Artificial intelligence (AI) and machine learning have advanced significantly in recent years. Once the stuff of science fiction novels, AI and machine learning are gaining traction in the enterprise, offering tremendous promise for improving organizational profitability and efficiency.
First, it’s important to distinguish the differences between AI, machine language, and their various subsets.
AI, which had been focused around the use of inference engines in the 1980s and 1990s, is the use of computers to simulate human reasoning. A familiar current example of this is IBM Watson, which can examine thousands of pieces of text to identify trends and offer up conclusions. For its part, IBM recently announced that Watson will be involved in a year-long research project to detect and respond to cybercrime threats.
Today, there are two common models for AI. In supervised models, the systems are trained to tackle a particular problem. A good example of this is with facial recognition where an AI system is provided examples of faces to help make a match based on facial features. Meanwhile, unsupervised models are provided no such background information and must learn on their own.
By contrast, machine learning involves the use of algorithms to iteratively learn from and adapt to data, enabling computers to find hidden insights from data without being instructed where to look.
“Artificial intelligence and machine learning have progressed from the annals of science fiction to real-world applications, delivering tangible business impact.”
For its part, deep learning involves a set of techniques that allow data scientists to use many layers of neural networks across different parameters. Still in its early days for enterprise use, deep learning is showing tremendous promise in applications such as fraud prevention and speech/image recognition.
The Business Upside of Machine Learning
The methodology where enterprises can move the needle the most is with predictive machine learning algorithms. Machine learning technology providers are beginning to focus on specific business challenges, such as identifying and responding to top-line opportunities for sales and marketing teams.
Predictive machine learning algorithms also offer great opportunities for sales, marketing, and customer service teams to identify and immediately take the next best action with a customer or prospect. For instance, if a customer calls into the contact center for a wireless service provider to cancel his service, the agent can be prompted to offer a bundle of services that are aimed at retaining the customer.
Meanwhile, machine learning can enable a sales associate to determine the most effective content and messaging to share with a prospect or customer based on their current position in the sales pipeline.
A growing number of Platform-as-a-Service (PaaS) providers are now emerging, offering support for both general-purpose platforms (applications which include self-service tools) as well as vertical industry applications. Vertical-focused applications in this space include:
• Risk modeling for customer loans in financial services.
• Detection of fraudulent credit card usage.
• Recommendation engines used by companies such as Amazon and Netflix to determine the products and services that ‘lookalike’ customers most likely will want.
• Calculating which prospective members will deliver the greatest potential long-term customer value to healthcare insurers.
PaaS-based predictive machine learning algorithms offer a number of operational and business benefits to enterprise companies. For starters, by using a cloud-based approach to PaaS-based predictive machine learning, enterprise companies can focus their resources on solving business problems and not have to worry about coding algorithms on their own.
Moreover, by utilizing machine learning as a service, enterprises don’t need to hire or retain a pool of inhouse data scientists and other costly specialists.
Plus, under a cloud-based model, enterprises pay only for the resources and services that they use. In addition, in bypassing the setup that’s normally required for development, enterprise companies can also achieve faster time to value. Furthermore, a cloud-based model enables easier integration with existing data sources.
Another benefit to using a cloud-based predictive machine learning algorithm is that it can scale to handle the biggest time sink in the data science timeline which is data preparation. This includes gathering, cleansing, and extracting data, which represents up to 80 percent of the time consumed in data prep.
Tapping Open Source
Open source AI and machine learning frameworks also offer cost-effective alternatives for enterprises. Thanks to open source offerings such as Google TensorFlow, OpenAI, and PredictionIO, AI and machine learning initiatives can be less expensive for enterprises by leveraging a larger pool of servers for compute power via the cloud. Meanwhile, as more people contribute to an open source AI platform, this helps accelerate the evolution of deep learning systems.
A terrific example of an open source predictive modeling platform is Kaggle. Launched in 2010, Kaggle hosts crowdsourced predictive modeling and analytics competitions in which companies and researchers post a business or technical challenge and data miners known as Kagglers compete to produce the best models. Kaggle is a novel way for enterprises to engage an army of data miners without having to build their own models.
Another intriguing approach to machine learning in the cloud is DataScience.com. Effectively positioned under a data science as a service model, DataScience.com can unleash its team of on-demand data experts to tackle a company’s pressing business challenge (e.g. identifying the root cause behind churn with a specific set of customers). Under this model, enterprise companies don’t have a dedicated team of data scientists. But they can obtain resources as needed through a monthly subscription model.
The potential for applying AI and machine learning in the enterprise has been talked about for years. Thanks to advances in the accuracy and power of AI and machine learning engines, combined with wider availability of resources, predictive modeling is now within reach of enterprise users, without having to hire and build from scratch.