The starting point in a big analytics strategy has to be business need. This means that management has identified areas of business activity that might benefit from more accurate and efficient decisions. As such there is a need to coordinate investments so that these aims are achieved, with efforts to acquire relevant data, the necessary infrastructure and skills, and an enterprise wide decision management capability.
Big analytics need big data, which in turn needs low cost data. Very few organizations will willingly increase their data costs unless there are very well defined payoffs, and even then it is always best to get someone else to create your data for you. In the main this means customers through self-service mechanisms, and other sources such as social data. And the efforts to acquire low cost data should embrace diversity whenever possible, since this has been shown over and over again to be the key to better predictive models.
With a data acquisition strategy in place we can consider the infrastructure that is required to manage, store and process it. This is specifically the domain of big data and technology platforms such as that provided by Hadoop. This is an infrastructure issue, since big data alone is simply an additional cost which is borne in the belief that the data will yield some value.
Big data infrastructure is the necessary condition for big analytics. This is where we start to uncover value, and find more efficient and effective ways of dealing with customers, supplier and even employees. While big data is largely the domain of technicians and specialists, big analytics touches most parts of an organization – business managers, IT, data scientists, analysts, and of course the customer. Because of this it really is not sufficient to see big analytics as something that can be hived off to a team of analysts and data scientists, with the expectation that predictive models will appear which can then be deployed in a production environment. Regulatory requirements, performance monitoring, domain expertise, business process design and a raft of other issues need to be accommodated in the predictive model lifecycle. To this end an integrated decision management environment is essential.
Whether an organization should build or buy its big analytics capability depends on the nature of the application, as shown in the diagram below. This is a simple payoff matrix which shows that if an application is simply considered a cost of doing business, then it is probably best to buy, since there is very little room for a strategic advantage. However when predictive models can make a large strategic difference, then those opportunities should be grasped and bespoke solutions built. Of course it isn’t completely black and white, and a mixture of build and buy can be used when appropriate.
There are also questions which should be answered concerning the necessary infrastructure and tools. Some businesses might decide that they do not want the cost of supporting a big data and analytics infrastructure, and opt instead for a cloud based option. Others, typically larger organizations might opt for in-house capabilities or a mixture of both.
The use of big analytics does require some strategic analysis. Businesses that are happy to simply follow the herd where information systems are concerned, might have to accept the added responsibility of doing something a little bit different at times – simply to realize the benefits they anticipate. A quarter of organizations surveyed expect big analytics to deliver some form of competitive advantage. This is to be compared with only ten per cent who expect traditional process automation to deliver an advantage (Butler Analytics Survey). And so expectations are higher – and so they should be. Organizations which are prepared to grasp this particular bull by the horns often gain a significant advantage over their competitors.
The previous article in this series is Big Analytics Case Studies