The main areas of focus for improving decision outcomes over time should be on establishing a post-decision feedback loop and evaluating the success of the decision.

Machine learning

When feedback analysis is automated, it is known as machine learning. Context data services should be designed in anticipation of machine learning. The pattern of context discovery for each decision should be recorded and tagged (advanced data brokers should be employed to record their context aggregation patterns for later analysis). The calling application that is used looks for ways to evaluate the effectiveness of the decision and feed that rating back to context data service, completing the loop. To facilitate this process, the service should possess two different interfaces which are connected to the deciding application – the call for context data services and the call to communicate back the success rate.

Analyzing and altering

Algorithms and models should be analyzed and altered in order to improve the outcome. This should be done outside of the deciding application and behind the interface. There might be a custom code or call to an external service behind the interface. The feedback analysis is often carried by human data scientists, who alter the models according to requirements. Hence, the feedback loop and subsequent learning in algorithmic decisions change over time, from simple to more and more advanced.

Systematic approach

The need of the hour is a move towards a systematic approach for context-aware algorithmic business decisions in applications. A strong leadership would be required for this and would take a lot of effort and time for implementation. But the organizations that successfully achieve this would be differentiated form other as they would be more intelligent and would continuously improve their business operations. IT leaders should now begin to make necessary investment in this endeavor.

Immediate action plan

There are many consequences to a bad decision to be considered. The system becomes aware of a failure when the outcome of a decision is evaluated. Sometimes, a separate device is developed to subscribe to the feedback event, in order to facilitate some kind of compensation for the faulty decision. It may be programmed to send a control signal to a device so that the action could be reversed and also send an apology letter to a customer. Although the required course of action depends on the application involved, having a dedicated service that is all set to act in case of a faulty automated decision is strategic.

Today, most of the decision outcomes analyzing and refining of decision automation metadata is done by humans. But the investment in machine learning is growing and hence, strategic application planners should expect a better ability to automate the whole decision cycle loop.