Knowing when to place critical bets on emerging technologies is a delicate dance for IT leaders pursuing digital transformation. Dither and you’ll miss an opportunity to leapfrog competitors. Rush and you’ll drain precious resources without a proper payoff.
As 2024 gets underway businesses are grappling with how to thread this needle with generative AI, which emerged last year as a productivity turbocharger. GenAI’s automation of content creation could boost global corporate profits by as much as $4.4 trillion, McKinsey Global Institute estimated. 1
Yet 37% of organizations reported hesitancy about adopting GenAI, citing concerns about data security and governance, technical complexity and implementation costs, according to a Dell survey. 2
Another stumbling block: Organizations with little to no experience undertaking AI projects may not know how to get started. Thirty-eight percent of organizations surveyed by O’Reilly Media 3 said that their companies have been working with AI for less than a year and remain in the early stages.
If you’re among the “early stages” cohort, this playbook 4 can help you navigate your GenAI journey.
Getting Started on Your GenAI Journey
Communicate the vision. Transformational initiatives require the orchestration of people, process and technology. Define how your organization will leverage AI, soliciting feedback from the C-suite, line of business leaders and other key stakeholders. Ask what stakeholders want to accomplish with regard to content creation and what is required to support them. Begin crafting a strategic vision, prioritizing use cases and planned investments, along with guiding principles for future projects. Regularly communicate how IT will support and protect the business as it deploys GenAI solutions.
Assess organizational readiness. GenAI use cases require prudent infrastructure planning and deployment. First, you’ll capture the “as-is” state of your environment to develop topology diagrams and document information on your technical systems. Then you’ll craft a “to-be” blueprint of what you need to support your strategic vision, including targeted capabilities, future IT architecture and talent required to facilitate the work. Secure buy-in from key stakeholders and regularly communicate progress with them.
Get your data house in order. GenAI requires high-quality data. Ensure that data is cleansed, consistent and centrally stored, ideally in a data lake with pipelines to integrate data into a model. Your data should be flexible enough to support a two-pronged approach that balances tactical deployments to test and learn with a long-term strategy that accounts for potential future use cases. Fast-track approaches require data preparation, including anonymizing, labeling and normalizing data across sources. For long term operations, you’ll institute guardrails for data governance, data quality, data integrity and data security.
Pick a Deployment Model. With your data cleaned and prepped you can consider deployment models. Creating private instances of pre-trained LLMs, such as Llama 2, can help you get up and running quickly. For more focused data needs you can leverage retrieval augmented generation (RAG) to augment your model with domain-specific information, which may also reduce hallucinations. Some organizations requiring specialized capabilities may wish to build and train their own model, although this requires significant tech resources. Ultimately, you’ll want to right-size your models—LLM vs. domain-specific vs. enterprise-specific—to meet your organization’s needs. Your deployment choice will depend on your use cases and the business outcomes you wish to derive from them.
Choose a Workload Location. Where you host your model is a significant consideration. Choose the environment that makes the most sense based on your business requirements and technical needs. Public cloud is a great choice to get your GenAI service up and running quickly, making it ideal for test-and-learn pilots. However, GenAI services can generate a lot of data, which becomes harder to move as the volume grows. If you’re planning to deploy a solution into production you should consider running it on-premises, which will allow you to customize your model and support it with hardware optimized to handle heavy compute and storage loads. Bringing AI to your data checks a lot of boxes with regard to control over security, governance, costs and energy efficiency. Maintaining complete control over your own infrastructure and data will afford you more peace of mind as you embark on this transformational journey.
Now you’re ready to build and deploy your GenAI service and pressure-test it with your organization’s employees. As you embark on this project, be sure to articulate best practices for prompt engineering, a necessary skill for front-line practitioners.
One More Thing…
Enthusiasm over GenAI deployment remains strong and it can be tempting to accelerate adoption. Just be sure you’re crafting your initiatives responsibly. After all, GenAI adoption has critical implications for your data, as well as your compliance, governance and security postures.
GenAI is here for the long haul. Following a few basic but critical principles can ensure your organization is well positioned to reap the rewards.
Learn more about dell.com/ai.
1 The Economic Potential of Generative AI, McKinsey Global Institute, June 2023.
2 Generative AI Pulse Survey, Dell Technologies, Sept. 2023
3 Generative AI in the Enterprise, O’Reilly Media, Sept. 2023
4 Generative AI is Here: Are You Ready, Dell Technologies, Oct. 2023