On any given day, organizations of every type, size, and vertical face numerous operations and business process challenges. They’re looking for better, faster, cheaper: improved service, better products, more satisfied customers, and higher revenues. Increasingly, part of the answer is found in the application of AI to these challenges. Indeed, for some markets, AI is rapidly moving beyond a differentiator to table stakes, and “evolve or die” sentiment is growing.

But where should you start and how do you identify the opportunities for AI to solve both simple and complex problems, and to deliver on key goals? A good place to begin is to understand the various categories of AI and ML use cases, and how they’ve driven value in organizations in situations that might be comparable to yours.

For start-ups or internal teams trying to figure this out, I’m providing a comprehensive list of the most widely successful use case categories, together with common customer needs where AI can be applied based on the type of problem you’re trying to solve. I’ll tackle 12 uses case categories over the course of this and the following two posts.

1. Classification: As one of the primary use case categories for AI, classification means to identify, sort, label, or classify objects including pictures and human faces. In order to build a classifier, the AI learning algorithm builds a model using historical data, which has often been labeled by a human. The algorithm learns which patterns of features of the input data correspond to certain classes based on this data.

Classification is central to facial recognition systems, spam filters, fraud detection systems, and thousands of similar use cases. It’s also critical in applications involving optical imaging and vision systems, including robotic inspection and assembly line work, as well as systems that power safer autonomous vehicles and advanced driver-assist (ADAS) systems. For instance, your android phone uses AI image classification to identify all pictures of dogs, cats, and the like. And Alibaba has a system that detects whether sows are pregnant in a pig farm.

2. Prediction: AI models for prediction are based on analysis of current and historical data to make predictions about future or otherwise unknown events. Also called predictive analytics, methods for prediction include various statistical methods (like predictive modelling) and AI methods like deep learning). Many prediction models identify the likelihood of some future outcome, such as for health, stock prices, market fluctuations,  weather, loan defaults, customer churn, and more. For instance, Urbint predicts problems with gas lines to trigger preventative maintenance before there is a problem. And several companies use machine learning for stock price prediction (the most successful of these efforts are not described in public domain locations, for obvious reasons).

Classification and prediction use cases comprise the lions’ share of AI implementations today. However, as AI matures and moves into wider deployment in the enterprise and in smaller organizations, AI applications are expanding into a number of additional categories, which I describe below and later in Part II of this series.

3. Optimization: Optimization can be described as the achievement of outcomes against goals. This might involve calculating the maximum or minimum value of some function, by simulating a variety of options to find an optimal solution that can be used to guide customers toward the best decision under different circumstances. 

For instance, organizations looking to maximize profits, based on a set of parameters (costs, materials, labor, overhead, and price) can run various simulation models to help guide them to profit maximization through finding appropriate pricing models for products or services. 

AI is used in field workforce optimization, in transport for route optimization, and for logistics and resource allocation and management. Companies like Ambyint are using AI to help customers optimize and automate their well pumping operations. Start-ups like LogiNext are using AI for field workforce management. Celect is using AI to optimize inventories for retailers like Neiman Marcus, Anthropologie, and Urban Outfitter. Note that optimization is also a core goal of the field of operations research, which overlaps somewhat with AI methods

4. Robotics and control: AI is increasingly being used for a variety of applications that require control of robotic devices such as unmanned vehicles that map ocean floors or planetary surfaces. Other uses include bomb detection and monitoring, and inspection and intervention within several research fields and industries such as  oceanography, marine biology, military, and oil and gas exploration, obstacle detection, avoidance and recovery operations. These applications may have other AI capabilities as components or subsystems within an overall observe-orient-decide-act control loop. Probably the most well-known robots in this category are built by Boston Dynamics, which was recently sold by Google.

5. Monitoring, Anomaly Detection, and Situational Awareness: AI is heavily use for monitoring situations or systems to detect potential issues and help humans understand the whole system without drowning in a flood of data. Examples include monitoring large-scale operations and equipment, such as detecting aircraft system anomalies, or monitoring and detecting patient health conditions. 

Beyond accidental issues to manage, there are also bad actors to worry about.  As a result, AI and machine learning applications that can automatically find unexpected anomalies in data or that detect intrusions and look for unusual patterns of behavior are proliferating. 

Specific use cases include those for monitoring and detecting fraudulent financial transactions, money laundering, and to detect and protect against potentially dangerous phishing expeditions that can install spyware, remote-access, Trojan horse attacks or ransomware within organizations.  DataVisor is helping the  financial industry to thwart potential breaches, bank fraud, and other types of criminal activity using an ML-based.AI solution, while innovative start-ups like IronScales are using ML for intrusion protection systems with an automated phishing-mitigation response

6. Decision making: More organizations are using AI to aid in decision making—to enhance their human intelligence, all the way up to and including the complete automation of decisions. AI is treading a similar path to analytics, moving from simply descriptive or diagnostic systems of the past to being more predictive and prescriptive. It’s providing visibility into and shaping the future and suggesting options and scenarios for decisions. 

The challenge going forward is how to make sense of the complex process by which these decision options are made and presented and whether they can be trusted. An increasing number of intelligence platforms are helping customers to monitor their operations—industrial, agricultural, technological, cultural—and provide informed options on which they can act/decide how to move forward or respond to changing conditions.

For instance, decision intelligence (DI) company Quantellia deployed its technology to help a large bank to optimize decisions around a 53-country technology transformation.  And Google has recently taught decision intelligence to over 17,000 engineers, helping them to move from AI systems that simply predict or classify to those that bridge from actions to outcomes.  

Stay tuned for my next post, where I’ll present an additional six categories of use cases that help in identifying and focusing on the types of problems organizations need to solve.