Machine Learning as a Service: Swift IQ Predicts the Future

SWIFT IQ belives Machine Learning as a Service (MLaaS) will become a key market in the near future as more providers look to integrate predictive APIs with their customer transaction data. After presenting at the recent API Strategy and Practice Conference, Swift IQ CEO Jason Lobel spoke with ProgrammableWeb about how businesses can use an API-driven approach to improve their "adaptive intelligence", and shares four techniques that can be tested immediately.

Winner of the 2012 Top Innovator Award in API Infrastructure at Data Week, Swift IQ has already been providing machine learning tools to help businesses convert their data assets to APIs. Now they are banking on a future where MLaaS will become a necessity for businesses that want to create compelling, contextual shopper experiences.

With the distributed data systems like Hadoop that are now available to all business - from the boutique to the enterprise - along with open source and proprietary machine learning algorithms, new big data software services are better equipped to analyze massive amounts of data on customer sales transactions and return a more personalized prediction of shopper needs, in realtime.

Jason Lobel, CEO of Swift IQ belives the skills shortage of data scientists will mean more businesses will look for machine learning services to do the analysis for them. This means developing a path towards building up a business' capacity for 'adaptive intelligence'. Lobel explains:

"Most companies do not have good machine readable data. Most business data is locked in spreadsheets for example. So our first product is to make your data machine readable", Lobel said.

Once businesses have made their data accessible via API (that is, machine readable), they can then draw on Swift IQ's Library of predictive algorithms to find one that best predicts the behavior of a business' specific customers.

"With Swift IQ, you can trial different algorithms. We usually have to understand the data, and then can apply a sample size. We then apply the algorithms to see if the algorithms are accurate. We want to see good response features in the models so we'll iterate with the variables first and then ask our client for a massive data dump [once the predictions are proving accurate]. That discovery process is part of our value add. We deliver all of this via an api so it can be delivered on demand and consumed in any application the client wants to consume it in."

Lobel belives that ecommerce businesses should pay particular attention. "The businesses that have the most to gain are the ecommerce companies. They can deliver a better experience with Machine Learning as a Service."

A Closer Look at Machine Learning Algorithms in Practice

In making a business' data machine readable, Swift IQ focuses on integrating data via APIs from four main sources:

  • JavaScript tags from a business' website
  • Product data from affiliate feeds or a business' own product catalog
  • CRM data on shoppers and customers
  • Segmentation data from external sources like the Acxiom.

Lobel classified four types of algorithms that can be used in machine learning to make predictions:

  • Recommendation Engines
  • Frequent Pattern Mining
  • Classification
  • Clustering

Recommendation Engines - like those used by Amazon - allow for a more personalized online shopping experience that helps retailers present the sorts of products that shoppers are most likely to be interested in. Lobel warns there is no "magic bullet" in recommendation engines, and the best value can often be acheived by presenting a number of recommendation algorithm results on the same page. He showed this example from Amazon where 8 different recommendation algorithms are applied to generate results:

Frequent Pattern Mining is particularly useful to supermarket chains who may want to organize shelving patterns to up and cross-sell products to store visitors. By analyzing what products shoppers buy in conjunction, supermarkets can make it easier for customers to remember what they want and get all they need. Up until now, it has been too costly for a supermarket to analyze the Big Data inherent in an individual shopping bag and know whether a customer is likely to want to buy milk when they next get a loaf of bread, for example.

Classification is used to better organize keyword search results to better suggest ranking of search terms based on their likelihood to be relevant to the searcher. Classification algorithms could be used in conjunction with knowledge about a customer's previous buying habits. For example, if a searcher had previously bought Ford car accessories, when they next search for "steering wheel covers" they could be returned with search results ranked by suitability to Ford car models first.

Clustering is a process to help identify high value customers. Perhaps buying specific items, or purchasing with a specific frequency is most common amongst particular customer segments. Clustering uses machine learning algorithms to uncover the hidden gold in customer transaction data that would let a business identify and better target their most valuable clients, even if this did not appear immediately obvious from any one-off purchase. Lobel notes that this data is extremely difficult to read and analyze without the help of machines, and often makes use of services like D3 to represent data results visually for better analysis by business customers.

Thanks to Jason Lobel for sharing images from his API Strategy and Practice presentation, available for viewing on SpeakerDeck.

Be sure to read the next Best Practices article: PW Interview Simon Chan PredictionIO