Roland Berger Toward Data-Centric AI: Strategy and Implementation

The Data-Centric AI Revolution

The artificial intelligence landscape is undergoing a fundamental transformation. As generative AI experiences unprecedented adoption rates, with enterprise spending increasing sixfold in 2024 compared to 2023, a critical shift is emerging from model-centric to data-centric approaches. This evolution represents more than just a technical preference—it’s a strategic imperative that will determine which organizations successfully harness AI for sustainable competitive advantage.

Roland Berger’s comprehensive research, based on in-depth interviews with 150 data and AI executives, reveals a sobering reality: while private consumers rapidly adopt tools like ChatGPT, enterprises struggle to translate AI investments into profitable and efficient business outcomes. The solution lies not in more sophisticated models, but in mastering the art of data-centric AI implementation.

The data-centric AI approach recognizes that AI model governance and data quality are becoming the primary differentiators. As large language models evolve beyond simple “stochastic parrots” to systems capable of human-like reasoning, the quality and strategic integration of training data becomes paramount.

From Model-Centric to Data-Centric Approaches

The traditional model-centric approach to AI focused primarily on algorithmic innovations, model architectures, and computational efficiency. However, as AI technologies mature, a subtle but fundamental shift is occurring. Access to sophisticated AI models and adequate computing power is no longer a significant barrier—these resources have become increasingly commoditized and accessible to most organizations.

This commoditization creates a new competitive landscape where data strategy becomes the key differentiator. Organizations that can effectively feed their proprietary knowledge into increasingly sophisticated AI systems will gain sustainable advantages over competitors relying solely on generic, publicly available data sources.

The data-centric approach emphasizes three critical pillars: data evolution alongside hardware infrastructure and algorithmic innovation. While hardware advances enable more powerful processing and algorithms become more sophisticated, data evolution encompasses the systematic capture, curation, and integration of proprietary organizational knowledge that cannot be replicated by competitors.

Enterprise AI Spending and Adoption Patterns

The scale of enterprise investment in generative AI is staggering. With corporate spending rocketing by a factor of six in 2024 alone, it’s clear that GenAI has moved far beyond experimental sandboxes into serious business strategy considerations. However, this massive investment has not yet translated into proportional business value for most organizations.

Research indicates that while 93% of executives are convinced that systematic AI implementation will improve their data management practices, the reality is more complex. Many organizations are discovering that their existing data infrastructure, governance frameworks, and management practices are inadequate for the demands of advanced AI systems.

The adoption pattern reveals a significant disparity between consumer and enterprise success. Consumer applications like ChatGPT achieve rapid widespread adoption because they operate on general knowledge and don’t require complex data integration. Enterprise applications, however, must integrate with existing systems, comply with regulatory requirements, and leverage proprietary data sources—creating substantially more complex implementation challenges.

This complexity manifests in longer implementation timelines, higher integration costs, and the need for comprehensive organizational change management. Successful enterprise AI transformation requires not just technology deployment, but fundamental shifts in data culture, governance practices, and organizational workflows.

Proprietary Knowledge as Competitive Advantage

As AI models become commoditized, competitive advantage increasingly shifts to organizations that can effectively mobilize their unique knowledge assets. This includes tacit expertise embedded in chat histories, internal discussions, employee interactions, and contextual intelligence that has been accumulated over years of business operations.

Proprietary knowledge encompasses several critical categories that traditional data management approaches often overlook. Tacit expertise includes the informal knowledge that employees develop through experience but rarely document systematically. Contextual intelligence represents understanding of industry dynamics, customer behaviors, and market conditions that cannot be gleaned from public sources.

The challenge lies in systematically capturing and integrating this proprietary knowledge into AI systems. Many organizations have vast repositories of valuable intellectual property scattered across email systems, collaboration platforms, internal documents, and employee expertise that remains largely untapped for AI applications.

Roland Berger’s research emphasizes that organizations must develop systematic approaches to feed this wealth of proprietary knowledge into advanced AI reasoning capabilities. This requires more than just data aggregation—it demands sophisticated curation, quality assurance, and integration processes that can transform informal knowledge into AI-ready data assets.

Multimodal Data Integration Challenges

The evolution toward multimodal AI capabilities presents both unprecedented opportunities and significant implementation challenges. As AI systems become capable of processing and integrating text, audio, visual, and structured data simultaneously, organizations must completely reimagine their data management strategies.

Traditional data management approaches were designed for structured data in isolated systems. The multimodal AI era demands integrated approaches that can handle diverse data types, maintain data lineage across different formats, and ensure consistent quality standards across all data modalities.

This integration complexity extends beyond technical considerations to organizational and process challenges. Different data types often reside in different systems, managed by different teams, with different governance standards and access controls. Effective data governance frameworks must evolve to address these multimodal requirements while maintaining security, compliance, and quality standards.

The multimodal integration challenge also creates new opportunities for competitive differentiation. Organizations that successfully integrate diverse data sources can develop AI applications that are more comprehensive, accurate, and valuable than competitors limited to single-modality approaches.

Data Quality vs. Data Quantity

One of the most critical insights from Roland Berger’s research is the fundamental importance of data quality over quantity in generative AI implementations. While synthetic data creation has effectively addressed many quantity limitations, quality and diversity remain the primary bottlenecks for successful AI deployment.

High-quality data ensures more accurate AI outputs, reduces hallucinations and errors, and enables more reliable business applications. Conversely, large quantities of poor-quality data can actually degrade AI performance, introducing biases, inconsistencies, and unreliable behaviors that undermine business value and user trust.

The quality challenge extends beyond traditional data cleansing to encompass contextual accuracy, temporal relevance, and semantic consistency. AI systems require data that is not just technically correct, but contextually appropriate and semantically rich enough to support sophisticated reasoning tasks.

This quality imperative has significant implications for data strategy. Organizations must invest in robust data curation processes, quality assurance frameworks, and continuous monitoring systems that can maintain high standards across diverse data sources. The focus shifts from maximizing data volume to optimizing data value and reliability.

Quality considerations also extend to data diversity—ensuring that training data represents the full range of scenarios, use cases, and contexts that AI systems will encounter in production environments. Addressing AI bias and ensuring representational diversity becomes a critical quality dimension that requires systematic attention.

Industry-Specific Implementation Barriers

Roland Berger’s research reveals significant variation in AI implementation challenges across different industries. Healthcare and retail organizations, for example, consistently report major difficulties in accessing data of sufficiently high quality for AI applications, while other sectors face different but equally significant barriers.

In healthcare, regulatory compliance requirements, patient privacy constraints, and fragmented data systems create complex implementation challenges. The industry’s need for extremely high accuracy and reliability standards means that data quality requirements are particularly stringent, while regulatory frameworks often limit data sharing and integration possibilities.

Retail organizations face different challenges, particularly around integrating data from diverse sources including point-of-sale systems, customer interactions, supply chain data, and external market information. The dynamic nature of retail environments means that data currency and real-time processing capabilities become critical success factors.

Financial services organizations must navigate stringent regulatory requirements while managing extremely sensitive data. The industry’s need for explainable AI and audit trails creates additional complexity in data management and AI implementation approaches.

These industry-specific challenges underscore the need for tailored implementation approaches rather than one-size-fits-all solutions. Successful data-centric AI strategies must account for industry dynamics, regulatory requirements, and sector-specific data characteristics.

Strategic Roadmap for Data-Centric AI

Implementing data-centric AI requires a comprehensive strategic roadmap that addresses both technical and organizational dimensions. Roland Berger’s research provides a framework for organizations to systematically approach this transformation, focusing on practical steps that build toward comprehensive AI-enabled capabilities.

The strategic roadmap begins with a comprehensive data asset inventory and assessment. Organizations must identify existing data sources, evaluate their quality and accessibility, and map potential integration opportunities. This assessment should encompass both structured and unstructured data, including informal knowledge repositories that may not be part of traditional data management systems.

Governance framework development represents the second critical phase. Organizations must establish clear policies, procedures, and oversight mechanisms for data-centric AI implementation. This includes data quality standards, access controls, privacy protection measures, and compliance frameworks that can scale with AI deployment.

The third phase focuses on infrastructure and capability building. Organizations need to develop or acquire the technical infrastructure, skills, and processes necessary to support data-centric AI at scale. This includes data integration platforms, quality assurance systems, and the organizational capabilities needed to maintain and optimize these systems over time.

Finally, the roadmap emphasizes iterative implementation and continuous improvement. Rather than attempting comprehensive transformation immediately, successful organizations typically begin with targeted use cases, demonstrate value, and gradually expand their data-centric AI capabilities across the organization.

Future Implications and Recommendations

The shift toward data-centric AI has profound implications for organizational strategy, competitive positioning, and long-term success in the AI-enabled economy. Organizations that successfully navigate this transition will develop sustainable competitive advantages, while those that continue to focus primarily on model selection and deployment may find themselves at a significant disadvantage.

The future competitive landscape will be characterized by organizations that have successfully integrated their proprietary knowledge assets into AI systems. These organizations will be able to provide more accurate, contextually relevant, and valuable AI applications than competitors relying on generic data sources and commoditized models.

Key recommendations for organizations embarking on data-centric AI transformation include: developing comprehensive data strategies that encompass both technical and organizational dimensions; investing in data quality and governance capabilities as foundational requirements; building organizational capabilities for continuous learning and adaptation; and approaching implementation through iterative, value-driven approaches rather than attempting comprehensive transformation immediately.

The research also emphasizes the importance of cultural transformation alongside technical implementation. Organizations must develop data-centric cultures that value data quality, systematic knowledge capture, and continuous improvement. This cultural dimension is often underestimated but represents a critical success factor for long-term AI implementation success.

As AI technologies continue to evolve, the organizations that master data-centric approaches will be best positioned to capitalize on new capabilities and maintain competitive advantages in an increasingly AI-driven business environment. The time for strategic data-centric AI planning and implementation is now—delaying these investments will only make the eventual transformation more difficult and costly.