Leveraging Generative AI for
Enterprise Automation

What is Generative AI?

Generative AI is a new frontier of AI. AI has long been a tool for analyzing and understanding data, identifying patterns, detecting anomalies, and making predictions. Generative AI’s status as the latest “disruptor” is unique in that it does things other AI technologies don’t.

Generative AI is a broad term for these technologies. One specific example is ChatGPT, which falls under Natural Language Processing (NLP) and Natural Language Generation (NLG) combined with the power of Large Language Models (LLMs).

Though Generative AI is based on language, it can do more than generate text. You can use it for images, NLP text generation, music composition, video creation, and more. Understanding how it differs from other forms of AI is crucial when considering how to apply it in your enterprise organization.

Continued research into Generative AI will lead to further evolution of its capabilities and reveal its limitations. The possibilities have enterprises optimistic and looking toward the future.

Enterprises and Leaders Are Optimistic About Generative AI’s Potential

The conversation about Generative AI is getting louder every day, and many businesses and their leadership teams are at least in the discovery phases regarding its use. 

A Gartner poll revealed key insights into where things are heading:

• 70 percent of organizations are actively exploring Generative AI.
• 68 percent of executives believe the value of Generative AI outweighs the risks.
• 19 percent of enterprises are in pilot or production mode for Generative AI use cases.
• 45 percent of businesses are investing more in AI to improve customer experiences, grow revenue, optimize costs, and strengthen business continuity.

From these data points, we can draw some conclusions:

• Enterprises recognize the potential value of Generative AI and are moving forward with minimal concerns over risks. Exploration is expanding, but because it is a new technology, its optimal use cases may still be unqualified. Those in pilot or production mode have defined the end goals they expect from adopting Generative AI. However, they may still face challenges when scaling it across the organization. 
• Investments in Generative AI cover many different business areas—customer relations and retention, sales, operations, IT, and more. These areas involve a variety of stakeholders with diverse perspectives, so defining how they expect Generative AI to reach objectives will require collaboration, compromise, and a holistic view of its value. 
• Currently, the most usage of the technology in organizations is at the individual level; it has been a boon for efficiency for the individual contributor. The enterprise, however, must plan its rollout strategically to support broader automation initiatives. In doing so, businesses must look at the opportunities and challenges associated with Generative AI. 

The Opportunities of Enterprise Generative AI

This smart and sophisticated technology can drive many benefits for an organization, including in these areas:

Supercharged Content Creation

Generative AI accelerates content creation. It can generate a nearly infinite range of content across a wide variety of formats, including text, images, video, and more. By working from accurate and relevant prompts—and further honing those prompts to shape outputs after the initial content results—you can scale content creation within an enterprise. 

Boosted Employee Productivity

Another substantial advantage of Generative AI in business is its capacity to be a super-powered AI assistant. Employees can be significantly more productive, saving them considerable time in their workdays to focus on meaningful work. 

A McKinsey report labeled Generative AI “the next productivity frontier.” Its insights concerning productivity include:

• Generative AI’s impact on productivity could add trillions of dollars in value to the global economy ($2.6-$4.4 trillion).
• Generative AI could change the anatomy of work and be a true companion or augmenter for workers when automating processes. 
• Labor productivity growth could increase from 0.1-0.6 percent annually through 2040 with the support of Generative AI. Combining it with other technologies could add up to 3.3 percent annually. 

Integration with Other Technologies 

In addition to profoundly affecting productivity, Generative AI and other technologies could expand the capabilities and impact of automation with plugins. For example, ChatGPT with a LinkedIn plugin could produce content for posts on the social media site. 

Another integration use case would be Generative AI and Machine Learning algorithms working together to unlock more analytics insights. Enterprises have massive amounts of data they want to analyze and turn into actionable insights. Algorithms are great at performing tasks related to optimization, such as predictive modeling. These techniques focus on identifying patterns and making predictions, but they can’t generate new ones—which is what GenAI can add to the field. 

Generative AI can recognize patterns and then deliver new content, mimicking the data from which it was trained. The progression here comes from knowing what something is and being able to create something new that still follows those same patterns. 

The Challenges of Enterprise Generative AI

Generative AI presents exciting opportunities for organizations. However, it is still an immature technology, and as such, some elements of it give the industry pause.

Human Intervention Is Still Necessary

At an enterprise level, companies have concerns about what Generative AI will create, especially for public-facing content or anything related to legal, compliance, or regulatory materials. As a result, there’s still a great need for human intervention in the process. A review of content for accuracy and appropriateness is still necessary in order to build trust and prevent any costly errors.

Early feedback about GenAI in the enterprise space has revealed limitations that signal the need for human intervention for the foreseeable future, including:

• Generative AI can produce confident statements that don’t align with training data.
• The data it consumes has inherited bias, similar to any learning model.
• Generative AI lacks human reasoning, which is critical in decision-making. 
• The current context window is narrow with only a few thousand words for the input and the output. 

It comes down to Generative AI not being a 100 percent trustworthy or reliable source. However, there are ways to navigate this with a human review of the output. 

Ungoverned Internal Data and Intellectual Property Sharing Are of Concern

For an enterprise to reap the rewards of Generative AI, internal data and intellectual property would need to be part of the inputs. There’s genuine fear about sharing it with these Generative AI engines in an ungoverned manner. 

For example, a developer could provide ChatGPT with a small piece of custom code to help identify a bug. By doing so, the employee may not realize this code is now part of the LLM training dataset. That means the code is now part of the public domain, and competitors can find it—or, worse, cybercriminals could use it to discover and take advantage of vulnerabilities in software. 

What Generative AI Produces Isn’t Exactly ‘New’

Most Generative AI tools run on LLMs. Their training requires terabytes of existing data. Therefore, the output from the tool isn’t entirely original, raising valid concerns about plagiarism and copyrights. There has already been plenty of backlash in higher education about the plagiarism aspect in particular, which has led to the use of tools that can detect ChatGPT content. 

For businesses, the problem is a bit more complicated. Legally speaking, there are arguments on both sides involving AI ethics considerations and no existing legal precedent. Avoiding these legal entrapments hinges on the content not completely regurgitating what already exists. 

Future Regulatory Rules Could Halt Enterprise-Wide Generative AI Adoption

Another potential problem is the possibility of industry-specific or governmental regulations surrounding Generative AI coming into effect. Highly regulated verticals such as healthcare and finance will have to be especially aware of these rules. Otherwise, it could prevent initiatives from moving forward. Proper research is a wise precaution before making substantial investments in these programs. 

Despite these risks, Generative AI could be a highly effective productivity tool with the proper enterprise guardrails in place. 

You now know the opportunities and challenges of Generative AI for enterprise use. Let’s see how Generative AI for the enterprise intersects with enterprise automation. 

Enterprise automation includes the three Es of value:

• Effectiveness: Effective automation should do the “right” things better and is critical for making consistent data-driven decisions. Effective automation allows an enterprise to quickly focus resources on the right objectives and initiatives. 
• Efficiency: Efficiency in automation looks like doing the right things with the least amount of effort. It’s about running leaner and faster. Realizing economies of scale through automation leads to process improvement, standardization, and cost optimization.
• Experience: Automation should improve the experience for your internal users or customers. Experience can be hard to measure and lag behind the measures involved with effectiveness and efficiency. By connecting efficiency and effectiveness, you can drive adoption and incorporate customer- or user-centricity. 

Achieving enterprise automation can be challenging for many reasons, which has companies seeking new tools and technologies for getting there. Generative AI is becoming a major contender in this respect. 

Generative AI’s impact on automation will follow one or more of these trails: creating better experiences for employees or customers, creating greater efficiency by developing content for workers, or creating a more effective process that serves all stakeholders. 

Generative AI becomes a building block for enterprise automation because it can play a role in developing content humans would have had to manage independently. However, for the enterprise to take advantage of the technology, it must realize there is more to the story. An organization must understand the subsets of Generative AI, create parameters, and keep humans in the loop. 

The Four Buckets of Generative AI

The adoption processes of Generative AI to advance and support enterprise automation fall into four different buckets. Each has its own value within an organization. 

Individual/Personal Automation

This type of Generative AI is at the individual contributor level. Someone gives ChatGPT a prompt—that is, a chat command—to create an article or image. For example, they may ask ChatGPT to write a business plan for moving into a new market. The results they get will depend on the quality of both their prompt and the content and datasets from which the technology learned. 

Stage One Automation

In this concept, natural language creation templates and guardrails are in place to guide the content creation process. Moderating and monitoring how the enterprise is using the generative tools is possible via human intervention. 

Locally Deployed LLMs

In this case, the data is internal, and companies can use these models to generate the content required to make the data more actionable. Locally deployed LLMs eliminate the risk related to sending data outside of any organization, which makes it well suited for the enterprise context. 

Public-Facing LLMs

Some datasets are available to anyone. They are anonymized but can be useful for organizations aiming to uncover insights. This could benefit many different stakeholders if it is truly open data. In this model, the risk regarding your data is at the highest possible degree. 

With these four buckets, you can realize benefits for both individuals and industries

Healthcare Virtual Assistants

Virtual assistants can be helpful in telehealth and for clinical decision support. They can deliver real-time, evidence-based recommendations for providers to improve patient outcomes. 

For example, ChatGPT could suggest specific treatment plans for someone newly diagnosed with diabetes. It would use anonymized patient history as context for flagging any potential drug interactions or other chronic diseases that may impact a care plan. This use case assumes proper care occurred during the process to protect confidentiality and protected healthcare information (PHI).

E-Commerce Self-Service

Generative AI-powered chatbots can provide immediate, personalized replies to customers’ complex questions. It’s a big step beyond traditional chatbots that only respond based on keywords. Improving the effectiveness and quality of these digital customer interactions enables human agents to focus on the most challenging inquiries—plus, a quick resolution helps build customer loyalty and satisfaction. 

Coding Assistants Accelerate Development for SaaS Companies

The software industry can get a boost in development and new releases when Generative AI plays the role of coding assistant. 

Software products are always in a state of continuous improvement with roadmaps for adding features, fixing issues, and improving the user experience. Giving developers a digital helping hand could revolutionize the process, reducing time to market and costs. Research on this topic shows positive results. A study by Bain & Company reports that 75 percent of generative AI coding assistant adopters say it has met or exceeded their expectations. Additionally, a study involving Microsoft’s GitHub Copilot, an at-scale AI tool that integrates with ChatGPT, revealed that developers using it completed tasks nearly 56 percent faster than those who didn’t. 

In these examples, you can see the value that Generative AI and enterprise automation can deliver when human intelligence works alongside artificial intelligence. 

Prompt Engineering Is Key to Realizing Value

Prompt engineering is a critical element of getting value from Generative AI. Essentially, it’s the ability to deliver the right inputs to the model or build upon those inputs using powerful contextual commands. This phase of Generative AI is not unlike the early days of search engines when queries had to be precise to get the most relevant results. 

How this works at the enterprise level goes back to the human-in-the-loop approach. For example, developers can issue cultivated prompts to discover bugs in their code or ask for examples of how to use a complex function. Developers are still vital to this process because they must review and validate the code. With the correct use of Generative AI, nearly any user can multiply their work output. 

Healthcare

Generative AI can have a significant impact on medical imaging review and diagnosis. Using these tools, you can reduce the burden of solely human reviewers, which can accelerate the process. Adding Machine Learning to the mix improves the process even more. 

Algorithms have already been useful in this use case. Generative AI training consists of images of healthy humans and those of abnormalities. It provides another layer of oversight in finding things in images that may be missed or overlooked. As a result, first-pass diagnosis accuracy could improve. 

Of course, “improvement” is a vague goal. To measure the value Generative AI adds to your organization, you’ll want to set more specific goals with clear target metrics. In this case, a starting goal could be to improve first-pass diagnosis accuracy by 20 percent over the first year.  

Finance

Two of the biggest concerns for financial institutions are fraud detection and risk assessment. AI’s ability to do this well depends greatly on the quality of its training data. In many cases, machine learning algorithms rely on a strong fraud signal, making them ineffective at scale. In these models, there are only two classifications: fraud or not fraud. Financial planning and analysis teams can leverage large language models to enable more scenarios in their analysis process to help improve effectiveness. 

Generative AI can expand training data because its output can continue indefinitely. It simulates synthetic data based on real data. Generative AI can use existing samples and produce new ones. It does this through what experts call “hallucinogenic outputs.” This is the phenomenon of ChatGPT providing inaccurate results, which is worrisome for many applications—but not fraud identification. It can create unique fraud patterns, which boosts the performance of fraud detection. 

One target outcome could be decreasing fraud-related expenses by 30 percent in the next fiscal year. 

Manufacturing 

In manufacturing, a great deal of time and resources flow into product design and optimization. Timelines to get these to market can be long and financially draining. Design teams are in constant iterative cycles and don’t have the capacity to consider alternatives. Generative AI can shake up this traditional route and add value throughout the product development journey, including:

• Testing new shapes and geometries, which could drive innovation.
• Improving the engineering process by providing more visibility regarding product performance, thereby reducing waste and timelines. 
• Assessing the geometries of the product to understand its manufacturability.
• Unlocking incremental value and greater margins in the product improvement stages. 

By leveraging Generative AI, a manufacturer could define a goal to decrease concept-to-floor time by 15 percent in the upcoming quarter. 

Customer Service

Customer service is a high-volume process that requires quick and accurate responses. The quality of a business’s customer support is often a major factor in whether or not customers stay loyal to the business. Human agents need their own support in the form of virtual assistants and chatbots to meet these demands. 

Traditional, rules-based chatbot automation has been a point of frustration for consumers and companies because it is limited in its ability to offer relevant support. Generative AI applied to chatbots is a different experience because it is conversational and much more expansive in terms of what information it can provide to resolve issues. Of course, Generative AI will need a high volume of data on products, policies, and FAQs to work effectively. 

Generative AI can also structure support tickets for quicker responses by summarizing and filling in information. Achieving success will require you to optimize the training dataset and implement fact-checking parameters. It becomes a virtual assistant for agents, who will then have more time to focus on the most complex issues and not feel overwhelmed by growing queues. Applying Generative AI to customer service could begin with the goal of improving service ticket outcomes by 20 percent over a six-month period. 

Energy

The energy sector can benefit from Generative AI in the area of maintenance. Regular upkeep of assets is critical in this industry. Any downtime equates to lost revenue and high costs. Energy companies have been adopting AI-powered automation in this use case for predictive maintenance and increasing the life of the equipment. Ensuring the performance of these assets can also improve safety. 

Generative AI can apply to maintenance by identifying valuable data via asset monitoring. Technicians can respond proactively to prevent equipment failure. High accuracy in predictions enables on-time maintenance reporting. Improving this by 50 percent is a good target objective.

To attain this value, enterprises in any industry must go through the steps of deploying Generative AI. 

Data Collection and Preprocessing 

You’ll begin this process by collecting and preprocessing data with these steps:

• Gather diverse and representative datasets that align with your desired outcome.
• Identify and remove errors or inconsistencies within the data.
• Convert the data to a common format.
• Add variation to existing data points to create new ones.

Model Selection and Training

The next step is choosing the architecture model. This is a critical decision because it will determine how a model learns and generates content. Common models include:

• Variational autoencoders (VAEs): A neural network for learning the distribution of a dataset. This works best for image- or text-generation tasks. 
• Generative adversarial networks (GANs): These are also neural networks that are good at generating realistic images. These work by training two neural networks against each other: a generator and a discriminator. The generator creates new content, and the discriminator determines if the context is authentic or fake. 
• Autoregressive models: These neural networks generate text or music. They predict the next word or note in a sequence. LLMs are autoregressive models. 

When you’ve selected the model, training can start with input data. The model learns to find patterns, distributions, and relationships in the data. During training, optimization of model parameters occurs via certain techniques, including:

• Gradient descent: This is an iterative algorithm that minimizes a loss function, which measures the difference between a model’s output and the ground truth data. It works by adjusting the parameters toward the negative gradient of the loss function.
• Backpropagation: This technique calculates the gradient of the loss function relating to the parameters of the model. It propagates the error from the output to the input layers of the model. 
• Regularization: This practice prevents overfitting, which happens when models absorb the noise in training data rather than its underlying patterns. It attaches a penalty to the loss function.

Evaluating Model Performance and Iteration

The next step is assessing and refining your models after training. You’ll do this by comparing performance to predefined benchmarks and metrics and evaluating the performance based on accuracy, precision and recall, quality, coherence, and realism. Based on your findings, you’ll iterate to improve all these calculations until you’re sure that it’s ready to integrate into existing systems. 

Deployment and Integration with Existing Systems

The Generative AI model is now ready to deploy to users via integration into applications, platforms, processes, and workflows. Continuous improvement is still the principle to focus on in this step. After integrating the model, you’ll want to gather user feedback to discern how it’s performing in the real world. 

Monitoring and Maintaining Generative AI Solutions

The final step is one that never ends: Continue tracking metrics to understand how outputs perform against expectations. Keeping an eye on performance allows you to make changes and adjustments as needed. 

Implementing Generative AI into your enterprise is the end goal—but you’ll need to be aware of the challenges ahead first.

AI suffered from inflated expectations that never came to fruition. Another blip happened every time a packaged application provider embedded AI into something new, focusing on fancy features instead of the three Es of efficiency, experience, and effectiveness. 

For Generative AI to avoid kicking off the next AI winter, enterprises need to experiment with the potential and focus on the outcomes they expect from Generative AI and automation. Additionally, enterprises must work out privacy, compliance, and security guidelines. To do this, you have to answer a few crucial questions. 

First, do you trust the data? The output from Generative AI is only as accurate as the data it consumes, and most organizations will want to leverage both public and private datasets to create competitive advantages.

Second, is the content safe to use (for various legal, copyright, or industry reasons)? Auto-generated, unrevised content opens an enterprise up to liability without checks and balances such as human review, content moderation, and data governance.

• Delivery of outputs that were not sourced from the data it was trained on—the hallucinogenic effect: Human review can check for inaccuracies or false assumptions here.
• Lack of understanding of AI explainability: Many decision makers may not fully understand what Generative AI is or its benefits in operations. As a result, they may be hesitant to invest resources in a technology they can’t explain.
• Inherent bias: Training data has its biases, so the outputs will as well.
• Absence of human reasoning or logic: Because Generative AI works only on the data it was trained on, it lacks the human ability to question the accuracy and quality of what it produces. 
• Data quality, completeness, and availability: Limits here will impact outputs.
• Ethical use and accountability: Generative AI raises ethical concerns relating to bias, fairness, and transparency. That’s why guardrails and parameters are critical in its deployment. 
• User acceptance and adoption: Some people will fear Generative AI because they think it could cost them their jobs. AI has and will continue to eliminate some jobs, but it also creates jobs and adds meaning to the work people do. Communication and education around the value of AI are key to employee adoption.
• Scalability and resource requirements: With any technology, there are challenges related to scaling. That’s the focus on implementation at the enterprise level, so this requires strategic planning. Having the expertise to do this is also a problem for many, but that resource doesn’t have to be internal; expert consultants can guide you.
• Security and privacy concerns: Generative AI requires large amounts of data to learn, and businesses may be hesitant to share their data with the model because of concerns about security and privacy.
• Implementation costs: Although many organizations are investing more in AI, some may think it’s cost-prohibitive. There are costs related to hardware, software, and training. You have to determine the ROI of these investments to determine whether it’s a sound initiative. 

It starts with a responsible approach called AI explainability. A decision maker or initiative leader needs to know why a model is outputting what it is. Until they have a greater understanding of this, they aren’t ready for enterprise deployment. 

To get your company to this point, you need support from a partner that specializes in AI and automation to look at all the opportunities and identify risks and potential. Plugging in Generative AI just because it’s the shiny new object won’t deliver efficiency, effectiveness, or experience.

Enterprises must devise a plan regarding how they will use this technology—and govern its use. This enormous task involves value stream management and mapping the desired outcomes from using Generative AI in enterprise automation. Process Mining and Process Discovery are pivotal during this process.

From there, you will determine what to automate. Is it business-rules-driven automation, workflow orchestration, or low-code applications? Generative AI can accelerate all these things. You’ll put your model into production and test it, which leads to refinement and continuous improvement. Until you see real-world results, automation-related benefits will remain theoretical, not practical.

Another area all stakeholders should watch concerns regulatory and legal frameworks. Earlier, we discussed the gray area of plagiarism and copyright infringement. As the technology grows in usage, there are bound to be legal cases that will arise and shape this area. Additionally, highly regulated industries with mandates for data collection and usage, such as healthcare, may see specific requirements around how they can use Generative AI.

Generative AI is already impacting the economy. Economic implications depend on the promise that this technology will add value to global economies because of productivity. The projections for this are in the trillions, and analysts believe these areas of business will realize the greatest effects: customer service, marketing and sales, software development, and R&D.

Finally, society’s acceptance of Generative AI will influence its future. There are those who see it as helpful and those who fear it. For human intelligence to perceive artificial intelligence as a threat is normal and natural, and an enterprise that aims to use Generative AI will need to consider this in its plan. After all, if people don’t use the technology, it won’t deliver value. Education, communication, and openness to employee feedback will be essential to the success of Generative AI in any organization. 

There is no turning back the clock on the AI era, so it's best to dive in and think about the implications and opportunities for the individual, the enterprise, and society.

As the vision of how organizations can use it becomes clearer, you can begin to work it into your enterprise’s automation framework with blueprinting and guidance. Getting the expertise you need to move forward will require a partner who knows all of the technology’s facets and implications.

Schedule a consultation with our experts today to plot your Generative AI and enterprise automation journey.