To build business value in the AI economy, start with your company’s data

The AI data revolution: Businesses are discovering new ways to leverage their proprietary data using artificial intelligence, particularly large language models (LLMs), to gain a competitive edge in increasingly crowded markets.

• McKinsey estimates that utilizing internal data for sales and marketing insights can lead to above-average market growth and increases of 15 to 25% in EBITDA.
• LLMs offer a unique method to extract value from company data, with the potential to transform businesses across various industries.

Quality over quantity: The effectiveness of AI models in enterprise settings relies more on the quality and relevance of data rather than sheer volume.

• Peter Norvig, former director of research at Google, emphasizes that “better data beats more data” in the context of AI applications.
• While frontier models are trained on massive public datasets, their utility for specific business purposes is limited without the integration of proprietary company data.

Data preparation is crucial: Ensuring data readiness is a critical step in leveraging AI for business objectives.

• Gartner reports that proper data preparation for AI can improve business outcomes by 20%.
• Poor data quality is a primary reason why 30% of internal AI projects are abandoned, according to Gartner.
• Data preparation involves removing corrupt entries, eliminating duplicates, and filling gaps in incomplete datasets.

Leveraging proprietary data: The greatest competitive benefits may be realized through the use of unique company data in AI applications.

• Valuable proprietary data may include anonymized customer information, purchasing patterns, feedback, web analytics, and supply chain data.
• Open-source data can supplement proprietary information but doesn’t provide a competitive advantage on its own.
• Using proprietary data, while adhering to privacy regulations, can also reduce legal complexities related to data sovereignty.

Adapting LLMs for business use: Most organizations lack the resources to build domain-specific models from scratch, but there are accessible methods for customizing existing LLMs.

• Prompt tuning and prompt engineering are common and relatively straightforward approaches that require fewer resources than building models from the ground up.
• These techniques allow companies to adapt LLMs to their specific needs without extensive modification of model parameters.

Real-world applications: Several companies across various industries have already begun leveraging AI and proprietary data to enhance their operations and customer experiences.

• Morgan Stanley trained GPT-4 on 100,000 internal documents to improve investment banking workflows and financial advice.
• BCG uses fine-tuned models to generate insights and client advice, with an iterative process to improve outputs and reduce hallucinations.
• ScottsMiracle-Gro created an AI-powered “gardening sommelier” using Google Cloud, trained on product catalogues and internal knowledge to provide gardening advice and product recommendations.
• Volkswagen of America developed an AI virtual assistant trained on vehicle instruction guides and connected car data to help drivers better understand their vehicles.

The growing importance of data assets: As LLMs become more commoditized and accessible, the value of proprietary data is likely to increase.

• Content owners are pushing back against free data collection by AI companies like OpenAI and Anthropic.
• Companies of all sizes should consider how to protect and leverage their internal data assets for competitive advantage.
• Even seemingly mundane assets like product catalogues and user manuals can be valuable resources when combined with AI technologies.

Looking ahead: The data-driven competitive landscape: As AI technologies become more accessible, companies that effectively leverage their proprietary data are likely to gain significant advantages in their respective markets.

• The commoditization of LLMs may lead to a shift in focus towards unique data assets as the primary differentiator for businesses.
• Companies should start viewing their internal data as a valuable resource that can be transformed into actionable insights and improved customer experiences through AI applications.
• As the AI landscape evolves, businesses that fail to capitalize on their data assets may find themselves at a competitive disadvantage in the increasingly data-driven economy.