Why is the concept of AnalyticOps important? There are four main reasons. Firstly, the data scientists who typically develop analytic models are often divorced from the IT environment tasked with putting their models into operation, and this can cause issues with that process. Moreover, the developers of AI models need to be able to deliver metrics to business owners in terms they understand. How else can those business owners be in a position to approve or not approve the deployment of those models?
Secondly, models can get out of date quickly and need to be replaced or updated on a regular basis. Formal processes need to be in place to support this. Thirdly, as predictive maintenance and Internet of Things (IoT) applications become more prevalent, companies are going to have hundreds, thousands or even tens of thousands of models (whether stand-alone or built into applications) and, with these sorts of numbers you are going to require some sort of management for that environment, not to mention real-time visibility into AI health.
And, finally, there are issues over bias – if you have a biased model your decisions based on that model will not be optimal and could lead to unfair outcomes for customers – and ethics and compliance. For example, the EU’s GDPR (general data protection regulation) requires that models be explainable.