Title: How to Manage Two Types of Generative Artificial Intelligence Resource URL: https://mitsloan.mit.edu/ideas-made-to-matter/how-to-manage-two-types-generative-artificial-intelligence Publication Date: 2024-12-04 Format Type: Article Reading Time: 5 minutes Contributors: Nick Meulen;Barbara Wixom;Kristin Burnham; Source: MIT Sloan Management Review Keywords: [Artificial Intelligence Technology Generative AI tools AI governance Vendor partnerships] Job Profiles: Academic/Researcher;Content Writer/Editor;Artificial Intelligence Engineer;Digital Marketing Manager;Chief Technology Officer (CTO); Synopsis: In this article, tech journalist Kristin Burnham discusses a research briefing by MIT Center for Information Systems Research researchers Nick van der Meulen and Barbara H. Wixom, exploring two distinct ways organizations implement generative AI to enhance business value. Takeaways: [Organizations are adopting two distinct approaches to generative AI, first, as a broadly applicable tool for personal productivity and then as tailored solutions aimed at addressing specific business needs., Broadly applicable generative AI tools, like conversational systems, are versatile but come with challenges such as biases, lack of context, and the potential for producing inaccurate outputs, requiring critical evaluation by users., To manage generative AI risks, organizations should develop clear usage guidelines, invest in training, and standardize tool access with a select group of vendors., Tailored generative AI solutions focus on strategic business objectives, offering the potential for scale but also creating challenges such as the risk of shadow AI and uncertainty around vendor control and model performance., Organizations can maximize the value of generative AI solutions by establishing transparent innovation processes, creating clear development guidelines, and forming strategic vendor partnerships.] Summary: The research briefing by Nick van der Meulen and Barbara H. Wixom from the MIT Center for Information Systems Research explores how organizations are leveraging generative AI in two primary ways: broadly applicable tools for personal productivity and tailored solutions for specific stakeholder groups. Broadly applicable tools, such as conversational AI systems, are versatile and pose challenges like context lack, bias, and security risks. To mitigate these, organizations should develop clear usage guidelines, invest in training, and standardize vendor selection. Tailored generative AI solutions are driven by strategic business objectives and face challenges like 'shadow generative AI' and vendor control over foundation models. Organizations can overcome these by establishing transparent innovation processes, formulating development guidelines, and creating vendor partnership strategies. Content: ## Introduction As organizations explore generative artificial intelligence (AI) to drive productivity and strategic value, they have adopted two primary implementation models. According to a recent briefing by researchers at the MIT Center for Information Systems Research, these models fall into: 1. Broadly applicable generative AI tools, which empower individual users across the enterprise. 2. Tailored generative AI solutions, which address the needs of specific stakeholder groups and strategic objectives. This analysis, grounded in roundtable discussions with a data research advisory board and interviews with industry executives, outlines each approach’s defining characteristics, unique challenges, and recommended management principles. ## Broadly Applicable Generative AI Tools ### Characteristics Broadly applicable generative AI tools—such as conversational systems and embedded digital assistants within productivity suites—are designed for universal use. They arrive as external products that organizations “privatize” to protect internal data. As one vice president and chief data and analytics officer at a global animal health company observed, “This is AI for everyone.” Users shape and refine these tools’ applications through everyday interactions. ### Key Challenges 1. **Output Quality and Relevance** • Large language models predict common word sequences, so results often lack specificity. • Users must craft precise prompts to obtain meaningful, context-appropriate responses. 2. **Accuracy and Bias** • Models may omit critical context, reflect or amplify bias, or present false assertions as fact. • Users must continuously audit outputs to avoid perpetuating errors. 3. **Data Security and Compliance** • Publicly available tools expose organizations to data loss, intellectual property leakage, copyright violations, and security breaches. 4. **Cost Management** • Licensing fees for multiple vendors can escalate once free trials and promotional discounts expire. ### Recommended Management Principles To enable safe, effective experimentation, organizations should consider the following actions: - **Develop Clear Usage Guidelines** Establish cross-functional teams—comprising technology, legal, privacy, and governance stakeholders—to define permissible tools, usage conditions, associated risks, and potential consequences. “You want to be innovative and speedy, but you also want to be risk aware, data secure, and compliant,” the data officer advised. - **Invest in Training** Implement programs that teach employees to formulate precise prompts, evaluate underlying models, and use generative AI responsibly. - **Standardize Tool Selection** Convene a representative user group to identify a small set of vetted vendors. Provide sanctioned licenses to foster a secure environment for innovation. ## Tailored Generative AI Solutions ### Definition and Scope Tailored generative AI solutions are business-case–driven initiatives built to meet specific strategic objectives and deliver value at scale. Organizations advance these solutions only after validating end-user desirability, technical feasibility, and business viability. As the data officer explained, “In manufacturing, for example, a solution might monitor processes and products in real time to ensure they meet quality benchmarks.” ### Unique Challenges 1. **Shadow Generative AI** • Independent stakeholder groups may deploy unsanctioned AI solutions with eager vendors, bypassing enterprise governance. 2. **Vendor Concentration and Model Opacity** • A limited number of vendors control most foundational models, complicating bias assessment, behavior prediction, and data security. 3. **Uncertain Long-Term Costs** • Fluctuating usage patterns, evolving model performance, and unpredictable pricing make it difficult to forecast operating expenses. 4. **Trade-Offs in Solution Delivery** • Whether organizations buy, enhance, or build models affects transparency, contextual awareness, and total cost of ownership. ### Recommended Management Principles To deliver successful, targeted generative AI solutions, organizations should adopt the following practices: - **Establish a Formal Innovation Process** Create clear governance structures with early, ongoing stakeholder engagement. Prioritize scalable solutions that align with strategic goals. - **Formulate Development Guidelines** Differentiate among buying, building, and enhancing models. Provide decision frameworks that clarify the benefits and limitations of each approach. - **Craft a Vendor Partnership Strategy** Engage in long-term, collaborative relationships with AI vendors to ensure mutual understanding, adaptability, and continuous improvement. ## Conclusion By distinguishing between broadly applicable tools and tailored solutions, executives can address generative AI’s diverse challenges and opportunities. Implementing structured governance, training, and vendor strategies will enable organizations to scale AI responsibly and maximize its business value.