Integrated Analytics Strategy

To integrate the analytics capability in an organization, there has to be something to integrate – obviously. The reality for many organizations, is that analytical activities are fragmented and partial. And since the ultimate goal of all analytical activities is to help automate decision making, either partially or completely, then this means that many decision making processes are still largely manual.

It is not uncommon to find business rules maintained manually in hard copy, and even then there will be modifications and exceptions that are undocumented and part of departmental folk law. In a busy business this is the line of least resistance, and so it is the one often chosen. Deploying business rules into a rules management system requires a level of intent and discipline that is hard to realize when day-to-day concerns are calling.

So, despite the illusion that businesses around the globe are automating their decision making by using both qualitative (data visualization) and quantitative (statistics, data mining) methods, the reality is somewhat different. And this is a good starting place!

To integrate analytical activities there are several bridges that need to be built:

1. Integrate qualitative methods with quantitative methods. Data visualization is a catalyst – the way we determine the state of play. Once anomalies have been identified, then we need the help of qualitative methods to establish significance and possibly causes. Without this link data visualization simply reveals interesting events and phenomena, but no means of understanding them.
2. Quantitative methods reveal patterns that give insight into the behavior of customers, suppliers, machines – or whatever is of interest. Once we have such patterns we can exploit them by building predictive models and/or implementing rules within a rules management system.
3. Deciding when particular rules and models should be deployed for best overall advantage is an optimization problem. And so we use prescriptive analytics methods to optimize decision making.
4. The next bridge is crucial, and in many businesses barely exists. Deploying models into a production environment is best achieved using a rules management system when appropriate, a model library, a deployment language such as PMML, or possibly an API. In reality however, the logic embedded in models is re-coded in Java, C++, or whatever the language of choice may happen to be. This is fertile territory for errors, delays and unnecessary costs.
5. Once models are deployed they need to be monitored, and so the analytical toolkit should support some form of model monitoring and reporting. Some suppliers provide this level of capability, but many don’t.
6. Finally, it is clearly desirable that the effect of business rules and predictive models should be visible to management, via some form of enterprise performance management capability. This is almost entirely lacking in most businesses.

So in reality the integration problem can be clearly stated, and the actions needed to achieve integration fairly simple to understand. But simplicity does not imply ease, and realizing such an integrated capability will mean organizational, cultural and conceptual changes. This is always difficult and generally comes about through the force of necessity. It also implies that integrated enterprise analytics needs very high level sponsorship, unless analytical activity is to become fragmented and departmentalized.

There are big gains to be made from integrated analytics. Money saving efficiencies can be easily realized. The effectiveness of models and business rules increased. Room for errors decreased. Regulatory requirements are more easily met. And perhaps most important, management become engaged because they see the commercial impact of analytical activities. Integrated analytics is an inevitability – and some will get there before others.