Posted by Paul Barsch, Big Data Services Marketing, Think Big.
In the race to adopt data science capabilities, it is common for companies and services to focus on the ‘operationalization’ of data science. However, this focus on operationalization can leave data science groups taking ‘operational analytics’ for granted. Not surprisingly, the similarity of these terms causes confusion. Let’s first focus on the differences between operational and operationalized analytics, and then discuss how they work hand-in-hand.
Operational analytics are models that measure and predict the business health of an organization and its operations. Operational analytics serve a dual function: First, they measure and predict the performance of a business, including leading indicators. For example, a marketing effort to existing customers that increases product engagement is a win, even if the revenue impact is well into the future. Second, they enable businesses to ask “what if” questions by simulating the impact of operational changes. As an illustration you could ask: “How will a tiered routing approach for our call center affect upsell close rates?”
On the other hand, operationalized analytics means focusing on the integration of analytics into business units in order to take specific action on insights. This is where the proverbial rubber hits the road; where analytics are integrated into both production technology systems and business processes.
As an example, one financial brokerage created attrition models to predict which customers are likely to close accounts. But this company knew that building a model isn’t enough; it must be put into production systems and then monitored for performance. The brokerage put the model into production and from there other operating business units, such as the marketing department, were able to take action on those insights to encourage customer retention.
The last mile of maturity is to perform operational analytics for our own analytics organizations! These analytics should measure model performance over time, identify drift (decay in model performance), model biases (segments where we underperform), etc. And to push the envelope even more, some firms are talking about creating an artificial intelligence “loop” that monitors operationalized models and learns from them to create self-managing analytics operations!
Closing the Gap
Whether the goal is to simply improve day-to-day business activities with analytics or create a futuristic self-learning system to manage analytics models, operational analytics are key to success; and they must be operationalized!
To do this, enterprises should have a strong technological ecosystem that supports easily ingesting throughput data systems, developing models, managing models, deploying and monitoring analytics. But keep in mind this important point; it’s not enough to have technology, it’s also critical to create an effective analytics culture that builds processes around the throughput KPIs that define your business.
It’s not “either/or” when it comes to operational versus operationalizing. The hard work of integrating both is the path to ensuring lasting business value.