Buy, boost, or build? Choose your path to generative AI

What you’ll learn:

Generative artificial intelligence solutions are business-case-driven and address a company’s strategic objectives. Companies have three options for acquiring them:

• Buying a solution allows for quick adoption without having to invest in development or fine-tuning. These solutions depend on vendors and are often geared toward a narrow context.
• Boosting a vendor’s solution with a specific or proprietary data allows for more accurate and relevant results but increases usage costs.
• Building a solution gives companies the most control and competitive differentiation, but it is also expensive and difficult.

Organizations are eager to benefit from generative artificial intelligence. But in the rush to deploy AI to meet business goals, some companies jump in without fully weighing whether it’s best to buy an existing model, purchase a model and enhance it with their own data, or build a custom AI solution from scratch.

Each requires different capabilities, partnerships, and levels of customization, according to Nick van der Meulen, a research scientist at the MIT Center for Information Systems Research. “Understanding your organization’s current strengths and weaknesses is essential because it informs the strategic choices that you make about how to acquire or develop your generative AI solutions,” van der Meulen said at MIT Sloan Management Review’s Work/25 conference in May.

Three pathways for acquiring generative AI solutions are outlined in a research briefing by van der Meulen and Barbara Wixom, a principal research scientist at MIT CISR. The researchers distinguish between two types of generative AI: broadly applicable tools that enhance personal productivity, and solutions that are business-case-driven and address strategic objectives.

Here’s a look at the benefits and detriments of buying, boosting, or building your generative AI solutions.

Buy: The fast path to market

Buying means adopting an off-the-shelf generative AI solution from a vendor. The vendor provides, runs, and maintains the model, which enables organizations to quickly adopt generative AI without investing in model development or fine-tuning. Pricing is typically based on use, and the AI model is often geared to a narrow context, such as a specific business function or industry.

The benefits are speed and simplicity, while the drawbacks are limited differentiation and dependence on the vendors that control the underlying models, van der Meulen said. So if vendors plan to discontinue a version or push a major update, organizations may suddenly need to rework processes or offerings that rely on that solution or adapt to the quirks of a new model.

Rather than treating vendors purely as suppliers, organizations should view them as strategic collaborators, van der Meulen added. This approach helps vendors improve their offerings and gives their customers a chance to shape the solutions they rely on.

Boost: Customize with proprietary data

Vendor solutions might be too basic for the needs of some organizations. In these instances, boosting might be the best option.

In this development approach, the vendor provides, runs, and maintains the generative AI model, but the organization enhances it, often using proprietary data. Boosting often means either fine-tuning a model so it performs better in a specific setting, or using retrieval augmented generation, which involves feeding company-specific information into the generative AI model to improve its accuracy. This method provides more accurate and relevant results but also increases prompt lengths and usage costs.

Though boosting offers more customization than buying, it requires strong data governance, robust validation processes, and a tolerance for ongoing operational expenses.

Build: Differentiate through custom models

Building AI solutions is the most ambitious route, given that organizations must take full responsibility for developing, running, and maintaining the solution. This option gives companies complete control and the chance to create real competitive differentiation through customization — for example, companies can leverage proprietary data to create solutions for specific use cases. This is expensive and difficult, so many companies stick with fine-tuning or adapting open-source foundational models instead.

Building does lower usage costs, but it also requires significant upfront computational investment and advanced data monetization capabilities.

Generative AI Business Sprint

Attend Online

REGISTER NOW

Succeeding with generative AI solutions

The researchers have three pieces of advice for organizations as they explore whether the buy, boost, or build methodology is the best fit:

• Establish a formal, transparent generative AI innovation process. Organizations need clear governance structures, early and consistent stakeholder engagement, and a focus on scalable solutions.
• Formulate guidelines for generative AI development decisions. Leaders need to differentiate generative AI development approaches to help teams make informed decisions.
• Create a generative AI vendor partnership strategy. These partnerships rely on mutual understanding and long-term collaboration. Vendors benefit from direct feedback about what organizations are willing to pay and insights into how they will use their offerings to create value, while organizations gain from vendors’ transparency, advice, and custom support. Viewing these relationships as partnerships fosters adaptability and continuous improvement.

The strategic choice between buying, boosting, or building will shape how — and how quickly — organizations turn AI potential into measurable value.

“Rigorously prioritize these generative AI solutions based on strategic alignment and measurable value potential,” van der Meulen said. “It means balancing grassroots creativity with this discipline so your efforts truly count.”

This article is based on research by Nick van der Meulen and Barbara Wixom. Nick van der Meulen is a research scientist at the MIT Center for Information Systems Research. His research targets the challenges of senior-level executives, with a specific interest in how companies need to organize themselves differently in the face of continuous technological change. Barbara Wixom is a principal research scientist at the MIT CISR. Her work explores how organizations generate business value from data assets. She has deep expertise in data and analytics techniques and technologies, with particular interest in data and analytics strategy, capabilities, and success.