GenAI: From Concept to Production

It's now widely recognized that Generative AI (GenAI) will profoundly impact both professional and personal spheres of human beings. Any field involving content—whether it's text, images, video, music, voice, graphics, or entire virtual worlds—stands to be affected directly or indirectly by this technology. In short, GenAI models can learn patterns via large datasets and create outputs that mimic the input data's style and structure. This innovation has drastically changed our approach to content creation, communication, digital experiences, and insights. While 2023 was about exploration, 2024 marks the beginning of GenAI initiatives, with 2025 and beyond expected to shift from experiment to execution.

Over the past 12-18 months, many organizations have focused on identifying and executing proof of concepts (POCs) with GenAI. However, many use cases have yet to reach the stage of production deployment. According to Gartner research, 70% of Global 2000 organizations will leverage Gen AI to enhance customer experience and provide valuable insights. In 2024, approximately 6.5% of the functional budget is expected to be allocated to various Gen AI activities, and 74% of enterprise leaders (business and IT) believe their organizations are well-positioned to capitalize on GenAI's potential.

Enterprises embarking on a GenAI journey have put significant effort into identifying the best use cases for POCs, ranging from AI-first/everywhere to productivity-driven initiatives. Gartner anticipates that by 2026, 8% of the workforce will be replaced by GenAI, and new business models will generate significant new revenue streams for organizations.

As the initial use cases for GenAI take shape, business & IT leaders must focus on measurement and governance to ensure these cases deliver tangible business value. Organizations must also implement robust controls to deploy AI responsibly and manage associated risks, informing a longer-term roadmap. Adopting GenAI will redefine the organizational structure, including people, processes, and technology, through a GenAI framework and methodology.

Before moving further, a simple explanation of a neural network, large language models (LLM), and small language models (SLMs).

Neural networks are a class of machine learning models inspired by the human brain's structure and function. They consist of interconnected layers of nodes (also called neurons), where each node performs a simple computation and has its weight and bias. When combined, these nodes enable the network to learn complex patterns and relationships within data, having an input, processing, and output layer with the ability to weigh each node.

Large language models (LLM) are artificial intelligence models for natural language processing (NLP) designed to understand, generate, and manipulate human language in a coherent and contextually relevant manner. These are trained on vast amounts of text data and use deep learning techniques. Some well-known examples of LLMs include OpenAI's GPT series (like GPT-3 and GPT-4), Google's BERT, and Meta's LLaMA.

Small or Specialized Language Models (SLMs) are a subset of language models with fewer parameters and are less complex than large language models (LLMs). They are typically designed for more specific or lightweight tasks, making them easier to deploy and less resource-intensive.

Based on early experiences, the journey toward everyday GenAI will require iterative and transparent business decisions that are methodical and repeatable. Here are six key steps or pillars to consider when moving GenAI initiatives from concept to production:

1. Data and Vector Stores: The quality of GenAI outcomes is directly tied to the data input into LLM models. Organizations possess vast amounts of structured and unstructured data, including PDFs, Excel files, PowerPoint presentations, chat logs, product videos, and customer service interactions. While organizations have long prioritized the quality of structured data for analytics and insights, the ability to ingest large volumes of unstructured data presents new challenges in maintaining data quality and access. The demand for better data will drive the development of new data sources, including synthetic data (artificially generated). Gartner projects that by 2030, synthetic data will dominate real data in AI models and make business & IT leaders rethink data catalog, classification, and usage with strong content governance and critical in-house capabilities of data engineering.

2. GenAI Models and DevSecOps: Over the past 18 months, we've seen rapid innovation and the emergence of various model variants with improved learning of complex patterns from vast datasets. Training LLMs require significant computing power and engineering investment, and currently, AI frameworks & libraries are optimized for Nvidia's CUDA platform. Given the pace of innovation, business & IT leaders need an agile and flexible framework to adopt plug-and-play LLM models based on use cases and evolving technologies, including security considerations. This ongoing process will involve model accuracy, performance, and cost trade-offs. Organizations will need skill sets to compare open-source versus commercial models and integrate APIs for deployment and regular monitoring as ongoing enterprise activities.

3. Audit and Compliance: When things go wrong, organizations need a plan for diagnosis and recovery, which is essential for risk management. Publicly listed and private enterprises are subject to internal and external audits as part of compliance, governed by regulations like SOX, GDPR, and other regional laws. While Gen AI can assist with tasks like drafting communication, enhancing customer experiences, or reviewing documents for policy violations, its use in financial modeling or other high-risk areas will be limited due to the potential impact on intellectual property (IP), brand reputation, trust, and ethical/legal risks. Business leaders must carefully manage Gen AI use cases to avoid regulatory surprises and audit challenges. Until then, critical business decisions will still rely on human judgment.

4. Talent and Human Capital: The human element in AI adoption focuses on how AI will augment, not replace, human roles. Training is crucial, not just in the technical aspects of using GenAI models but also in developing new skills and roles. For example, prompt engineering will likely become a standard practice for knowledge workers. By 2027, 60% of labor hours across most jobs will be disrupted, driving an organization's need to refresh talent and human capital. Employees must be experts in their fields and acquire new skills to thrive in the GenAI era. This shift will also spark discussions around intelligence quotient (IQ) and emotional intelligence (EQ), emphasizing the need to balance engineering and decision-making.?

5. Organizational Change Management: GenAI's ability to create realistic content raises ethical concerns, including deep fakes, misinformation, and the impact on creative industries. Change management within organizations is equally important as there is widespread anxiety about how GenAI will affect livelihoods, and companies must address these concerns by emphasizing how GenAI can empower human potential rather than replace it. This will be an ongoing process, with communication, upskilling, AI adoption, and training being essential components of the Gen AI journey.

6. Governance: Organizations familiar with business or digital transformation understand the importance of a center of excellence (COE) for developing and managing technology capabilities. In the GenAI era, business & IT leadership teams require a new approach, including a control tower with cross-functional participation to develop strategies and align resources and budgets accordingly. Given the broad range of GenAI use cases, centralized risk management and the challenge of maximizing business benefits with limited talent will be critical. An AI control tower approach allows for systematic AI implementation and maturity while effectively managing risks and increasing business value. Organizations shouldn't wait for regulatory bodies to catch up.

Generative AI represents a significant leap in artificial intelligence, offering exciting opportunities and challenging ethical questions. Its applications span various fields, from customer service to business operations, impacting all practical aspects of business and technology uses. As GenAI continues to evolve, ongoing research will aim to improve these technologies' quality, efficiency, and ethical use. Future advancements may lead to even more sophisticated models capable of generating complex and nuanced content across various use cases while maintaining a balance between IQ and EQ. Unlike previous digital and cloud transformations, the success of GenAI initiatives will depend on multiple pillars, many of which will be driven by new technology layers for monitoring and governance. This is just the beginning; with each new technology, new responsibilities emerge!