About analytics platforms
When we speak of analytics, we speak of a process that describes the data fed to it, or even makes predictions after analyzing the data, in either case providing useful business insight. So, when we think about ‘analytics’, the first thing that comes to our mind is a powerful set of algorithms, possibly using data science, that will process the data and create useful ‘models’. Although the algorithms indeed form the ‘heart’ of analytics, a practical analytics solution requires many other critical components before the users can use it in production in a practical way.
To begin with, we need a robust way to input data for analysis, either through uploads, or through components that directly read the business data or data lake.
The raw data from source systems usually need to be classified into numeric (or continuous values), categorical (some specified distinct values), free form text, or dates, and such like. Related fields (amounts, for example) need to be converted to a common order of magnitude or unit. More sophisticated solutions will be able to point out data outliers, or have some treatment for sparse or missing data. All of these are referred to in the analytics language as data preparation.
The data thus prepared or engineered will form the input to core analysis engine or algorithm. This is the workhorse of the solution, and runs a set of defined, or customized analytics algorithms to create the desired business output from the data. This can describe the historical data, revealing previously unknown truths or details. Alternatively, the analysis might diagnose the root causes of the business behavior. Or, predict future behavior based on the past, using data science tools. Or even, prescribe what needs to be done next.
Finally, the output from the core analysis needs to be presented in an insightful way to the user, preferably using graphical visualization. The output data might also need to be extracted and fed into another system.
Therefore, a complete analytics project requires a good deal of software to be built around the core algorithms to support the above workflow from beginning to end. An organization’s data analyst or IT would spend a great deal of effort in basic software engineering for each significant analysis project, clearly wasteful and unsustainable in most business organizations.
Good analytics platforms help solve this issue by providing tools that support each of these steps, so that your analysts focus on what remains the fundamental job: analysis.
Business problems are many, as are the varieties in the business data in organizations. The sophistication of an analytics platform lies in the robustness of its software components, and the flexibility it offers in handling this variety. Each leading platform available today does some parts better than it handles the other parts of the full workflow.
So, do you really need an analytics platform? If you do, then which platforms are good for you? How should you go about choosing? Our next week’s story tries to de-mystify the complex and bewildering world of analytics platforms.
Watch this space.?