5 Ways to Add Cognition to Your BI Program
Reaching true system cognition remains in the future, but there are some projects that you can implement today that will improve your users' experience and increase the cognition level in your BI systems.
- By Troy Hiltbrand
- July 31, 2017
3 Flavors of Predictive Analytics AutomationThe term cognition is garnering a lot of attention in the computing world. To understand system cognition, you must begin by understanding cognition in general. Cognition is, by definition, the mental action or process of acquiring knowledge and understanding through thought, experience, and the senses. This is something that we do naturally every day, but it is different for systems.
In the past, systems did what they were programmed to do. Today, cognitive programming teaches systems to learn from experience and use this gained knowledge to guide their future behaviors. Because cognitive systems are the culmination of data acquisition, data analysis, and application of that knowledge to drive behavior, business intelligence is a prime environment for adopting cognition.
Full cognition, in the form of a system that autonomously ingests and transforms data as part of its learning process, is still years away. There are, however, five practices that can increase your BI systems' cognitive capabilities today and greatly enhance the effectiveness of the resulting data and your users' experience.
Prediction is a methodology that uses hidden patterns in the data to develop a model that provides increased insight. The two most popular predictive models are regression and classification.
Regression identifies a trajectory from past data to forecast future data. The most basic form of this (and one of the most commonly used) is linear regression. Linear regression creates a best-fit line through a set of historical data points. The resulting regression line is the model; future predictions are points on the line.
Other, more complex models of regression are available that take into consideration seasonality of historical data and nonlinear patterns in the data. These can be employed when trends are not linear in nature but still contain a pattern that can predict future data points.
Businesses often use regression when forecasting sales and managing budgets. Having an informed guess about where finances are headed is critical for strategic business decision making.
Although forecasting has been around for a long time, the application of more complex models explaining the historical trends and improving the responsiveness of these forecasts is where more cognitive capabilities are emerging.
Classification is another predictive model, but it is not dependent on a trajectory component. The process of classification extracts patterns from data where the category of the data is known and applies those same characteristics to instances where the category is unknown.
An example of classification is customer churn. With a churn model, the goal is to identify which customers will leave and never return. When a business can accurately predict who is likely to leave, it can expend limited resources to proactively prevent this exit and keep profitable customers generating revenue. In the historical data are many examples of customers who have stopped buying and examples of loyal customers. These are the instances with a known category. The goal of the prediction modeling process is to identify which attributes of these customers are the best indicators of who will stay and who will go.
The predictive model in the case of churn is a mathematical combination of these attributes in a way that accurately classifies these historical instances. Examples of attributes could include when they purchased last, how much they have purchased, how often they interact with customer service, or even whether they have talked about your company on social media.
With this model in place, you can evaluate your entire customer base to determine who is most likely to churn. Combining this prediction with the customer value, you can then implement actions that will reduce that likelihood of departure. Knowing which attributes have the strongest impact on the model also provides insight into what actions can be taken to reduce the chance of churn.
Adding prediction to your analytics provides users with a previously unknown facet of the data so that they can make more informed decisions.
Another method of adding cognition to your business intelligence systems is by applying prescriptive analytics, aka optimization. This entails running through different configurations to generate all potential outcomes and then determining which one is optimal based on a set of measures. One evolving area in the field of optimization is identifying methods that not only generate an optimal solution but do so with the best performance. The speed of the optimization is often as important as the final answer because it determines how easily the process can be integrated into the system for ongoing results.
Cognitive systems perform both the number crunching associated with optimization and the application of the path to the best outcome into production without the need for a human to interpret.
Consider, for example, inventory management. After running through multiple scenarios to identify the right levels of inventory, a cognitive system can automatically place orders to ensure that inventory levels are maintained at a level that is most profitable for the company.
3. Interactivity/Natural Language Understanding
Users are demanding that their interaction with systems be more comfortable -- more like other interactions in their lives. They want to be able to have a conversation with these systems and garner the information they need.
Voice recognition and natural language understanding allow end users to interact with data and business intelligence in a whole new way. Chatbot frameworks from companies such as Google, Facebook, Microsoft, and Amazon let users interact with systems using natural language and get the correct results. Using inputs such as voice in Amazon Alexa or Google Home or chat through Slack or Facebook Messenger, these chatbots can transform requests into usable results.
4. Dynamic Segmentation
Business intelligence users are not homogenous. Being able to dynamically differentiate and segment users based on how and when they interact with your system can help you optimize your user interface. Segmentation of users based on behavior can optimize the level of detail provided to the user and the format presented. This goes beyond role-based security (limiting what they can see) to a system that dynamically determines how they should see the data to optimize its utilization when making a decision.
5. Intelligent Delivery
In addition to providing the right analytics to the right person, intelligent delivery uses past interactions to identify the optimal delivery mechanism and timing for when analytics should be delivered. Are users more responsive when the information is delivered via email, through the business intelligence portal, SMS, or Slack? Do they respond better when the information is delivered as events are happening in real time or on a predefined schedule? What times of day do they respond most effectively?
Analyzing patterns of when and how users have responded in the past can help you create a model that will ensure that information delivered will have maximum impact. This can help to reduce the glut of information overload your users face.
A Final Word
True cognition in a business intelligence system is still in its very early phases, but the application of these five methods will allow you to raise your systems to a new level where you can provide a dramatically enhanced user experience. Pick a practice and find where in your BI stack it can be applied today to make your BI program the most effective it can be.