Business firms are highly conscious of both the potential and the risk of driving the next wave of business transformation using digital analytics. The potential comes from the ability to take faster and better decisions using AI and driving business automation even in areas where human judgement is involved. The risk comes from the inherent complexity of such solutions. They are successful only when a lot of pertinent & high-quality data is used, the right algorithms are trained, the deployment is quick, and the decisions can be consumed easily. Additionally, the models should be explainable, their performance measurable and they should be able to adapt to changing business scenarios quickly.
Infosys has successfully delivered potent business solutions in the areas of forecasting, customer analytics, vision analytics etc. to its clients, by leveraging the powerful capabilities of AWS. Infosys has mitigated the risks mentioned above using AWS’s superior algorithms and the out of box support for the end-to-end AI lifecycle.
AWS’s three-layer AI capabilities have helped to deliver greater value to our clients. This 3-layer AWS ML stack represents different levels of abstraction that balance speed to market against customizability of AI solutions. The AI Services level provides powerful pre-built AI algorithms through API calls. These span computer vision, speech, natural language processing, chatbots, forecasting etc. The ML Services level provides managed services and resources for machine learning to data scientists. These types of services enable you to label data, build, train, deploy, and operate custom ML models without having to worry about the underlying infrastructure needs. The ML Frameworks and Infrastructure level is intended for expert machine learning practitioners. In AWS, you can use open-source ML frameworks, such as TensorFlow, PyTorch, and Apache MXNet. Infosys uses the appropriate layer to realize the best solution for its clients.
For example, a client wanted to improve the license usage processes for their customers and wanted to understand the problem areas in the process by extracting insights from ticket and customer survey data. We leveraged Amazon Comprehend, an inbuilt AI service, for key word extraction and initial topic modeling. The inbuilt capabilities of the service allowed for quick insights from the data. Topics were refined further using Sagemaker algorithms such as Neural Topic Modeling and LDA. These high-performance algorithms could be deployed quickly using AWS sagemaker.
In another example, we leveraged Amazon Forecast for generating demand forecasts at scale for a manufacturer. This advanced forecasting service was also used for parts demand forecasting for another manufacturer. We also leveraged Sagemaker’s optimized algorithms such as XGBoost to create alternate forecast models.
Beyond the algorithmic performance, we were able to automate the end-to-end ML pipeline and make forecasts available for consumption using AWS Glue, S3, Angular JS and EC2 instances.
AWS’ AI services help avoid reinventing the wheel since they already contain high performance algorithms made available through a simple API call. This allows us to focus on simply fine-tuning the approach. For example, the use of AWS’ AI services like Amazon Forecast allows us to tap into the best practices of large-scale forecasting from Amazon.com’s own experiences. Amazon’s neural network-based models allows for pattern learning across hundreds of SKUs.
AWS’ IOT Analytics that allows for pre-built transformations on IOT data, the ability to analyze large scale data, with auto-scaling and the inbuilt support for analytics. This native support has helped provide client solutions with lesser time and effort.
Leveraging AWS Sagemaker’s automated capabilities for training, storing, and deploying models we have helped accelerate client activities significantly. We have been able to combine these capabilities with ease, provision infrastructure and deliver solutions to clients in rapid iterations.
AWS’s Model Monitor service, the ability to log output to Cloud Watch and the inbuilt capabilities for A/B testing and experimentation offer potent infrastructure for detecting and correcting model and data drifts. We leverage these as a key strategy for model maintenance.
Maintainability and automated operations models must undergo changes to meet with evolving business conditions. For example, we deployed a machine learning based customer segmentation application for a manufacturer in Europe. When that model had to be scaled to other geographies, it had to be analyzed for alignment to the business operations in those geographies. Such agility and maintainability can only be achieved through provenance tracking, audit trails and logging. With the automated infrastructure capabilities, and native integration with Cloudwatch, AWS and Sagemaker made it easy to determine which forecast came from which source data and modularity of the pipeline.
Point of View