Key Strategies for Profitable Business Analytics

817

Business analytics serve only one purpose – that of helping people make better decisions. These decisions might occur at the level of transactions, tactical operations, and strategy. Business intelligence for example is largely concerned with analysis of prior performance and support for diagnostics. It is mainly used to support tactical decision making by managers at various levels in the business, and is usually not useful for transaction based decisions or strategic decisions. These latter generally tend to consider macro factors such as economic conditions, competitor activity, market trends and so on.

The business analytics space is becoming quite crowded, with machine learning, prescriptive analytics and artificial intelligence adding to the analytic mix. Machine learning concerns itself mainly with applying algorithms to historical data, in an attempt to detect patterns of behavior that might be useful in future activities. In loan approval for example, we interrogate historical data looking for characteristics that might indicate a loan applicant will have no problem repaying a loan. Many loan approvals are now processed automatically with very little human involvement. Clearly there is considerable scope here for adding intelligence to operational applications dealing with customers, suppliers, employees and trading partners. As such the intelligence needs to be embedded into these applications so they are available at the point of work. This is also true of business intelligence, and particularly the embedding of various visual artifacts (charts, graphs, dials etc) into the applications that are used day after day in a production setting.

The current fascination with all things visual is understandable, but a business will not realize the efficiencies that business intelligence and machine learning can deliver, until analysis is embedded into production applications. Such analysis can speed up processing of transactional activity, and ultimately will completely automate a good deal of it.

Prescriptive analytics is fundamentally different from BI and machine learning in that it establishes how processes should execute to make best use of resources. BI and machine learning are concerned with what has happened or will happen. Prescriptive analytics is concerned with how to best use resources given various forecasts and plans. The optimal deployment of sales reps, factory scheduling and even product pricing are all activities that can benefit from prescriptive analysis, and optimization technologies are key to this.

That the big payoff from analytics comes when the analysis is embedded into production applications is well demonstrated by the emergence of the Decision Model and Notation (DMN) standard. This provides diagrammatic and integration methods for embedding decisions into business processes. The decisions may be manual, semi-automated to fully automated. What is important is that the decision logic can be integrated into a production environment.

Finally we come to artificial intelligence (AI), a broad collection of technologies and methods that support automation of processes and decisions. Machine learning happens to come under the umbrella of AI, and more advanced techniques such as deep learning are enabling highly accurate image recognition, language processing and speech processing. Again AI does not sit in a world of its own, but is most useful when it can assist with and possibly automate day to day operational activities. Examples include chatbots that interact with customers via messaging platforms for customer support and helpdesk applications. AI is also being used to automate business process improvements and execute real-time marketing optimization.

In all of this the key issue is that analysis and intelligence should be integrated into the working environment. Some suppliers call it smart applications, presumably in contrast to dumb applications that do little more than act as system of records applications. The emergence of Internet of Things (IoT) data streams only emphasizes the point that analysis needs to execute in the production environment, and of course that the inevitable trend is toward real-time analysis.

Much greater demands will be made on infrastructure and integration technologies, and on the various skill sets within the organization to work together. And absolutely critical to all of this is the creation of processes that allow analytical models to be designed, built, implemented, monitored and modified, so that the organization has a firm grip on the decisions these models are making. Anything less and a business may well find that analysis has become out of sync with reality, and all that implies. In reality the technology is the easy part. Disciplines, culture and methods as always will be the real challenges.

Thank you to GoodData for supporting this article.