Most enterprises spend a significant part of their IT budget on application development and maintenance (ADM). As per Gartner, ADM accounts for 34 percent of IT budgets. The cost grows as the technological complexity along with business processes and application size grows. IT managers focusing on reducing ADM costs face challenges from business owners looking for new features and processes, which can result in more complex and thicker applications. The more complex and thick the application, the higher the overall ADM cost. This increase in ADM cost coupled with constant pressure to reduce overall IT cost, forces IT managers to explore other avenues like predictive analytics.
More than 50 percent of ADM costs go into application maintenance and remediation due to ad hoc faults and failures. With teams spending the majority of their time on these critical failures, predictive analytics can drastically help to reduce costs. Some of these failures potentially bring the IT systems down resulting in business losses. IT managers have tried various techniques like regression testing and application monitoring to minimize the cost and losses from these failures. However when it comes to ADM, the majority of enterprises use staff augmentation, which can be problematic because few teams usually think beyond providing quick reactive fixes.
Because the cost of fixing these failures can be significantly reduced if they are predicted in advance, predictive analytics becomes doubly important. This is not a new concept. Predictive analytics has already been used in various domains like banking, insurance, and retail. As in these industries, the majority of critical errors are driven by issues happening earlier.
For instance, a failure in a settlement system of an investment bank could be a result of a particular type of trade happening for the first time or an error that has already happened in upstream systems, such as trading and clearance applications. These errors in various systems can be associated to predict potential critical errors in downstream applications.
Another interesting example is database bottlenecks resulting in queries that run slower than usual. Due to the slow execution of these queries, the average number of concurrent requests—due to backlog—goes up. Eventually application servers run out of threads and hangs. This brings down the application. As we can see, this could be a common problem for many enterprises. Application down time can be reduced by proactively resolving associated issues, in this case database slowness.
A popular predictive analytics technique known as association, also called Market Basket Analytics, is ordinarily used by retailers, but can be applied to application health and behavior data to get insight on failures. Application analytics and monitoring platform can be constructed to reduce defects, and drive software and hardware optimization. The platform collects various types of log files that can be stored together in a clustered environment to avoid any space constraint. Data can be used to monitor applications in near-real time mode and also provide proactive recommendations for optimization and potential failures. The prediction of performance and failures can be made using analytics techniques such as association rules and regression.
Therefore, we can say that predictive analytics plays a vital role in finding out critical application failure areas well in advance, helping enterprises excel in the dynamic market landscape.
Though a lot of work still needs to be done in this area, predictive analytics will play a vital role in finding out critical application failure areas well in advance helping enterprises excel in the dynamic market landscape.
The article was originally published on Software Magazine and is re-posted here by permission.