Tarun Srinivasan, VP, Global Operating Leader, Analytics at Genpact, discussed steps to build cost-saving analytics pipelines for clients during his talk at the fourth edition of the Machine Learning Developers Summit (MLDS) titled,“Delivering analytics at scale for the end to end business chain of clients”. He dilated on the application of conventional analytics tools/techniques and new age AI/ML technologies in analytics services like data preparation, business intelligence, forecasting, optimisation and advisory services.
“Less than 0.5 percent of the data generated in the world is really ever analysed. We are generating more and more data everyday. I can ensure that we will all have jobs in the analytics space in the future even if we analyse only 0.5 percent of the total data. Because of COVID-19, we have seen accelerated adoption of analytics than ever before. We are seeing a conscious movement from autonomous to autonomy”, Srinivasan said.
Sign up for your weekly dose of what's up in emerging technology.
He outlined the common issues clients raise, including:
- How can we scale our analytics efforts?
- We don’t have enough good data
- We have outdated legacy systems which are not talking to each other
- We don’t have enough analytics talent
- The IT organisation does not speak to the business
At the bottom of the pyramid is the foundational layer. “As we go up, the complexity increases but the value also increases exponentially. Unfortunately, most of the efforts are happening at the bottom of the pyramid. This restricts data scientists to spend quality time upward,” Srinivasan said.
The focus should be on how to use artificial intelligence and machine learning to help in the processes of the foundational layer. Once ML simplifies the bottom layer, data scientists can spend their energy on other priorities like descriptive analytics, predictive modeling etc.
Building scalable analytics systems
Srinivasan elaborated on the two foundational elements to build scalable systems:
- Strong governance model: Make sure the analytics end -to-end value chain is in alignment with the company’s vision and a strong analytics engagement model is in place. The model should have a constant interaction mechanism with the end customers. Incubating bilingual skills (data scientists also have expertise in a certain domain such as finance, supply chain, manufacturing etc) is also crucial.
- Strong execution model: Ensure the right infrastructure such as automated data pipelines, leading tools for data analytics and preparation is in place. Building the right experience for the customer is also critical.
Srinivasan outlined how analytics can be used in the business processes to save money for the clients. Companies possess a huge quantity of spend data from the suppliers. The historical spend data can be crunched to extract business insights.
“This is mostly spreadsheet analytics. The power lies in its simplicity.When you analyse the spend data, you get a lot of powerful insights: How the price of the supplier has been varying over a period of time; is there any dependency of the quantity of the parts purchased; dependency of location; dependency of various pricing indices etc. Once you are able to understand the reasons for the price variations, you will be able to predict the price and also use it as a negotiation mechanism with the suppliers.”
Most of the clients have hundreds and thousands of contracts, be it with end customers or suppliers. Due to its sheer volume, the contracts do not have good visibility. Not many industrial clients have good contact management technology solutions. The contracts may be in the form of scanned images and pdfs.
At Genpact, computer vision OCR technologies are used to read the documents with great accuracy in many languages. NLP techniques are deployed to extract insights from the data. Then, machine learning is used to classify and create a structured database.
Diagnostic and prognostics
If an aircraft runs into an issue mid-air, a message is relayed to the ground station and usually troubleshooting happens after the aircraft lands.
Analytics can improve accuracy of forecasts and reduce downtime. “We observed that the algorithms and rules used to generate alerts are rudimentary and all design-based. We have used analytics to make sure the rights alerts are generated at the right time when the aircraft is landing. We had a large history of alerts and work done previously to work with to achieve this. It is a classical case of supervised ML as we could bring AI to build algorithms to dramatically improve accuracy,” Srinivasan said.
Analytics can greatly benefit medical professionals, especially radiologists in times of COVID-19. Due to the pandemic, the stress on radiologists has increased due to the sheer volume of scan requests. Radiologists can leverage analytics to parse the historical scan data to do the differentials for new scans. The goal is not to eliminate radiologists but to assist them in making informed decisions.