This last year has seen a proliferation of predictive apps which Forrester has described as “the next big thing in app development.” These apps, backed by predictive APIs, have brought machine learning to the masses. Machine learning is a powerful technique where a system learns from observing data related to a desired behavior.
Wednesday at APIcon UK, Louis Dorard spoke about the value and the use cases for predictive APIs. With Gartner analysts predicting that “organizations that use predictive business performance metrics will increase their profitability by 20 percent by 2017,” Dorard found an audience eager to learn about these tools.
The Business Possibilities Enabled by Predictive APIs
At the heart of the session was the business value driven by predictive APIs. While Dorard has previously outlined the questions that an organization’s decision maker should ask about prediction APIs, getting to that point will require an understanding of the business possibilities that are enabled. Some of the possibilities include:
Churn analysis - This is the ability to observe what customers are doing and whether users are leaving your service or not. Businesses can record this activity and that becomes training data for their machine learning system to make predictions on who might leave next. Once those people are found, action can be taken against it. Businesses that charge recurring fees such as any subscription based Saas offering are ideal candidates for using this model.
Upsell - This is the ability to detect people likely to be interested in other products that are sold by the business.
Optimizations - This includes pricing optimization, which is determining the best price for a product given the characteristics of said product. Sales optimization is the ability to predict the sales of a product when given the characteristics of the product at various price points.
Fraud detection - Credit card companies often use this to detect fraud as early as possible in an effort to save money in the long term.
Credit scoring - This allows a company to review applications for credit and predict whether it is a good risk to extend credit to that person.
Different predictive APIs for different needs
Understanding the business value is the primary concern for decision makers but it’s also important to realize that one size does not fit all when it comes to predictive APIs. These APIs can work quite differently and suited for different needs. If looked at on a spectrum of abstraction from low level to high the APIs fall into the following categories:
Machine Learning Algorithm - In the simplest terms, this type of API exposes a machine learning algorithm so that domain experts can make use of it. At this level, some prior knowledge of machine learning is required. For users that have been using machine learning already, this type of API can make it easier to deploy predictive models. Examples of these APIs include ErsatzLabs.com, BigML.com, Wise.io and Microsoft Azure ML.
Automated Prediction - These APIs are much easier to use than the Machine Learning Algorithm APIs. Use cases for Automated Prediction APIs include recommendations, classification (ie. determining if a specific email is important or spam), and regression (ie. predicting the value of a house based on a set of available data). Examples of this type of API includes BigML.com, Google Prediction, WolframCloud.com, DirectedEdge.com and Prediction.IO.
Text Classification - This kind of API allows users to create their own text classifiers to be used for applications such as language detection, spam filters and sentiment analysis. Examples of these APIs include uClassify.com, MonkeyLearn.com, and Coritcal.io.
Vertical Prediction - Vertical Prediction APIs focus on one particular type of problem. These include some of the aforementioned problems such as churn analysis, fraud detection, and lead scoring. They can also be used for query classification which gets used in personal assistant applications such as Siri. Examples include Framed.io, SiftScience.com, Infer.com and Wit.ai.
Fixed Model Prediction - This kind of API already has a prediction model in place that doesn’t require any training of the system to learn a desired behavior. This allows functionality such as topic classification from an article, gender detection, or Siri-like features but for pre-defined queries. Examples include Datumbox.com, AlchemyAPI.com, Semantria.com, Maluuba.com and Nsure.io.
The Sorts of Predictive Apps that can be Built
Many well known programs and applications today are already leveraging predictive APIs. Gmail’s priority inbox is able to make determinations between regular and important email and places the important emails higher in the user’s inbox. Google Now tries to predict where a user will go next based on the location of a user and environmental data surrounding that location. A tool such as Trackur can look at a user’s Twitter mentions and asses whether a tweet has a positive, neutral or negative sentiment. Crowd predictions is a functionality used by Snips that forms a prediction based on user habits. An example is if a user rides a train regularly, Snips can take into account factors such as the weather or nearby events taking place and determine how busy the trains are going to be at the time the user normally rides them. Zillow uses predictive APIs for it’s house value estimation feature on its website. Google Translate can automatically detect the language on a given web page using text classification.
Predicting the Future is Happening Now
As decision makers come to realize the value that predictive analytics, the trend will continue to grow. Janet Wagner referenced a Transparency Market Research report that forecasts “the global predictive analytics market to reach $6.5 billion (USD) by 2019. From 2013 to 2019, the market is expected to grow at approximately 17.8% compound annual growth rate.” With the barrier of entry for machine learning effectively removed by predictive APIs, the time is now for businesses to take advantage of leading indicators made available by Predictive APIs or else fall behind in an ever competitive market.