Generative AI for Business

Before we dive into the benefits of generative AI for the enterprise, it’s important to understand the key terms in this field:

• Natural Language Processing (NLP): A branch of AI that enables a computer to understand, analyze, and generate human language.
• Transformer: A neural network architecture (introduced by Google in 2017) that can train significantly larger models on ever larger datasets than before. It allowed the emergence of recent LLMs to process and learn from sequences of words, recognizing patterns and relationships within the text.
• Large Language Model (LLM): A deep learning model specialized in text generation. Today almost all LLMs are based on the transformer architecture (with variations from the original). As the name indicates, the real revolution is in the scale of these models, the very large number of parameters (typically in the 10s to 100s of billions), and the very large corpus of text used to train them.
  • Examples of commercial / closed-source models: OpenAI’s GPT3/3.5/4, GCP’s PaLM, AWS’s Bedrock, Anthropic’s Claude, etc.
  • Examples of open-source models: Google FLAN-T5 & FLAN-UL2, MPT-7B, Falcon, etc.
• Pre-trained Model / Foundational Model: An LLM trained from scratch in a self-supervised way (without labels) on a very large dataset and made available for others to use. Foundational models are not intended to be used directly and need to be fine-tuned for specific tasks.
• Fine-tuning: The process of taking a pre-trained model and adapting it to a specific task (summarization, text generation, question-answering, etc.) by training it in a supervised way (with labels). Fine-tuning requires much less data and compute power than the original pre-training. Well fine-tuned models often outperform much larger models.
• Hallucination: A situation where the LLM generates a wrong output; the root cause is that the LLM uses its internal “knowledge” (what it has been trained on) and that knowledge is not applicable to the user query.
• Multi-modal AI: a type of AI with the ability to process and understand inputs from different types of inputs such as text, speech, images, and videos.
• Retrieval Model: a system (typically a Transformer) used to retrieve data from a source of information. Combining retrieval models with large language models partially addresses the hallucination problem by anchoring the LLM in a known corpus of data.
• Vector Store: a type of data store specialized in managing vectorial representations of documents (known as embeddings), those stores are optimized to efficiently find nearest neighbors (according to various distance metrics), they are central architectural pieces of a Generative AI platform.
• Enterprise AI: Providing predictive insights to expert users for improving performance of enterprise processes and systems. (See What Is Enterprise AI?)
• Generative AI for Enterprise Search – Leveraging generative AI’s intuitive search bar to access the predictive insights and underlying data of enterprise AI applications, improving decision-making for all enterprise users.

The ability for generative AI to create high-quality, contextually relevant content—text, images, videos—in a fraction of the time it takes today, is transformational for a wide swath of businesses and specific functions. Generative AI applications and use cases span practically all industries and organizations across manufacturing, healthcare, energy, retail, transportation, government, financial services, and so on.

With generative AI, marketers can rapidly create a broader set of personalized campaign content without adding more writers; financial analysts can produce granular custom reports for executives in minutes. Such advances will lead to dramatic cost savings, better customer experiences, and increase sales velocity. Those are just a few of the potential advantages.

Core to transforming a business is the positive impact generative AI can have on the enterprise search experience. Imagine using a search engine to access exactly what you need within your business, making it easy for users to access the most pertinent information, portions of reports, and predictive analytics from your enterprise data and external systems. By making data, analytics, and predictions broadly available across an organization through an intuitive search bar—and not just to the data analysts in the company— generative AI can vastly improve decision-making at every level in the organization. Suddenly, people throughout the ranks of an enterprise can take advantage of this powerful AI technology, boosting efficiency, productivity and, importantly, the ability to plan.

Take a machine operator as an example. Machinery operators typically monitor equipment performance and manufacturing conditions at a control board. They are responsible for triaging alarms, responding to urgent issues, and ensuring that operations safely and reliably meet production targets and quality specifications. It’s a demanding role. As a result, operators do not have time to read detailed manuals or aggregate information across systems to identify trends in performance. Furthermore, many manufacturers face an aging workforce, where deep expertise is leaving the organization as operators retire.

Generative AI poses a unique opportunity to overcome these challenges. A large language model (LLM) can be trained on a corpus of enterprise data – such as historical machine failures, work order logs, inspections, production performance, and OEM operating manuals – to synthesize information and make recommendations for less experienced operators.

While working directly from the control board, a machinery operator may ask a generative AI application: “The conveyor belt on production line A is broken. How do I fix it?” The Generative AI application will quickly return the exact troubleshooting steps from the equipment’s Standard Operating Procedure (SOP) document, along with additional commentary from recent work orders on the production line A conveyor belt.

With this generative AI application, lesser experienced operators instantly gain access to the knowledge and experience of the operators who came and learned before them – without requiring decades on the job. The AI application synthesizes all relevant information and makes it available in a useful and easy to grasp format, all from entering a simple prompt into the enterprise search bar. The result is an operator new to the job becomes more efficient, more effective, and can deliver better outcomes for the business.

Many business functions can benefit from applications of generative AI.

A significant breakthrough in generative AI came with the development of transformer models—neural networks that can learn the context of sequential data—to rapidly create high quality, comprehensive responses to user queries. Transformer models were first described in an important Google research paper in 2017 called “Attention Is All You Need,” and they have been advancing ever since.

The capabilities of generative AI have both amazed and scared the public, some of which detracts from the opportunity for businesses with generative AI. Because the current versions of popular generative AI models, such as Chat GPT and Google Bard, have been trained on a wide body of publicly available data, they lack the nuances required for specific domains, which is where domain models come in.

For a manufacturer or health care service to take advantage of generative AI, they need a system that generates content—reports, analysis, etc. — based on data relevant to their domains, including such data sets as terms specific to a company. Most of an enterprise’s specific data is proprietary, sensitive, and sits behind a firewall. As such, applying generative AI to the enterprise requires domain models.

Domain models combine the fluency and comprehensibility of emerging transformers with the domain-specific data and knowledge of an enterprise. With access to the corpus of data, best practices, and policies specific to an enterprise, generative AI models can, for example, answer inventory-related search queries for supply chain analysts, create brand-compliant graphics for marketing managers, and identify potential compliance exceptions in new contracts.

Domain models will generate the greatest value for enterprises by finding contextually relevant and policy-compliant responses to user queries. Sophisticated domain models have the greatest value potential for enterprises by not just tapping into existing enterprise data, but by calculating metrics and analyzing that data to produce value-added content for enterprise users.

Supply chain executives trying to identify how well their company is fulfilling customer orders will be less interested in looking at each order’s fulfillment history. Rather, they’d want to see On-Time-In-Full (OTIF) history by product line, customer, or region. The ability of domain models not just to find and regurgitate data, but to combine data and create business-relevant metrics analysis that are predictive in nature is where generative AI models can become invaluable to enterprise users.

Domain models also build upon existing analytical frameworks and business logic to find correlations between data, identifying cause-and-effect and generating analytic-backed insights. In the above example of a supply chain executive investigating service levels, domain models will not just produce an OTIF report, but also an analysis of potential drivers of poor OTIF, caused perhaps by a specific transportation provider or linked to the timing of orders relative to the planning cycle of the company.

Domain models will be able to support enterprise users not just in analyzing the past but in taking proactive actions by using predictive insights. The supply chain executive in the above scenario will be supported by their domain generative model by identifying the future orders with the highest risk of delays along with explanations of the drivers of each order’s risks.

Enterprises leveraging domain generative models will capture outsized value not just by effortlessly generating insights for their users, but by also giving them the ability to act on those insights. By integrating generative AI capabilities with personal productivity tools and enterprise applications, companies will give their users the ability to:

• Take personal actions such as generating reports or setting alerts
• Initiate collaborative actions by kicking off a new project or create the agenda of a group meeting
• Kickoff workflows in enterprise applications such as accelerating supply orders, creating support tickets, or generating a customer discount.

Let’s get back to our supply chain executive investigating customer delivery fulfillment. Using domain generative models, the executive can identify past track records and risks to future deliveries, as well as making changes to production schedules and transportation arrangements to expedite critical orders at risk of delay.

Organizations looking to adopt generative AI technologies for enterprise use cases stand to gain a major productivity boost. Yet as with many leading-edge technologies, several factors exist that business leaders should consider when choosing a generative AI technology partner.

Data Availability and Quality
Generative AI domain models require large amounts of data to be trained effectively. Difficulty in obtaining or collecting relevant data, particularly for specialized or rare situations can limit the scope and effectiveness of the model. Furthermore, limited availability of high-quality data can lead to poor performance and accuracy of the model. Consequently, enterprises should consider both the availability and quality of data to train and operate generative AI systems.

Security and Privacy Risks
In addition to availability and quality of data, unauthorized use of sensitive data for training or operating the model can also expose the enterprise to security and privacy risks. Vulnerabilities in the model architecture could be exploited by hackers to gain access to sensitive data and cause harm to individuals and enterprises. Moreover, lack of transparency and accountability in data handling and use could lead to potential regulatory noncompliance.

Intellectual Property Risks
With public-facing generative AI systems, enterprises may be exposed to intellectual property infringement risks. By design, public generative AI systems are trained on internet data. Consequently, generated content similar to existing works that are protected by intellectual property laws could lead to legal actions and potential financial damages. Ongoing debate around the ownership and authorship of generated content from public generative AI systems could lead to legal disputes and challenges.

Bias and Inaccuracy
As the generative AI model is trained on available data, bias and inaccuracy in the training data can be perpetuated or even amplified in the output of the model, leading to discriminatory, legally risky, or unsafe outcomes. Furthermore, inaccurate answers due to outdated or irrelevant data or model hallucination can lead to unintended consequences and exacerbate an existing issue.

Domain Models
With a customized generative AI model trained on the enterprise’s own data and deployed on top of enterprise data and applications, enterprises can reap the benefits of generative AI systems while minimizing risks.

First, many enterprises likely have a wealth of high-quality data that is subject to rigorous data management practices and quality checks. Secondly, enterprise data is often subject to access controls and security measures, which can help prevent unauthorized access to sensitive data. Moreover, as enterprise data is typically owned by the organization, generating content based on enterprise’s own data eliminates intellectual property rights risks. Diverse, representative, and high-quality enterprise data can also help minimize bias in model outputs. Finally, while limiting hallucinations is an ongoing research area, generative AI systems that have a built-in feedback loop with human evaluation, such as the C3 Generative AI Product Suite, will be able to eliminate hallucination and inaccuracy over time.

Advances in technology are once again promising to transform business and society – and this time it’s happening at hyper speed. Generative AI presents a new epoch of human efficiency and effectiveness, affecting society and industry in ways we have yet to understand. When we look back, as discussed above, this moment will be as groundbreaking as rollout of the personal computer and introduction of the Apple iPhone.

With generative AI solutions, business leaders will drive their organizations to previously unattainable and unthinkable speed and scale of execution. Companies will automate and optimize business processes, generate novel insights, and deliver new customer experiences faster than ever before.

While the general public has become enraptured by charming and relatable examples – from writing puns to creating a photo of an astronaut riding a horse – business leaders need to focus on well-scoped problems that generate economic value.

Developing and implementing generative AI technology for business transformation requires a thoughtful and deliberate approach. First, business leaders must identify tractable use cases where the solution will be useful, drive value, and where stakeholders agree to measurable success metrics.

Tractable use cases include automating repetitive tasks, or synthesizing insights from unstructured data and documentation. This could be achieved through streamlining high-volume tasks to save time, or unlocking once all-but-impossible to access information deep in documentation and disparate data sources. Success metrics will measure the business impact, and could be tied to economic benefit, customer service, sustainability outcomes, or business efficiencies.

As you can see, all this is leading to major changes within a given enterprise. To scale these technologies and solutions, leaders will need to evaluate existing workflows of people who already perform job functions generative AI can transform. Successful enterprise implementations of generative AI will embed solutions into existing customer and employee workflows, rather than serve as standalone tools.

The simplicity of using generative AI—for a consumer or a business user—should make adoption a non-issue. The possibilities really are limitless, and enterprise adoption should occur in months rather than years, ultimately making generative AI a moment in technology history whose significance will soon be undeniable – for enterprises everywhere.