Azure ML end -to-end Operationalization Competency

Posted By:

Azure ML is strategically positioned for building and rebuilding, an agile and customized solution

Azure ML’s speedy and adaptive end to end operationalization – ability to revise the implementation strategy based on changes in market conditions, social signals or business change is one of its biggest Unique Selling Proposition, which personally to me is the most exciting.

Coming from a financial services background, where the time to train model and implement the strategy takes around 6 months to one and half years! So by the time the model is trained and implemented it may already have lost some of its predictive power and population dynamics may have changed. And in case we need to do revision, it’s again a significant delay.

Azure ML offers a stark contrast to this with its flexible and more agile operationalization. In the current era of data overload and customer personalization, we want to refine our models and strategies to make sure we are aligned with changing market conditions as well as customer preferences to the second if possible.

With Azure ML, as the data and experiments are already in cloud and, it can be as easy as click of a button to refit the model to be representative of changing conditions or re-building model by incorporating new data. The latter though time intensive is still more easily done in Azure ML, as part of workflow – data and experiment, is already in and we have an array of predefined modules to use to rebuild the model. Once we finish with rebuilding the model, we can operationalize the service with the click of a button!!

This is particularly helpful in customer analytics, campaign management and inventory management. To call out a specific example, we built an Inventory forecast model to incorporate both historical data and current social media traffic to predict the optimal forecasting levels. The scenario was for a new phone launch, where the demand is high and the sales are unpredictable as there are lot of dynamic elements – social signals, phone quality, brand loyalty etc. and hence we needed to re-fit models to ensure inventory is representative of current sales and customer response based on social media data.  Additionally the model also recommended inventory transfer actions between nearby stores in order to rotate inventory, ensuring we were minimizing losses associated with sold out stock and unsold excess inventory. With Azure ML we were able to seamlessly consume the high velocity and high volume social Twitter data highlighting customer response and current phone sales to re-predict optimal inventory levels every day.