunities for businesses to gain insights from large, diverse datasets. Traditional business intelligence addressed some questions, but big data and advanced analytics can answer all key questions. This includes predicting future trends and determining the best actions for improvement. Opportunities exist in using big data to personalize offers, predict customer behavior, and optimize operations. However, many companies face challenges around data management, analytics skills, and making the technologies user-friendly for business users.
2. What is Big Data?
• “Big data," is a group of data technologies that
are making the storage, manipulation and
analysis of large volumes of data cheaper and
faster than ever.
• Types of “Big data”
– Transactional Data
– Data from mobile app
• Location data , Profiles
– Data from Social media
• Blogs, Facebook, Twitter and other social media apps
2
3. Big Data Challenge
• Managing the three “V”s of big data
– Volume
– Velocity
• The speed at which data is coming and changing
– Variety
• Text, Audio, Video
• Big Data is mainly unstructured data
• Technology to store big data
• Technology to analyze big data
3
4. The Business Needs
• Traditionally business wanted answers to Five
Questions
• Traditional BI answers two of those questions
– What Happened? – Reports and Ad-hoc Queries
– Why it Happened? – Analytics, Cubes
• Dash Boards and Score Cards Answer the third
– What is happening Now?
• Data Mining and Predictive Analytics Answer the last
two
– What is going to Happen in Future? – Data Mining
– What can I do to stop it or make it better in future? –
Predictive Analytics
4
5. Big Data Opportunity
• The relational databases has limitations
– Data needs to be modeled
– Need to know the business needs to create good
data models
– Data needs to be structured to support queries
• Can we do analytics on big data and answer all
Five business questions?
5
12. Big Data Opportunities
• McKinsey projects that in the U.S. alone, there will be a need by
2018 for 140,000 to 190,000 “data scientists”
• Steep technical learning curves and a lack of qualified technical staff
create barriers to adoption
12
13. Big Data Opportunities
• Need for another 1.5 million data-literate
managers
– Formal training in predictive analytics and statistics.
• The technologies in the big data area are not
Analyst Friendly
– Need Programmers with knowledge of
Hadoop, Statistics and analytics
• Companies Retraining programmers and database analysts
to get them up to speed on advanced analytics.
• Getting started with Hadoop doesn't require a large
investment as the software is open source, and is available
instantly through the Amazon Web Services cloud (Elastic
MapReduce service)
13
17. DMA Campaign Response Rates 2010
• Email to a house list averaged a 19.47% open rate, a 6.64% click-through
rate, and a 1.73% conversion rate, with a bounce-back rate of 3.72% and an
unsubscribe rate of 0.77%.
• Direct mail: Letter-sized envelopes had a response rate this year of 3.42% for a
house list and 1.38% for a prospect list.
• Catalogs had the lowest cost per order of $47.61, just ahead of inserts at
$47.69, email at $53.85, and postcards $75.32.
• Outbound telemarketing to prospects had the highest cost per order of
$309.25, but it also had the highest response rate from prospects of 6.16%.
• Paid search had an average cost per click of $3.79, with a 3.81% conversion
rate. The conversion rate (after click) of Internet display advertisements was
slightly higher at 4.43%.
17
20. Improving Offer Acceptance Rate: Algorithms to Personalize
Offers
• K-Means Clustering for clustering Users
– Cluster users based on brand preferences and
demographics
– Most popular Clustering Algorithm
• Logistic regression for finding the probability of
accepting an offer
• SVD (Single Value Decomposition) to reduce
dimensionality of data and to reduce noise
– Reducing the dimensions to a few improves
performance and reduce accuracy
– The noise reduction which happens when the
dimensions are reduce helps to improve the
accuracy of prediction
20
22. How Does The Model Work?
– Classification Algorithms learns from Examples in a process known as Training
– Need Training Data and Decide on Training Algorithm
• Choose between Logistic Regression and Google’s combined regression and ranking
– Need to specify the input values (Predictors) and output values (Target) in the
training data
• Predicting Clicks probability is the Target variable
• User and Item features are the input variables
22
23. Choosing Products for customer and Ordering
Customer
Details
Click Prediction
Sale Items Model for Product
Items Display
Chosen Order
23
24. Conclusion
• On the basis of our on-line surveys, face-to-
face survey and analysis of studies done by
others we conclude that the opportunity for a
Marketing application based on Big data and
Machine Learning is great. In a scale of 1-10
we rate this opportunity at 9
24