Innovation is in our DNA

Innovation is in our DNA


From the onset, we set the bar very high. We established a tradition of excellence, and our data scientists and analysts prove the depth of their expertise every year.

Competitions and Awards

Competitions & Awards

To be the best, we must always push the limits of science and technology.


Kaggle Genentech
Cervical Cancer Screening
1st place

What we did: Predicted and identified at-risk populations for cervical cancer.

Techniques used: neural networks, gradient boosting (xgboost), blending

Read More ›

KDD Cup 2015
1st place

What we did: Predicted student dropouts on massive open online course (MOOC) platforms.

Techniques used: neural networks, gradient boosting (xgboost), blending

Read More ›

Kaggle Avazu
1st place

What we did: Predicted effectiveness and conversion of mobile ads.

Techniques used: factor models, neural nets, blending

Read More ›

Kaggle Expedia Hotel Search
1st place

What we did: Predicted the most accurate hotel rankings based on customer preferences and likelihood to click.

Techniques used: normalized discounted cumulative gain, learning to rank, gradient boosting, neural networks, blending

Read More ›
Video Camera

Netflix Prize
Tied for 1st place

What we did: Improved predictive accuracy (RMSE) by 10% over existing recommender system.

Techniques used: SVD, RBM, ASVD, AFM, SVD++, KNN movie, KNN user, SVD time, global effects, blending

Read More ›



Business Impact

See how our machine learning science and artificial intelligence are driving real results for our customers.



Customer Acquisition

Targeting New Customers

Created high-precision targeting system to drive customer acquisition for leading global cruise line

Impact: Achieved $200 million annual impact through an increase in booking rates of 56% via email and 97% via direct mail campaigns.

What we did: Implemented automated customer scoring, which provided high-precision targeting for customer acquisition.

Key innovation: Created automated predictive scoring solution to consistently establish individual’s propensity for booking specific destination trips; used reinforcement learning and multi-variate optimization.

Read More ›

Customer Experience

Gaining Visibility

Created a universal customer journey dashboard and streamlined data processes at major US Airline

Impact: Reduced by 30% the amount of time and cost of labor associated with reviewing customer claims.

What we did: Implemented data analytics platform to drastically reduce time spent on data extraction, preprocessing, and variable creation, specifically as it pertained to customer management. Provided a comprehensive view of each customer’s trip, allowing for a much more personalized experience and a true 1:1 relationship with each customer.

Key innovation: Applied sequential analysis in a nontraditional way to segment customers; used that to identify how the sequence and combination of events predicted future behavior. Used supervised learning to identify how much the sequence and combination of events predicted future behavior, which helped determine the recommended action.

Read More ›

Customer Experience

Forecasting Consumer Needs

Predicted consumers’ needs for a US-based, top-five telecom company

Impact: Reduced churn by 1%, improved marketing campaign efficacy by 20%, and generated millions of dollars in profits.

What we did: Created a dynamic learning framework to gradually update the recommended treatment for customers based on behavioral feedback. Using need states for both prepaid and post-paid customers, the framework learns to assign the best treatment — special discounts, payment terms, bundling offers — to each need state, or segment.

Key innovation: Created new concept for clustering customers and used it to systematically help marketers design new campaigns. Developed dynamic learning framework to continuously learn customer responses.

Read More ›

Customer Retention

Preventing Attrition

Predicted and prevented attrition of noncontracted customers

Impact: Improved customer relations efficiency and reduced the cost of maintaining a stable customer base for better growth.

What we did: Developed an approach to identify fading customers in industries with a low frequency of repeat purchase behavior and in noncontractual markets.

Key innovation: Created a method to continuously identify people willing to stay from the high–churn-risk group; used clustering and dynamic learning framework.

Read More ›

Customer Retention

Re-energizing Fading Customers

Detected and remediated early signs of attrition for top global airline

Impact: Generated a boost in profit margin of 0.75%, improved load factors across select routes by 2%, and reduced customer churn by 3%, generating millions in profits on an annual basis.

What we did: Implemented analytics platform to help airline consolidate its data assets into a data lake, generate personalized offers, improve operational efficiencies, and detect and remediate early signs of attrition. Extracted and continually managed 4,000+ insights across customers, commercial efforts, and operations.

Key innovation: Compared short-term behavior variances against long-term behavior patterns to define fading customers hierarchically; used reinforcement learning and supervised learning.

Read More ›

Customer Retention

Predicting Customer Behavior

Predicted store trips and recommended intimate personalization tactics for top US retailer

Impact: Grew revenues by $200M in first 12 months; improved margins by $20 million in first 12 months; on track to close Aspirational Value gap in 3–4 years.

What we did: Helped national retail chain consolidate its data assets into a data lake, providing a 360° view of its customers. Improved predictability of store trips by 30% and helped develop more precise offers that trigger a shopping trip and swell customer baskets. Delivered more accurate tactics in intimate personalization and helped develop strategies to move customers up the loyalty ladder.

Key innovation: Incorporated domain knowledge into kernel method and combined it with a gated network to optimize offers; used logistic regression, deep autoencoder, hierarchical clustering, K-Nearest Neighbor, and regression trees.

Read More ›

Customer Retention

Improving Customer Spend

Implemented new loyalty program to increase spend for top credit card company

Impact: Increased spend by 10X ($1 billion) with passion-driven targeting for loyalty program.

What we did: Improved customer profitability with personalized offers and recommendations, prevented attrition, optimized pricing, and surveyed competitors.

Key innovation: Applied image recognition modeling (in the form of convolutional neural networks, or CNN) and natural language processing (in the form of recurrent neural networks, or RNN) in financial models; also used behavior-based segmentation, and look-alike modeling with Z-scaling.

Read More ›




Demand Forecasting

Forecasting Hot Products

Predicted product winners and losers for top shoe retailer chain in China

Impact: Increased sales by up to $60 million and improved margins by $18 million through 20% inventory reduction.

What we did: Predicted demand and developed pricing strategies at the SKU level for better inventory management. Broke down products into individual attributes and simulated cognitive human processes through analytical models.

Key innovation: Identified individual attributes to best represent products; used vector auto-regression, polynomial regression, gradient boosting machine, online learning, and cannibalization analyses.

Read More ›

Demand Forecasting

Predicting Demand

Forecasted demand across retail, hospitality, travel, and CPG

Impact: Improved demand forecasting accuracy by 5–20% depending on prediction time period.

What we did: Created a unified framework for data-driven demand forecasting that can be applied and reused across multiple industry segments.

Key innovation: Pulled methodologies from specific solutions in multiple industries to build a reusable framework that can be flexibly applied to various demand forecasting problems; used time series models and logistical regression.

Read More ›

Price Optimization

Recommending Bids

Built real-time pricing recommender engine for car manufacturer

Impact: Helped increase the landed price by $150 to $200 per car, with an estimated increased profit of $30–40 million per year.

What we did: Supplied real-time pricing recommendations to auctioneers at a top US car manufacturer based on make, model, year, mileage, location, and condition of the vehicle.

Key innovation: Used Kalman filters to dynamically learn a pricing system. Patent granted. Also used KNN.

Read More ›

Price Optimization

Detecting Price Sensitivity

Determined store-to-store price sensitivity for major retailer in Asia

Impact: Generated revenue uplift by identifying 10% gap of stores that qualified for higher pricing.

What we did: Created a top-down approach to better determine price sensitivity by location among stores in a retail chain, allowing for more accurate pricing across product categories to maximize revenue at each store.

Key innovation: Developed nontraditional method of demand forecasting and applied it to pricing optimization; used generalized linear modeling framework.

Read More ›

Product Recommendations

Returning Better Search Results

Accurately ranked hotel search results according to customer preferences

Outcome: Placed first in the Expedia Hotel Searches competition.

What we did: Developed an analytics model that not only returned results that reflected the preferences of the user but also ranked them according to which ones the user was most likely to click on and purchase.

Key innovation: Deployed factorization machines to construct model-based features; used normalized discounted cumulative gain, learning to rank, gradient boosting, neural networks, and blending.

Read More ›





Improving Data Sharing

Broke down data silos to improve customer identification and quote processing at insurance agency

Impact: Generated a 30% increase in response rates for one product line and quadrupled conversion rates in a second product area, resulting in millions of dollars in increased annual revenue.

What we did: Combined four business lines into one large customer base of services and sales to manage lifelong customers’ personalization; matched lost-touch customers in one business line with new contact information.

Key innovation: Developed graph-based customer identification and matching mechanism by integrating multiple data sources; used predictive modeling, reinforcement learning, descriptive modeling, segmentation, stepwise iteration, and learning to rank.

Read More ›

Risk Management

Automating Document Reviews

Scored and prioritized 30,000 documents for review for risk and noncompliance

Impact: Reduced contractual risk and noncompliance through automated review of all documents. Helped enable portfolio view of risk exposure and review contracts for recommended mitigation actions.

What we did: Leveraged natural language processing (NLP) to automate review of all documents and create a prioritized list for high-risk documents; Identified 10–15% high-risk documents for review queue that would otherwise not have been captured by the rules-based system.

Key innovation: Extracted 150+ nonstandard terms from master service agreements along with the corresponding changes in statements of work; used topic modeling, dependency parsing, semantic role labeling, convolutional neural networks, and sequence alignment.

Read More ›


Optimizing Schedules

Created data-driven movie scheduler for top cinema chain

Impact: Generated 4% lift in profits in one year through increased attendance.

What we did: Built an automated scheduling solution that produces profit-optimal schedules by forecasting attendance for each movie at all times and by considering revenue and cost elements such as ticket sales, distributor costs, supply costs, and labor costs.

Key innovation: Predicted demand at each session level by location, time (five-minute intervals), screen type (2D/3D/IMAX/Extreme), film type, release week, demand shifts, and other often-overlooked details; used nonnegative matrix factorization, BILOG linear models, K-means, support vector regression, singular value decomposition, and dynamic programming.

Read More ›