The good news about the Big Data market is that we generally all agree on the definition of Big Data, which has come to be known as data that has volume, velocity and variety where businesses need to collect, store, manage and analyze in order to derive business value or otherwise known as the "4 V's." However, the problem with such a broad definition is that it can mean different things to different people once you start to put some real values next to those V's.

Let's be honest, Volume can be a different thing to different organizations. To some it is anything above 10 terabytes of managed data in their BI environment and to others it is petabyte scale and nothing less. Likewise velocity can be multi-billions of daily records coming into the enterprise from various external and internal networks. When it really comes down to it, each business situation will be quite different not only from a size and speed perspective but also more important from the business use-case or requirement. A large bank's Big Data problem could be very different to that of an online retailer or an airline. If you compare what say a hospital is trying to do collecting and analyzing all the sensor patient data compared to a utilities provider running a smart-grid or a telecommunications operator. True, all could be categorized as machine generated or raw data but the exact type of data might be different not to mention the volume or growth rate. Probably the one unique common denominator across all aforementioned industries is that everyone is keeping the data for longer time-periods. No one is throwing it away - not even the detailed data.

The Many Cost Factors to Consider
Costs will of course vary depending on the individual allocated IT budget but regardless, how the company allocates IT budget dollars to new Big Data initiatives needs consideration. Let's face it, enterprise buyers didn't suddenly come into a bunch of newfound IT assets or line items on their budget and the current world economic situation would certainly not suggest so. More likely existing budgets are being re-allocated and instead of spending more on say existing traditional data warehouses or appliances, monies are being allocated to new projects running on open source projects including Apache Hadoop which promises both low cost, ease of scale not to mention the obvious best approach to managing and analyzing multi-structured data sets. The difficultly then arises how do you integrate or have your Hadoop environment co-exist with the established BI or DW environment that the business has grown to love and rely upon?

Leverage What You Already Have
Let's assume you have a data warehouse or data mart in place today and you already use various ETL or data movement tools and BI dashboard, analytics or reporting tools and you don't want to disrupt business users which could not only impacting performance levels but also training up on a new set of tools. In fact you already likely beholden to strict SLA's around response times for the various business reports and KPI's. However, at the same time the business is demanding access to new data sets in order to glean better insights either directly analyzing this data or co-mingling it with existing customer data. This could take the form of web-logs, click stream data or social media data from various interactive sites the business is now leveraging and tracking. The promise of impacting profit margins and gaining a competitive edge just cannot be avoided.

As we all know, traditional relational or columnar databases can't handle the unstructured data types so IT needs to rollout a different solution to satisfy the business demands. Evaluations can take many forms but typically will start with which Hadoop distribution, which NoSQL or NewSQL database and what query access tools in addition to MapReduce. It is certainly no easy task as there are a large number of technology solutions on the market today that claim to run on or with Hadoop providing MapReduce or SQL-like capabilities which all satisfy the requirement of managing volumes of unstructured data. Some are more mature than others; some proven and not all are low-cost. Open source on the surface looks very low cost but as soon as you require any level of support, which lets face it once it's live and relied upon as a business critical environment, you will need to allocate a line item on your budget. The Big Data line item won't just be one line as it will need to include all components required to properly rollout a Big Data solution to truly satisfy the business demands. Just like any other IT environment the obvious pieces will include: Software licensing and support, hardware, skilled dedicated resources, professional services and training and the dedicated time of business users to provide input on key requirements including specifying types of reports, queries and analysis which will naturally change and evolve over time.

Big Data Costs Can Quickly Creep Up
In terms of the hardware expenditure required to manage the new Big Data set, you may start out with a Hadoop cluster of say 10 nodes and yes that is certainly manageable but if your data velocity is significant, you can quickly reach 100+ nodes and now you will face a number of other expenses including additional headcount and skilled resources to manage the environment proactively in addition to tools for managing the cluster including system management and alerting and potentially add-on software which can vary by business use-case but might cover real-time analytics against streaming data for say fraud detection or detection of unusual patterns. You may also need a business tool to provide a front-end GUI dashboard to track specific KPIs or data visualization tools so business users can quickly understand what is going on. Very quickly the costs become less about the storage and hardware and more around the software that focuses on getting the most value from this newly collected data set.

There is no denying the fact that Big Data presents great new opportunities but reaching the point of a quantifiable ROI in a fast time frame is still a very real challenge. Everyone is talking about Big Data and all the innovative technology approaches to tackling it but it is still difficult to find lots of business success stories within any one-industry sector. It's still fairly immature but the good news is that its moving at a much faster pace than any other IT project today and certainly our data warehouse and BI forefathers have provided lessons learned over the past two decades.

Big Data Is Big Business but It Comes with Strict Requirements
If we want to examine more closely the main areas of expenditure for a Big Data project, it is probably best to look at it through the lens of a specific type of business and use-case. Let's take a large financial institution that has a number of existing traditional data warehouse / BI environments but because the business doesn't want to throw any data away (well let's face it regulations don't allow that for a number of years) and realistically the business wants to retain specific data sets for ongoing trending and analysis. This includes examining questions such as "what constitutes a low-risk client based on spending behavior patterns over a specific time period cross-referenced with customer demographics" which will help the institution better target a particular segment of the market.

Given the IT budget doesn't allow for increased spend that correlates with data growth rates, they need to seriously reduce costs and so decide to go the route of a Hadoop-based environment given its promise for low-cost scale and the fact that it can provide insights into customer patterns by capturing semi- and unstructured data. Front-ending the warehouse with a dedicated Hadoop cluster is the preferred architectural approach but the business users still want access to both the Hadoop environment and the existing traditional data warehouse environment.

Given we are talking about a financial institution, the question of security and availability quickly come to the top of the requirements list. At the same time, if business users want to access that data, SQL query access and using the current BI tool against that new set of data is also a requirement. If you can avoid having to the move large chunks of data on a frequent basis from one to the other, it will not only reduce costs but also latency. In an ideal world, being able to leverage the skill sets you already have and avoiding duplication of work is key.

Below is a quick table outlining the main cost factors to be considered and a set of comments against each of these areas that could reduce costs.

Summary: Consider All Factors and Get Business Buy-in Quickly
Big Data is fundamentally a business problem. If you begin with the question of "what is the business trying to achieve by collecting, storing and analyzing this new set of data...", you will start down the right path to realizing business gains. Whether you outsource the initiative or bring in external consultants and vendors to manage the project, the same questions will arise and in order to leverage what you already have which includes both existing IT environments and skills, you will be better able to contain costs.

Furthermore, we all love the promise of new innovative technologies including Hadoop and MapReduce but without leveraging tried and tested standards we have come to love and respect, it doesn't make a whole lot of sense from both a technical or economic sense. As you start on your Big Data journey or project, be sure to ask what exactly the business requires and how can you leverage what you already have today. We all know, getting business user buy-in and success is half the battle to a successful rollout.