Skip to content Skip to footer
Search

Artificial intelligence explained

Example pattern for mobile
Example pattern for desktop

Summary

  • What is artificial intelligence and how is it becoming critical for business today?
  • AI comprises computer systems that use multiple technologies to perceive the world, process information and take action.
  • Beyond AI’s core capabilities, which can automate formerly manual data-handling tasks, AI can help companies make better decisions and help protect their data, equipment and even reputation.

What is artificial intelligence (AI)? How does it work? What can it do for my company? Those are all questions our AI teams hear a lot. We’ve got answers. 

AI is usually defined as computer systems that can perceive the digital or physical world, process what they perceive and, in most cases, take the kind of action that normally requires human intelligence. For business, there are big benefits available from machines that can see or hear or sense, draw conclusions, then make smart choices and act on them. 

Most business applications fall into three categories. Some make it easier for people to access important information. Some provide insights to support faster and better-informed decisions. Some automate increasingly complex activities. Some do all of the above.

What follows are the most common types of AI applications for business, brief explanations of how they work and a look at how AI manages to be so intelligent.

Giving you the data you need: information processing

AI can intelligently read, see, listen and sense to find the data and generate the insights that your business needs.

  • Natural language understanding and generation. AI can read or listen to everyday language (such as a newspaper, a social media post or a consumer talking on the phone), to interpret, understand and possibly act on such data. That can enable chatbots to have a conversation with people. It can also record and transcribe meetings. 
  • Information extraction. AI can quickly and accurately extract the precise information you need from data locked in paper or electronic formats. AI can, for example, read through stacks of contracts to pick out purchase terms and conditions, fee structures, exclusions, termination clauses, and more.
  • Computer vision. AI can see — and understand what it sees. It can observe machinery or infrastructure and then advise on performance and maintenance. One insurance company, for example, is using computer vision to detect and assess damage to vehicles. As a result, it can automate much of the claims process. Computer vision can also scan news broadcasts or read and assess text that appears in videos. It can even (with appropriate compliance and respect for privacy) identify faces for marketing and security.  
  • Smart sensors. AI can make dumb sensors smart: ingesting information on light, vibrations, temperature and sound. That can support predictive maintenance for machinery, as AI might recognize (for example) a certain vibration as a sign of potential problems. It can optimize climate controls to cut energy costs and keep machines running optimally. It can help drones and cars drive themselves.

Making humans smarter: decision support

AI can help people make decisions by providing highly accurate, continually improving forecasts and models — and by putting the right information in front of the right person at the right time. 

  • Neural networks for data analytics. A type of machine learning that imitates the human brain structure, artificial neural networks can support AI’s perception capabilities (see above). It also can support sophisticated data analytics: mining data for relevance, classifying it, detecting patterns and correlations, and disseminating insights. Financial institutions, for example, often use neural networks to spot errors or breaks in transaction data and notify the right people to remedy the error.
  • Digital twins for forecasts. With AI, you can make a highly sophisticated digital simulation of a physical asset (such as a power plant), a digital asset (such as an investment strategy), a person or set of people (such as a set of consumers), or a combination of all three (such as a supply chain). The resulting "simulation models" can produce highly accurate forecasts of maintenance needs, marketplace behaviors, supply chain risks and more.
  • Strategy games. AI can “gamify” even highly complex strategic questions, such as how to go to market and pursue acquisitions. Using reinforcement learning (see box), AI models multiple scenarios — in some cases, hundreds of thousands. That allows executives to test the results of possible strategic choices. They can dynamically model not just how their own choices might work in the current environment, but how those choices might impact the marketplace and their competitors. 
  • Adverse event monitoring. You need to know when people report adverse reactions to your products or services — and you need to know if there’s a deeper problem. Machine learning can extract and classify information from sources ranging from medical texts to social media, helping find problems fast. That’s especially critical for pharmaceutical companies where lives may be at task. It’s important too for financial institutions seeking to identify bad actors and any consumer-facing company concerned for its brand.
  • Process intelligence. Beyond the base case of predictive maintenance for machines, AI can also analyze and improve how humans are using machines. It can, for example, follow employees’ keystrokes and patterns of application use, then detect opportunities for automation and greater ease of use.

 

Grey pattern

Inside the black box: Understand how AI operates

For most business applications, the brain of AI is machine learning: techniques that allow computer programs to perform better (“learn”) as they receive more data. Machine learning, as used in business today, most often works in three main ways.

Supervised

In the most straightforward machine learning, humans give AI labeled historical data and outcomes. An example could be the history of a factory tool’s operating temperature, operating speed, maintenance record and malfunctions, with explicit correlations made. 

With enough data of this kind, AI can make predictions even for situations different from what the historical data shows. It might predict, for example, that the tool would malfunction even at slower speeds if the temperature were higher. As AI becomes operational and receives new data from its own sensors, it will continue to improve the accuracy of its predictions. 

Unsupervised

Many companies have unlabeled, poorly organized data which the right kind of machine learning can structure and interpret. A retail chain, for example, might have reams of data on sales of different products in different stores at different prices as well as on economic conditions and the weather. 

Unsupervised machine learning could read through all this data, identify patterns and correlations, and then (for example) offer demand forecasts to help improve stocking and pricing strategies.

Reinforcement

Machine learning can hone its skills in a fast-changing environment, improving through the feedback that this environment (or human teachers) give it. This feedback reinforces correct decisions. 

A securities firm, for example, might ask AI to guess commodity prices based on a set of variables. At first, the AI is often wrong, but it receives feedback by comparing its forecasts to real-world prices. It learns from its own pattern of correct and incorrect guesses until its accuracy improves. 

 

Grey pattern

Doing white-collar work: intelligent automation

AI can act: It can automate even complex tasks that traditionally require human intelligence. The following examples of current, real-world solutions are just a beginning. The complete list of AI-enhanced automation is far longer — and growing every day.

  • Data reconciliation. As AI ingests information from multiple bots — APIs, PDFs, or other sources — it can standardize, compare and reconcile disparate kinds of data. That can support everything from compliance to finance to fraud detection to the consumer experience, sparing humans tedious work while also increasing accuracy. 
  • Cybersecurity. AI can automate defensive actions, reducing reaction times and helping compensate for the shortage of cyber talent. Common automation examples include spotting abnormal identification patterns, prioritizing vulnerabilities so humans only spend time on the most urgent cases and phishing and deepfake detection through image recognition and automated tracking.  
  • Autonomous machines — and machine maintenance. Self-driving cars aren’t yet mainstream, but AI is already helping drones fly, robots act and electrical power flow. It’s also helping machines maintain themselves: not just predicting needs, but adjusting temperatures and speeds, or sending in robots to execute repair work.
  • Social listening. Rather than humans spending hours peering through social media to determine sentiment on brands and products, AI automates the “listening” as well as the reading and watching. Companies both save human resources and get fast, continually updated analysis. That’s an especially big benefit for M&A, since it helps potential acquirers quickly evaluate a target company’s products.
Emerging Technology

Emerging Technology

Unlocking the business potential of emerging technologies

Learn more

Robert N. Bernard

Director, Products and Technology, PwC US

Email

Anand Rao

Global AI Lead; US Innovation Lead, Emerging Technology Group, Boston, PwC US

Email

Next and previous component will go here

Follow us