Embedding AI in your enterprise: An introduction for the C-suite

In March 2023, Bill Gates declared GPTs “the most important advance in technology since the graphical user interface . . . as fundamental as the creation of the microprocessor, the personal computer, the internet, and the mobile phone.”

Research by OpenAI found that 80% of workers in the United States (U.S.) could influence or enhance 10% of their work using large language models (LLMs) like ChatGPT. A fifth (19%) of those employees could see up to 50% of their work affected.

A new corporate paradigm has arrived. So, what does this mean for the C-suite and top management? Consider this: what if everyone in your enterprise was ten times more productive? Or ten times more creative? Ten times more valuable to the company?

The size of the artificial intelligence (AI) prize is enormous. Goldman Sachs expects generative AI alone to increase global wealth by 7% (nearly $7 trillion) and boost productivity growth by 1.5% over the next 10 years.

At all levels of business, one main constraint in boosting wealth and productivity is a lack of time. How many ideas have you given up on just because you didn’t have the time or resources to think of a way forward? This is the true value of AI: not only automating mundane tasks, but also delivering new and more efficient ways to work.

AI upscales your operations and transforms your quality of service. Successfully implementing AI can create new revenue streams, growing your profit margins and maximizing return on investment.

As a business leader, AI is a huge opportunity for you—as critical as outsourcing in the 1980s, or the adoption of the personal computer at work. Can you remember the office before email? Soon, you won’t remember work before AI.

C-suite leaders need a strategy to make AI work—to operationalize the technology, establish the framework, and embed it in the enterprise.

First, let’s recap AI’s journey to today.

AI: How far have we come, and what’s next?

AI technology has quietly revolutionized over the last few years. The advances have become apparent in highly sophisticated LLMs. OpenAI’s ChatGPT is probably the most famous example. Yet these advances wouldn’t be possible without the continual evolution and aggregation of data, advancing GPU architectures and—more recently—the advent of deep learning and transformer frameworks.

Deep learning focuses on training complex AI systems to learn and make predictions from large datasets. Transformers, meanwhile, introduced an architecture better able to process sequential data and capture the complex relationships within those datasets. The combination of deep learning and transformers has paved the way for today’s advanced, context-aware AI systems.

There are four major AI paradigms executives should be aware of. Each paradigm is enabled by deep learning and transformers:

1. Supervised learning

Supervised learning is the best-known approach: AI models are trained on labeled examples. It creates accurate and reliable AI systems by learning the known and desired outcomes in the labeled training set. However, supervised learning requires substantial amounts of labeled training data, which can be costly and time consuming to obtain. Supervised models also stumble when they encounter new or complex situations that deviate from their training.

2. Unsupervised learning

Unsupervised learning is when a model learns patterns and structures from unlabeled data without specific guidance. AI models trained this way excel at uncovering valuable insights and patterns without relying on labeled examples. However, the hands-off training process is at the mercy of the model’s internal reward systems. As such, unsupervised learning can sometimes produce tangential results when trying to solve a specific use case or business problem.

3. Reinforcement learning

AI models can train alongside human subject matter experts. This is called reinforcement learning. It's usually defined by AI models receiving feedback from human agents to produce a desired outcome. From a business perspective, this type of AI model training excels at creating accurate AI models customized to a specific use case. Reinforcement learning produces reliable insights which can then be safely integrated with automation to deliver value. One caveat is that reinforcement learning does take time. However, the training process can be greatly accelerated through active learning. This is where the AI model prioritizes only the most informative example data for the human agent to label. Improving performance with less human effort.

4. Generative AI

Generative AI describes a set of algorithms capable of generating new content from training data. Examples include ChatGPT, DALL-E, Cohere, AI2, and DreamFusion. Generative AI can automate ‘creative’ work, like composing a customer email from scratch, or images, and music. However, generative AI models are difficult to customize to specific use cases. They're also prone to ‘hallucinations’ (producing incorrect, bizarre, or even offensive content). To get the most value from generative AI, it must be attended—that is, at a moment of discernment, a human should review every output before it’s consumed by the rest of the business.

Driving innovation with AI-powered automation

The four AI paradigms have big implications for the enterprise. The vision of artificial general intelligence, capable of mimicking a human’s work entirely, remains very far-off. But we’ve comfortably arrived at the point of AI being used as a general-purpose technology—like the printing press or steam engine before it.

The latest AI models signal a new era for enterprise automation. Businesses seeking more value from automation and technology can now integrate generative AI in their process automation stack. With AI-powered communications mining, automations can understand and extract the important information (like sentiment, tone, and intent) from a customer email or chat log. While also understanding if the message had come to the right place to begin with. With generative AI, that same automation can then write a response to the customer, notify management of a potential outcome, and update core systems of record. That’s true AI-powered automation.

We’ve reached a stage where employees can comfortably hand off important, customer-facing work to integrated AI-powered automation systems. Businesses can now solve capacity and productivity challenges that have plagued the economy for decades. Solving those issues means gaining new operational capacity to upscale their operations, avoiding costs associated with a growing workforce, and immediately starting to reduce operational expenditure.

Even before ChatGPT landed on the scene, sports betting leader Flutter improved customer service through automation powered by generative AI. Flutter was able to predict incoming customer issues and service them end-to-end—containing costs, enhancing service levels, and boosting their net promoter score by more than 40 points. Flutter's Head of AI & Automation Strategy Oonagh Phelan took to the stage at FORWARD 5 to tell the story, which you can see in our Best Bits from FORWARD 5.

With AI-powered automation embedded as part of the team, your employees are free to think, create, and focus on the most complex business challenges and important customers. In my next blog post, I’ll provide concrete guidelines for embedding AI-powered automation in your enterprise. In the meantime, you’ll find great advice on operationalizing AI for automation in our on-demand sessions from UiPath AI Summit.