Unlearn to Unleash Your Data Lake

16 Sep

The Data Science Process is about exploring, experimenting, and testing new data sources and analytic tools quickly.

The Challenge of Unlearning
For the first two decades of my career, I worked to perfect the art of data warehousing. I was fortunate to be at Metaphor Computers in the 1980’s where we refined the art of dimensional modeling and star schemas. I had many years working to perfect my star schema and dimensional modeling skills with data warehouse luminaries like Ralph Kimball, Margy Ross, Warren Thornthwaite, and Bob Becker. It became engrained in every customer conversation; I’d built a star schema and the conformed dimensions in my head as the client explained their data analysis requirements.

Then Yahoo happened to me and soon everything that I held as absolute truth was turned upside down. I was thrown into a brave new world of analytics based upon petabytes of semi-structured and unstructured data, hundreds of millions of customers with 70 to 80 dimensions and hundreds of metrics, and the need to make campaign decisions in fractions of a second. There was no way that my batch “slice and dice” business intelligence and highly structured data warehouse approach was going to work in this brave new world of real-time, predictive and prescriptive analytics.

I struggled to unlearn engrained data warehousing concepts in order to embrace this new real-time, predictive and prescriptive world. And this is one of the biggest challenge facing IT leaders today – how to unlearn what they’ve held as gospel and embrace what is new and different. And nowhere do I see that challenge more evident then when I’m discussing Data Science and the Data Lake.

Embracing The “Art of Failure” and The Data Science Process
Nowadays, Chief Information Officers (CIOs) are being asked to lead the digital transformation from a batch world that uses data and analytics to monitor the business to a real-time world that exploits internal and external, structured and unstructured data, to predict what is likely to happen and prescribe recommendations. To power this transition, CIO’s must embrace a new approach for deriving customer, product, and operational insights – the Data Science Process (see Figure 2).

Figure 2:  Data Science Engagement Process

The Data Science Process is about exploring, experimenting, and testing new data sources and analytic tools quickly, failing fast but learning faster. The Data Science process requires business leaders to get comfortable with “good enough” and failing enough times before one becomes comfortable with the analytic results. Predictions are not a perfect world with 100% accuracy. As Yogi Berra famously stated:

“It’s tough to make predictions, especially about the future.”

This highly iterative, fail-fast-but-learn-faster process is the heart of digital transformation – to uncover new customer, product, and operational insights that can optimize key business and operational processes, mitigate regulatory and compliance risks, uncover new revenue streams and create a more compelling, more prescriptive customer engagement. And the platform that is enabling digital transformation is the Data Lake.

The Power of the Data Lake
The data lake exploits the economics of big data; coupling commodity, low-cost servers and storage with open source tools and technologies, is 50x to 100x cheaper to store, manage and analyze data then using traditional, proprietary data warehousing technologies. However, it’s not just cost that makes the data lake a more compelling platform than the data warehouse. The data lake also provides a new way to power the business, based upon new data and analytics capabilities, agility, speed, and flexibility (see Table 1).

Data Warehouse Data Lake
Data structured in heavily-engineered structured dimensional schemas Data structured as-is (structured, semi-structured, and unstructured formats)
Heavily-engineered, pre-processed data ingestion Rapid as-is data ingestion
Generates retrospective reports from historical, operational data sources Generates predictions and prescriptions from a wide variety of internal and external data sources
100% accurate results of past events and performance “Good enough” predictions of future events and performance
Schema-on-load to support the historical reporting on what the business did Schema-on-query to support the rapid data exploration and hypothesis testing
Extremely difficult to ingest and explore new data sources (measured in weeks or months) Easy and fast to ingest and explore new data sources (measured in hours or days)
Monolithic design and implementation (water fall) Natively parallel scale out design and implementation (scrum)
Expensive and proprietary Cheap and open source
Widespread data proliferation (data warehouses and data marts) Single managed source of organizational data
Rigid; hard to change Agile; relatively ease to change

Table 1:  Data Warehouse versus Data Lake

The data lake supports the unique requirements of the data science team to:

  • Rapidly explore and vet new structured and unstructured data sources
  • Experiment with new analytics algorithms and techniques
  • Quantify cause and effect
  • Measure goodness of fit

The data science team needs to be able perform this cycle in hours or days, not weeks or months. The data warehouse cannot support these data science requirements. The data warehouse cannot rapidly exploration the internal and external structured and unstructured data sources. The data warehouse cannot leverage the growing field of deep learning/machine learning/artificial intelligence tools to quantify cause-and-effect. Thinking that the data lake is “cold storage for our data warehouse” – as one data warehouse expert told me – misses the bigger opportunity. That’s yesterday’s “triangle offense” thinking. The world has changed, and just like how the game of basketball is being changed by the “economics of the 3-point shot,” business models are being changed by the “economics of big data.”

But a data lake is more than just a technology stack. To truly exploit the economic potential of the organization’s data, the data lake must come with data management services covering data accuracy, quality, security, completeness and governance. See “Data Lake Plumbers: Operationalizing the Data Lake” for more details (see Figure 3).

Figure 3:  Components of a Data Lake

If the data lake is only going to be used another data repository, then go ahead and toss your data into your unmanageable gaggle of data warehouses and data marts.

BUT if you are looking to exploit the unique characteristics of data and analytics –assets that never deplete, never wear out and can be used across an infinite number of use cases at zero marginal cost – then the data lake is your “collaborative value creation” platform. The data lake becomes that platform that supports the capture, refinement, protection and re-use of your data and analytic assets across the organization.

But one must be ready to unlearn what they held as the gospel truth with respect to data and analytics; to be ready to throw away what they have mastered to embrace new concepts, technologies, and approaches. It’s challenging, but the economics of big data are too compelling to ignore. In the end, the transition will be enlightening and rewarding. I know, because I have made that journey.

Source: http://cloudcomputing.sys-con.com/node/4157284


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: