Infosys Nia Advanced ML increases data scientists’ productivity by order of magnitude by applying automation to the data science workflow. It offers a broad range of machine learning algorithms with industry-leading speed and scale. Moreover, data analysts, developers and even business users with a limited knowledge of data science can also build high-performing ML models as a result of its easy-to-use ML workbench. Combine this with our strong capabilities in offering next-generation services, it has enabled our clients to significantly cut down turnaround time from data to insights leading to improved decision making. Targeted marketing, lower customer churn, reduced frauds in banking transactions or optimized asset efficiency —these are just a few of the ways Nia Advanced ML opens a new world of opportunities for businesses.

Features

Nia Advanced Machine Learning enables exponential increase in the data scientists’ productivity by applying automation to the data science workflow . It also supports a broad range of ML algorithms offering industry leading Speed, Scale and Predictive Accuracy.

Conversation Engine

GUI-based features

Data analysts, developers and business users can build accurate and high-performing ML models.
machine learning

Integrated enterprise framework

for data preparation, modeling, deployment and reports
bot studio

Extensibility

for data preparation, visualizations, ML methods
omni channel

Scalable and fastest:

ML algorithms. Suite of HPC Implementations in C++ for key ML methods.
natural language processing

Streaming Fast Predictions

using Infosys Nia Prediction Server
analytics dashboard

Self-service provisioning

with elastic scaling for cloud deployments

Infosys Advanced ML Delivers Business Benefits

 

Nia ML Business benefits

Success Stories

Real-Time Fraud Detection

Motivation

  • A major global financial company was looking to detect fraudulent transactions in real time. They were looking to block these transactions or notify the customers immediately, thereby avoiding the heavy cost of dealing with fraudulent transactions.

Solution

  • Historic data was ingested and a fully non-linear, supervised machine learning model was build and deployed. The model detected fraud in real time. Incoming transactions were assigned a probability based on the likelihood they were fraudulent. The model also provides details on which variables were significant in predicting fraud. The dataflow is automatically recorded for regulatory compliance. The model is especially designed to deal with highly imbalanced datasets that are typical of this domain.

 

Benefits

$50M+ annual cost savings

Improved customer satisfaction with 10% lift in accuracy

Update fraud models once a day instead of once a year

Insight into reasons for fraud, with full audit trail

Data Center Optimization

Motivation

  • A data center for a major payments company had high equipment costs. There was a huge and rapidly growing machine data volume from thousands of feeds, leading to capital expenditure waste from overprovisioning to cover anticipated peaks in usage. There were outages which went unnoticed until customers complained, which was hurting user satisfaction. The company was in a reactive mode to their situation.

Solution

  • A model was deployed to detect patterns and anomalies and take action before the user experience was negatively impacted. This model allowed the company to take all the data, put it together, and correlate events across those streams. This allowed them to provision only equipment that was really required, and monitor thousands of systems and social media feeds at data rates of over 25 terabytes per hour

 

Benefits

Provision only what’s really required

Take action before merchants complain

Monitor thousands of systems and social media feeds

Estimated $20-$30m savings per year

Recommendation

Motivation

  • A leading handset manufacture was looking to monetize content delivery and built out a mobile video streaming service. They wanted to promote additional sales/rental of movies and wanted to tune their legacy recommendation engine running on an older version of Hadoop with Mahout. Major pain points were the speed and accuracy of their existing approach, despite having a large team of modelers within the company.

Solution

  • We built a machine learning model using our recommender system. In less than a week, the solution delivered a 1500x improvement in runtime and improved the accuracy of the model by 20%. The model provides an interpretable system that optimally understands user preferences.

 

Benefits

20% improvemen t in recommendation relevance

1500x execution speedup

Optimally understand user preferences

Allow for diversity in recommendation without excessive obscurity

Customer Acquisition Via Micro-Targeting

Motivation

  • A Fortune 100 financial customer was looking to do a micro-targeting marketing application to extend high end credit card offers to high net-worth individuals. They were using a legacy system integrated with a 100 node Hadoop solution with 1200 cores. Despite the large system, the speed and accuracy of their existing analytics approach were major pain points.

Solution

  • With our solution on a single node and 12 cores, they were able to realize a 12.5x improvement in execution speed. When they scale this across their Hadoop cluster with 100 nodes, our near-linear scalability would provide about 100x that performance or 1200x execution speedup.

 

Benefits

Remove pain points of existing legacy analytics system

5% accuracy improvemen t translates to large financial gains

1200x executions peedup

Allow their analysts to build models instead of writing software

Customer Churn in Retail

Motivation

  • An online retailer with over $100M in revenue and 19 million opt-in registrants had lots of customers disengaging after the initial sign-up. They could not identify a clear reason for this passive customer churn via the lack of engagement. Their in-house system used analytics and off-the-shelf statistics but no machine learning.

Solution

  • We developed an advanced analytics solution that not only provided churn prediction, but also identified the variables that most reliably contribute to an accurate prediction. This gave the customer insight into why their customers were disengaging, which in turn allowed them to take action to improve retention. The result was a large reduction in churn combined with a large predicted revenue increase.

 

Benefits

Churn reduced by 55% – 68%

Predicted revenue increase of 200 – 400%

Insight into reasons for churn enables corrective action

Identify which members are likely to churn

Real Estate Property Valuation

Motivation

  • A large provider of residential real estate valuation in Canada wanted to improve their property valuation using machine learning (ML). They had less data than their competitors and therefore had to be particularly diligent in its usage. The desire for this improvement was at the CEO-level on down through the company. Their existing tools included homegrown analytics and off-the-shelf statistics but no machine learning.

Solution

  • We built a supervised machine learning model that improved the accuracy of their Automatic Valuation Model property valuation to enable them to predict property values 2-3 years into the future. Our software production-ready models off-the-shelf, and our model are deployed in production.

 

Benefits

1st residential property company in Canada to use ML

Improved valuation accuracy despite less data than competitors

Production system using our off-theshelf software

Went from behind to ahead by taking their analytics to the next level

Awards

Best In Show Award
tmforum

Singapore

Contact Us