Predictive Model Factory

Cisco is a multinational technology company that develops and sells a wide variety of networking equipment including routers to switches. More recently cisco has also started to sell a variety of analytics and software solutions such as data center analytics and automation software.

Challenges

• Build models to deliver accurate predictions about customer propensity to buy across Cisco's extensive product portfolio.
• Overcome speed and scalability challenges associated with analyzing an exploding amount of information about buying patterns.

Solution

• Deploy H2O's pre-built ready-to-use algorithms.
• Streamline the Cisco prediction factory to a much simpler input controlled using R.
• Leverage H2O's in-memory compute engine to minimize the need for expensive storage resources
• Incorporate a greater amount of up to date customer buying information.

Results

• A dramatic reduction in processing time from more than a month to two days - despite a dramatically larger dataset
• The ability to analyze the entire set of customer data, rather than just a small sample -delivering far greater accuracy
• Immediate incorporation of new buying data with no need to prepare models in advance.
• The power to deploy of advanced machine learning algorithms (including GBM and deep learning) - rather than simple decision trees - for more

Inside Cisco’s Predictive Model Factory

How Cisco’s 20-person advanced analytics team modernized data preparation, model training, and score deployment procedures for 60,000 customer intelligence models, across hundreds of millions of observations.

From an artisanal approach to mass production of predictive models

Cisco has long-embraced the power of predictive analytics. Every quarter, the networking leader’s 20-person advanced analytics team deploys a set of propensity to buy (P2B) models, which predict whether certain companies will buy certain products within a given timeframe. Marketing and sales teams rely on these predictions to focus on their highest potential revenue opportunities. What started as a two person effort to develop a predictive model for a couple of Cisco products back in 2007, evolved from an adhoc, “statistician intensive” process to a semi automatic factory, a tight production process that creates, validates and deploys 60 thousand specific predictive models every quarter. Streamlining data preparation steps and repeating common sense methods for training machine learning models have allowed Cisco to predict purchases for very customized combinations of markets, customer segments and sizes, technologies and relationship maturity with Cisco. Since its inception the P2B factory has been associated with more than $3B in revenue, a tremendous demonstration of the power of data science when combined with excellence in marketing and sales execution.