AI flying blind: Why large corporations no longer understand their own data

The underestimated AI risk: Why data chaos can cost companies millions

The hype surrounding artificial intelligence is increasingly giving way to a sobering reality in the C-suite. While the language models of OpenAI, Google, and others are becoming ever more powerful, their productive use in practice usually fails due to a massive, yet often ignored, problem: internal data chaos. When contracts, chat histories, emails, and spreadsheets are scattered unstructured across countless silos and departments, even the most intelligent AI agents are left with nothing or simply hallucinate. It is precisely at this vulnerability that a rapidly growing, multi-billion-dollar market is emerging. Specialized startups and established tech giants are building so-called semantic data layers – the new, central operating system for enterprise data. But this technology not only brings enormous efficiency gains and revolutionary payment models, it also harbors tangible risks in terms of cybersecurity and dependency. This article offers a deep dive into the architecture of the AI ​​future and explores what this transformation means for the economy – from large corporations to medium-sized businesses.

Related to this:

• Unframe AI | AI-ready data. Built from the systems you already run.

A billion-dollar market is emerging from an old problem

The debate surrounding artificial intelligence in business has fundamentally shifted in the past two years. While the initial focus was on which language model provided the best answers, a far more pragmatic aspect has now taken center stage: the quality, timeliness, and accessibility of the data these models are meant to work with. A survey of one hundred senior data and technology executives from corporations with over two billion US dollars in revenue and more than five thousand employees paints a sobering picture. Nearly all respondents—ninety-nine percent, to be precise—struggle to define consistent business metrics across different tools and departments. Almost eighty percent of data teams spend more than half their time preparing data instead of actually gaining insights from it. These figures reveal that semantic inconsistency is no longer a marginal issue, but rather the very operating system on which many large companies still run.

A new category of providers is stepping into precisely this gap, positioning themselves as a link between the fragmented reality of businesses and the demands of modern AI systems. Their promise is to consolidate all of a company's data sources—from emails and contracts to ERP and CRM systems—into a continuously synchronized, governance-enabled layer. The success of this promise is evident in the case of the Californian-Israeli-German startup Unframe, which, within twelve months of its official market launch, secured more than one hundred million US dollars in contract volume. Its latest funding round of fifty million US dollars, led by Highland Europe, brought the total amount of funds raised to one hundred million US dollars. Such growth figures are a remarkable signal of a genuine, not merely claimed, need in a market segment that has only existed for a few years.

Why traditional data architectures are hindering AI

To understand the implications of this development, it's worth looking at the historical function of data warehouses. For decades, these were optimized for a single purpose: retrospective reporting. Quarterly figures, sales statistics, key performance indicators for management – ​​all of this could be satisfactorily represented using periodically updated, highly structured databases. However, artificial intelligence, especially agent-based systems designed to make independent decisions or trigger actions, requires something fundamentally different. It needs operational information that is up-to-date in real time and linked to actual business processes. An AI agent intended to support a supply chain decision cannot work with data that is three weeks old and has been prepared solely for retrospective analysis.

This discrepancy is exacerbated by another structural problem. Estimates from major technology companies suggest that up to 90 percent of corporate data is unstructured, scattered across documents, chat histories, support tickets, invoices, and calendar entries. This vast amount of information largely lies outside the realm of traditional, spreadsheet-based data repositories and was simply inaccessible to traditional analytics tools. Without a unified semantic layer that consistently represents both structured and unstructured content, it remains virtually impossible to consolidate information from diverse sources and provide it to an AI model as a reliable context.

A separate benchmark project assessing data maturity for generative AI arrives at a structurally similar finding, albeit with different weighting. In this study, 59 percent of the surveyed managers cited data quality and consistency as the biggest hurdle to the productive use of generative AI, followed by fragmented data landscapes at 50 percent and data security and privacy issues, also at 50 percent. It is also noteworthy that only 13 percent of companies already using generative AI can prevent sensitive data, such as customer or employee information, from flowing unfiltered into language models. This points to a governance gap that extends far beyond technical compatibility issues and leads directly into the realm of data protection and regulatory liability.

The business model behind the new data layer

The fundamental principle pursued by providers like Unframe can be described in four functional steps. First, enterprise systems, applications, data repositories, files, and APIs are connected to the platform. Next, these heterogeneous sources are transformed into a single, governance-enabled data layer that applies to the entire organization. The third step involves continuous synchronization with the source systems, ensuring that the data layer doesn't become just another outdated snapshot, but rather reflects the actual state of the business. Only in the fourth step is this foundation used to power artificial intelligence, analytics, and automation – all built on the same consistent basis.

Conceptually, this approach resembles a kind of digital twin of the organization, but not primarily for simulating physical processes. Instead, it serves to map customers, products, contracts, and transactions in a consistent representation. The economic appeal of such a model lies in the fact that each new AI application doesn't have to start from scratch to access relevant business data. Instead, each application draws on the same, already cleaned and enriched foundation. This potentially reduces the costs of each subsequent AI initiative significantly, as the most complex and often underestimated part of an AI project—data integration and cleansing—only needs to be done once.

Unframe describes its offering as modular, reusable building blocks that can be assembled into customized solutions without requiring separate model training for each use case. The so-called Framery functions as an open platform layer that can connect to any Software-as-a-Service offering, API, database, or file format, regardless of whether the infrastructure is operated in the cloud, on-premises, or in a hybrid environment. For European, and especially German, companies, another aspect is of considerable importance: the platform operates independently of any specific language model provider, meaning there is no tie-in to a particular provider. Furthermore, compliance with the European General Data Protection Regulation (GDPR) and the EU's AI law is explicitly advertised. Data should, by default, not leave the customer's system boundaries unless explicitly requested.

Outcome-based pricing as a trust signal in a distrustful market

A particularly revealing aspect of the business strategy lies in the pricing model. Unlike traditional software providers who charge license fees regardless of actual benefit, Unframe relies on a results-based billing model. Customers only pay when a solution demonstrably works and delivers results. This decision is by no means accidental, but rather a direct response to deep-seated skepticism in boardrooms. As the company's co-founder has publicly stated, companies are tired of funding solutions that fail in the vast majority of cases, which is why they demand a model where payment is made for actual success.

This observation aligns with the broader experience of many corporations that have been experimenting with so-called proof-of-concept projects for years without integrating them into production systems. A significant backlog of promising AI use cases, numerous pilot projects, and yet almost nothing actually used in day-to-day operations – this summarizes the starting point for many companies turning to this new category of providers. The economic core of this promise lies in shifting the risk of a failed project from the customer alone to at least part of the provider's own risk. This fundamentally alters the incentive structure and is likely one of the key reasons for the observed growth dynamics, particularly the aforementioned net revenue retention of 400 percent, which means that existing customers have quadrupled their spending on average within a year.

Who the customers are and what that reveals about the level of market maturity

The mentioned reference customers paint an interesting picture of the various industries. Among those named are the real estate companies Cushman & Wakefield and Avison Young, the cybersecurity company Armis, and the Swiss publishing house NZZ. Additionally, use cases are reported for airlines to optimize crew scheduling and for retailers to plan sales promotions. This distribution across real estate, media, security, aviation, and retail suggests that this is not a niche solution for a single industry, but rather a cross-industry infrastructure problem that evidently arises in very different economic contexts.

Also of interest is the company's geographical and organizational structure. Headquartered in Cupertino, California, with additional offices in Tel Aviv and Berlin, Unframe embodies a transatlantic or transcontinental setup typical of the technology sector, combining American capital, Israeli technological development, and proximity to the European market. For a company with a German focus, it is particularly relevant that the operational co-founder explicitly works from Berlin and that the company has repeatedly participated in European digital events such as the Mind the Tech conference in Berlin. This signals a deliberate strategic focus on the European market, which, due to stricter data protection requirements and a generally more cautious attitude toward the uncontrolled transfer of company data to American cloud providers, has different requirements than the North American market.

The competitive landscape and the race of the tech giants

The trend described is by no means limited to specialized startups. Established technology companies have also identified the problem of fragmented enterprise data as a key growth area. Dell Technologies, together with Nvidia, has developed an AI Data Platform whose stated goal is to transform decades of fragmented enterprise data into actionable intelligence. A senior product marketing executive at Dell aptly stated that the solution to the data problem must not only address the storage level, but must encompass the entire pipeline from data ingestion and curation to orchestration, so that the data ultimately reaches the computing power of the graphics processors reliably.

IBM, in its assessment of key data trends for 2026, also emphasized that without a converged platform offering unified access to structured and unstructured data, companies will be unable to implement analytics and autonomous automation with the necessary speed and reliability. This assessment from one of the world's oldest and largest IT companies underscores that this is not a niche issue for small vendors, but rather a fundamental structural challenge identified as a critical bottleneck across the industry. Significantly, IBM explicitly stresses that no single vendor can solve this problem alone, as the integration of distributed data and the development of the necessary infrastructure for autonomous AI systems require a broad ecosystem of partners and technologies.

This assessment sheds important light on the actual competitive dynamics. A market is not emerging in which a single winner dominates the entire value chain, but rather a multitude of providers with different focuses, some complementary, some competing with one another. Specialized providers like K2View focus explicitly on operational, production-ready data architectures, while infrastructure companies like Dell and Nvidia provide the underlying computing capacity, and platform providers like Unframe bridge the gap between raw data and deployable AI applications.

Managed AI Platform - Image: Xpert.Digital

Here you will learn how your company can implement customized AI solutions quickly, securely and without high entry barriers.

A managed AI platform is your all-inclusive, worry-free solution for artificial intelligence. Instead of dealing with complex technology, expensive infrastructure, and lengthy development processes, you receive a ready-made solution tailored to your needs from a specialized partner – often within just a few days.

The key advantages at a glance:

⚡ Rapid implementation: From idea to ready-to-use application in days, not months. We deliver practical solutions that create immediate added value.

🔒 Maximum data security: Your sensitive data stays with you. We guarantee secure and compliant processing without sharing data with third parties.

💸 No financial risk: You only pay for results. High upfront investments in hardware, software, or personnel are completely eliminated.

🎯 Focus on your core business: Concentrate on what you do best. We take care of the entire technical implementation, operation, and maintenance of your AI solution.

📈 Future-proof & scalable: Your AI grows with you. We ensure continuous optimization and scalability, and flexibly adapt the models to new requirements.

More information here:

• Managed AI Platform

Governance as an underestimated competitive factor

One aspect often overlooked in the public debate on artificial intelligence is the issue of governance—that is, control over who has access to which data, how it is processed, and how it can be traced back to specific AI decisions. The survey of one hundred data managers cited above shows that eighty-seven percent of respondents demand greater transparency regarding how artificial intelligence uses and interprets their data. This is a remarkably high figure and suggests that trust in the reliability of AI systems is not a side issue, but rather a key prerequisite for their wider acceptance within organizations.

The medium-term priorities of the surveyed companies for the next three to five years confirm this picture. Scalability across various data sources is cited as a priority by 92 percent, the portability of data and key performance indicators (KPIs) by 83 percent, and governance and observability of artificial intelligence by 82 percent. These three priorities are closely interrelated. Without scalable data structures that are consistent across system boundaries, reliable governance cannot be established, and without governance, any scaling remains associated with significant legal and reputational risks, particularly with regard to the European General Data Protection Regulation (GDPR) and the now-effective AI law of the European Union.

This is particularly important for German and European companies, as the regulatory framework is significantly stricter than in the United States or much of Asia. European companies implementing a data layer for artificial intelligence must consider from the outset how personal data will be protected, how processing procedures will be documented, and how AI system decisions can be made transparent in the event of a dispute. Providers who incorporate these requirements into their architecture from the beginning, for example, by allowing data to remain entirely within their own system boundaries, gain a structural advantage over competitors who primarily rely on US cloud infrastructures with less stringent regulations.

Economic implications for German SMEs

While the reference customers mentioned so far are predominantly large, internationally operating corporations, the legitimate question arises as to the relevance of this topic for German SMEs, which traditionally form the backbone of the German economy. At first glance, the situation for smaller companies appears less dramatic, as they naturally have fewer systems and smaller amounts of data. However, closer examination reveals that SMEs that have grown organically over the years and implemented various isolated solutions for accounting, inventory management, customer relationship management, and communication are confronted with a structurally similar fragmentation problem – just on a smaller scale.

The crucial economic question, therefore, is not whether the problem exists, but whether an investment in a unified data layer can be economically justified given limited IT budgets. The emergence of results-based pricing models, as pioneered by Unframe , is likely to play a significant role here, as this model considerably reduces the financial risk for smaller companies. At the same time, it can be assumed that leaner, more specialized solutions will become established for medium-sized businesses rather than the comprehensive platforms designed for multi-billion-dollar corporations. However, the fundamental logic that a consistent, continuously synchronized data layer is a prerequisite for any meaningful AI application remains valid regardless of company size.

Risks and open questions regarding the new data layer

While this development deserves considerable attention, the associated risks must not be overlooked. Migrating the entire corporate data landscape into a unified, central layer inevitably creates an extremely attractive target for cybercriminals. A successful compromise of this central data layer could have far more extensive consequences than a breach of a single, isolated system, as it could instantly grant access to customer, contract, and financial data for the entire company. The security architecture of such platforms must therefore be exceptionally robust, and companies should critically examine how granular access rights are controlled and encryption standards are implemented when selecting a provider.

Another, more strategic risk lies in the dependence on a single platform provider. While Unframe explicitly emphasizes that it is not tied to a specific language model, the underlying data layer itself still represents a significant structural dependency. Should a company decide to switch to another provider, migrating a company-wide data layer that has grown over years is likely to be a considerable undertaking, similar to the well-known challenges of switching large ERP systems. This type of lock-in effect is naturally rarely addressed by the providers themselves, but it should be factored into the evaluation of every strategic decision.

Finally, the question of the actual quality of the results remains. Even the most advanced data layer cannot guarantee that the AI ​​decisions derived from it are error-free. Particularly with autonomous, agent-based systems that independently initiate actions, the question of responsibility in the event of errors remains unresolved. If an AI agent makes an incorrect business decision based on a faultily synchronized data layer—for example, misinterpreting a contract clause or making a flawed supply chain decision—the question of liability arises immediately, a question that has so far been inadequately addressed both contractually and regulatoryly.

A sober assessment of actual progress

A critical assessment of this development warrants caution regarding exaggerated promises of salvation. The notion that artificial intelligence can be implemented and made production-ready within an established corporate environment within a few days may be true in selected, clearly defined use cases, but for the vast majority of complex business processes, it is likely a significant oversimplification of reality. Experience with previous waves of digital transformation, such as the introduction of cloud infrastructure or the implementation of ERP systems, shows that even technically sophisticated solutions regularly fail due to organizational resistance, insufficient data quality in the source systems, and a lack of willingness to change among the workforce.

At the same time, the underlying diagnosis is undeniably correct. The fragmentation of enterprise data is real, well-documented empirically, and indeed represents one of the key bottlenecks for the productive use of artificial intelligence. The question, therefore, is less about whether a unified, AI-compatible data layer makes sense, but rather which specific path to achieving it is the most economically viable for a particular company with its unique historical system landscape, risk tolerance, and available budget. For companies that have been hesitant to address this issue so far, market developments over the past two years provide an important clue: those who continue to experiment with isolated, uncoordinated pilot projects will fall behind competitors who invested early in a consistent data foundation in the medium term.

Breaking down data silos: The path to interoperable enterprise AI

The development of the coming years will depend significantly on whether uniform standards for interoperability between different data layers and AI platforms are established. If the industry succeeds in developing open protocols that allow different providers of data layers and AI models to work together seamlessly, this would considerably reduce existing dependency risks and intensify competition to the benefit of customers. Initial steps in this direction are already visible, for example, through the support of open interface protocols that enable integration with existing, in-house AI assistants without requiring their complete replacement.

For German businesses, which have traditionally been rather hesitant in adopting new digital technologies, yet possess a data protection culture that is strict by international standards and exemplary in many respects, this development presents an interesting strategic opportunity. Providers who integrate European compliance requirements into their architecture from the outset could gain a significant advantage in terms of trust, particularly in German-speaking countries, over purely American providers whose data processing practices are traditionally less transparent and subject to less stringent regulations. German companies that establish a consistent, governance-ready data layer early on will gain not only a technological but also a regulatory competitive advantage, which is likely to prove increasingly valuable in the coming years as the practical implementation of European AI legislation gains further momentum.

Konrad Wolfenstein

I would be happy to serve as your personal advisor.

You can contact me at wolfenstein∂xpert.digital or

Just call me on +49 7348 4088 965 .

LinkedIn