Generative AI strategy dilemma: Buy, build, or partner?

Perhaps the most exciting aspect of cultivating an AI strategy is choosing use cases to bring to life. This is proving true for generative AI, whose ability to create image, text, and video content from natural language prompts has organizations scrambling to capitalize on the nascent technology.

To that end, you IT leaders are grappling with some critical questions as they pursue GenAI application development.

What model(s) do you choose? What infrastructure do you run it on and where? Will you have access to enough GPUs to power your solutions? Can you deliver your applications on time and on budget? 

The answers will vary per business, of course. Yet IT organizations are also mulling the classic strategic questions about whether to build, buy, or partner as they craft their GenAI applications and services. New data suggests a mixed-bag approach.

Seventy-nine percent of organizations surveyed by KPMG are buying or leasing technologies (50%) or building, buying and partnering (29%). Another 12% of organizations are building GenAI solutions themselves, citing cost savings, customization requirements, and IP protection. 

As with most things, a happy medium is key.

“As businesses continue to explore the possibilities of GenAI, finding the right balance between building and buying solutions will be key to securing a competitive edge through better performance, higher quality, and enhanced customer loyalty,” said Todd Lohr, principal, U.S. technology consulting leader, KPMG.

From use cases to tools and techniques

What this suggests is that IT departments still need plenty of help in building and deploying solutions that may incorporate commercially available, partner-led, and internally developed technologies. 

For many of you, this is the white-knuckle time; the wrong decision can set your GenAI strategy back months. Naturally, you’ll consider the scope of your use cases, including what architecture, processes and tools will help you achieve the outcomes you seek.

Maybe your use cases include a digital assistant that retrieves relevant information about your organization, or your company’s products and services. Perhaps it’s picking an AI copilot to help programmers generate code in Python or other languages. Or maybe it entails building a tool to help staff navigate a digital twin from within the digital twin. Maybe it’s a mix of the above.

Whether the use cases are straightforward or meta, such initiatives hold great potential for helping your organization leapfrog competitors.  

When you’ve identified your use cases (and prepared your data) it’s time to choose a language model that will feed your solution the data it requires to be actionable is a critical decision. 

You’ll want to pick a pre-trained model, preferably one that also features some fine-tuning and one that you can run in your datacenter or at the edge if necessary. This will help you control such variables as performance, security, and costs. 

Meta’s Llama open-source LLM makes for a solid choice; enterprises such as Goldman Sachs, AT&T, and Accenture use Llama for customer service, code generation, and document reviews. 

Of course, you’ll only get so much mileage out of pre-trained models, which are trained on publicly available online data. You’ll want to combine the pre-trained model with retrieval augmented generation (RAG), a popular technique for generating content that includes data specific to your organization. 

Such output is critical for summarizing information, retrieving relevant documents, as well as analyzing data that helps inform collateral for marketing, sales, product development, and other business functions. It also breaks down the knowledge siloes that have long plagued enterprises.

Picking trusted partners

Identifying use cases, preparing your data, and choosing models and infrastructure are critical to the success of your AI strategy. Your C-suite peers may be banking on it, as executives surveyed by KPMG cited revenue growth as their top driver for GenAI investment. 

Whether you choose to buy, build, partner, or embrace all three approaches to launch GenAI applications and services, the journey starts with a cohesive strategy and the right use cases that can help your organization grab the competitive edge it seeks. 

Yet GenAI technologies and processes are new to organizations, many of which lack the resources to execute these steps alone. This is why it’s prudent to pick partners that can help steer you through the curves, from bringing AI to your data and choosing the right infrastructure to leveraging the growing AI ecosystem and bringing professional services to bear on your chosen use cases.

The right partner can mark the difference between solutions that rule the day and those that are just DOA. Ask yourself: Are you confident in your approach to helping AI flourish across your organization?

Learn more about the Dell AI Factory.