Analytics Out of the Box: Towards optimized scale, speed and skill
By : Dr. Prateek Kapadia
Chief Technology Officer
An analytics practice uses data to deliver knowledge – as actionable insights, recommendations, smart data visualisations and partially or fully automated actions so as to deliver optimum business impact. The analytics practice is typically constituted by a specially nominated cross-functional (internal or out-sourced) team with the required core competencies – data sciences, decision sciences, data engineering and business analysis – within a business organization. This article offers the author’s views on how modern analytics practices can efficiently scale up their knowledge delivery with the use of appropriate technology.
The supply-demand imbalance
The gap in the demand and supply for relevant skills in analytics practice is a major challenge. There is a perpetual and ever growing need for data scientists. The academic institutions are rapidly coming up with new courses in the field to create adequate supply. The parody “how many engineers does it take to fit a light bulb?” is an apt metaphor for this context. The answer may be simple if we take it in the literate sense, but the same could lead to different answers if we think beyond it and interpret the problem in different ways. If we take the viewpoint that analytics practices are light bulbs and data scientists are engineers, the following statements are plausible:
- The time-and-material effort as well as the expected quality standards for fitting each (analytics practice) light bulb is significant enough to require more than a few (data scientists) engineers;
- Fitting a light bulb is a complex operation that requires multi-specialised and multi-skilled engineers;
- There are too many light bulbs to fit but the engineers are too busy, hence there is always a waiting time as the availability of engineer is a challenge.
The functioning of an analytics practice is often misunderstood in the same way –the number of analytics outcomes that can be delivered by the practice is mostly perceived to be proportional to the number of data scientists it employs!
The quest for efficiency
The acquisition of right skills by personnel augmentation is necessary for analytics practice to function. However, analytics practice also need to focus on attaining the desired efficiency and scale in their processes so that they deliver more impactful knowledge and respond more dynamically and faster to the demands of their parent businesses. A common way to increase the efficiency of any delivery pipeline is to optimize production resources by maximizing product reusability and simplifying customizations of the delivered product. These efficiency goals are best specified for the analytics practice by orienting its output as ‘analytics product’ to connote packaging for ready consumption and ease of use. There is a deliberate attempt here not to call it an ‘analytical model’. It can be much more than that. Sometimes multiple models can be packaged along with the execution environment, necessary validations and a delivery/consumption mechanism put together in a box as an ‘analytics product’.
This ‘analytics product’ philosophy resonates in our title Analytics Out-of-the-Box – a term chosen to describe packaging and delivery mechanisms. (The pun “out of the box” thinking is deliberate.) An analytics product must have the following characteristics:
- Ease of use: easily configurable user/application interfaces for analytics consumers (humans and applications);
- Repeatability: the ability to be instantiated and reused efficiently across multiple use cases. In addition to meeting these goals, analytics practices also need to increase the efficiency of analytics product lifecycles – from conception to delivery, support and optimisation.
The making of ‘analytics product’
Churning out well-defined analytics products on an efficient lifecycle requires the production line to have the following characteristics:
- Packaging appropriate data connectors and data transformation logic
- Ability to develop and package multiple analytical models into one analytics product;
- A common execution environment for the iterative testing, deployment and product usage – this avoids repeated retro-fitting of products to different execution environments;
- An ability to continuously deliver multiple analytics products for one set of consumers;
- A feedback mechanism to feed in performance data to optimise the analytics product and
- Sufficiently abstracted controls for in-service configuration and performance management of analytics products in use.
Analytics practices would do well to choose one software platform/product that allows these functions in their analytics product lifecycles.
The traditional and modern Business Intelligence vendors also talk about ‘prepackaged analytics applications’, however this is too limited in its scope. They typically envisage providing a standard set of prepackaged dashboards and reports for helping solve standardized business problems. The ‘analytics product’ has to go beyond this, it needs to have flexibility ingrained to account for different data sources, data models, analytics techniques beyond statistical ones, types of users, output delivery mechanism, consumption resource and so on. The need is for ‘analytics products’ that can help the analytics practice to bring in speed and efficiency for solving even the most complex business problems like churn. Going forward the ‘analytics product’ is also expected to self-optimize with feedback loop also incorporated within its design.
Analytics practices implementing these recommendations will see significant scale in efficiency and quality of delivery, reduction in friction with their consumers, and avoidance of people dependability.