Building a Strong Foundation

Responsible use of high-quality data is at the core of GenAI success. Firstly, it is critical that those using GenAI systems can have confidence that data is being used appropriately, so integrating ethical and responsible AI practices into your strategy is important, not just to meet compliance standards, but to drive long-term adoption. Secondly, poor quality data is literally draining the value from GenAI initiatives: Gartner reports that organizations lose an average of $15 million per year due to poor data quality—a figure that is only set to rise as reliance on data-driven AI solutions grows. Errors in data can mean suboptimal employee support recommendations, greater anxiety, or privacy violations, all of which undermine the promise of AI-driven EX. Getting to grips with both parts of this problem is core to the successful implementation of GenAI.

Recent scientific research (INRIA, 2024) strongly suggests that responsible AI – of which data management is a key part – is critical to building trust, ensuring fairness, and gaining societal acceptance.  It provides the policies, standards, and technologies needed to ensure data is managed, protected, and used consistently across the organization. It also serves as the backbone for high data quality, security, availability, and compliance, ensuring that the datasets feeding AI models are reliable and free from bias. 

At Orange, data protection was our main concern when we rolled out our internal GenAI tool – Live Intelligence – across the entire company. This was not only important for us as an employer but also for our staff.  When they were first introduced to the new tool, the first questions people asked were usually, "Where is my data going?" and "What data can I use?" We implemented two key measures in response to these concerns.  

First, we established a rigorous technical and legal framework to ensure that any data submitted to a large language model (LLM) is never stored or repurposed—it is processed solely to generate a response to the user's prompt. This strict standard has been consistently applied to every LLM editor integrated into Live Intelligence. Furthermore, we made a firm commitment to keep all our data hosted within Europe, and, when adding web search functionality, we carefully selected a search engine that fully complied with these data protection protocols. 

Second, we introduced a Responsible AI Charter that clearly outlines the guiding principles for AI across the company. This charter not only details both the risks and opportunities associated with AI technology, but also defines governance mechanisms, explains the robust measures in place to safeguard personal data, and specifies the terms and conditions for responsible use. For example, we have pledged never to use AI-generated images in any public-facing promotional materials—demonstrating our dedication to transparency and ethical standards in all aspects of AI deployment.

Of course, governance cannot be static and must evolve to take account of developments in the way organizations use GenAI.  For example, like many companies, Orange Business is exploring the use of Agentic AI (a recent survey has revealed that 51% of companies have deployed AI agents, with another 35% planning to do so within the next two years). We believe this will transform how we handle routine tasks across infrastructure management, customer support, and office productivity. By automating these processes, we not only allow employees to focus on higher-value tasks—but shift their role from an "in the loop" executor to an "on the loop" supervisor, overseeing and authorizing work carried out by intelligent agents. 

This shift, however, means that employees may not always have the opportunity—or the capacity—to verify every AI-generated response or intervene in real time. In high-stakes scenarios, such as providing customers with immediate troubleshooting guidance, trust in GenAI’s accuracy becomes non-negotiable. At Orange Business, we are committed to meeting this challenge head-on by pioneering solutions like AI systems that validate each other’s outputs and investing in the fine-tuning of Small Language Models to ensure reliability and inspire confidence in every interaction. 

The effectiveness of GenAI is not measured solely by technical performance, but also by its ability to produce fair and transparent outcomes. Successful GenAI implementation therefore demands ongoing monitoring and reduction of bias (George et al., 2025) as models trained on skewed or unrepresentative data can learn and amplify societal biases related to race, gender, or culture.  This can generate unfair recommendations or decisions that impact morale, possibly leading to discriminatory outcomes in sensitive applications like hiring, with potential legal ramifications. 

Ensuring transparency in how AI models make decisions and the provision of mechanisms for human oversight are essential to maintain trust and avoid ethical pitfalls. At Orange, we take a proactive approach to this issue, conducting systematic bias reviews, performing regular fairness audits, and maintaining transparent communication with all stakeholders. Our governance structure—anchored by the Data and AI Ethics Council and supported by dedicated, cross-functional teams—guarantees that every AI deployment is rigorously aligned with legal requirements, ethical principles, and core societal values. These measures are not just procedural; they are foundational to ensuring our AI systems consistently deliver fair and equitable results for every user.

Another critical challenge lies in the realm of data security and privacy. GenAI implementations create unique privacy and security risks due to their data-intensive nature, with systems often requiring access to vast, sensitive datasets—ranging from HR records to internal communications.  

This not only introduces the risk of employees or other users unintentionally placing confidential data in public GenAI tools but also makes them attractive targets for cyber threats – in both cases, organizations may be exposed to significant legal and reputational risks. (To address this, Orange Business has built and operates its own highly secure infrastructure, equipped with proprietary GPUs in a secure cloud—resources we also extend to our customers.) 

The picture is further complicated by the fact that models organizations often trained models on internet data without the explicit consent of the data's creators, raising legal questions over unauthorized data use; and by laws such as the U.S. Clarifying Lawful Overseas Use of Data (CLOUD) Act which grants authorities the power to request data from tech companies, regardless of where it is stored. This creates a complex web of compliance obligations, especially for multinational organizations, which must contend with a patchwork of data residency and privacy regulations. 

Safeguarding employee data is not just a legal necessity; it’s foundational to building trust in AI-driven solutions. Any breach or misuse of this information can quickly erode employee confidence, hinder adoption, and cause lasting damage. For more on this topic, read our "Trust by design" blog.

The potential of GenAI hinges largely upon the quality of data used to train and operate these systems. In an age where GenAI technologies like Microsoft Copilot and Live Intelligence promise to transform business processes, any cracks in the data foundation can have outsized consequences. Poor data quality—manifesting as inaccuracies, incompleteness, or inconsistency—can produce unreliable, biased, or even harmful outputs. The oft-cited mantra "Garbage in, Garbage out" rings especially true here: if GenAI models are fed flawed information, the resulting insights and recommendations may lead to costly errors, eroding both employee trust and operational efficiency. 

A persistent issue for many organizations is the existence of legacy, unstructured, or poorly labeled data. Integrating these disparate sources into a coherent, AI-ready dataset is akin to constructing a house on a shaky foundation. Without robust data integration and labeling, organizations risk feeding their GenAI models with inconsistent or irrelevant information, decreasing effectiveness, and increasing the risk of unintended consequences.  

Existing data governance frameworks are often inadequate for the speed and complexity of GenAI, making it hard for GenAI to function effectively without a unified data layer. Balancing speed and control is also crucial, as GenAI enables rapid experimentation, but this can clash with traditional, slower governance processes.

There is much talk of data "siloes", but these isolated data domains were created for a reason and are retained because they still provide value. Legacy systems offer a proven, reliable, and cost-effective platform for core business functions, CRM systems provide a centralized store for customer data, Operational Technology enables automation and the optimization of physical processes, etc. These domains only become siloes when they are required – but unable – to exchange data between them.  

Organizations have been grappling with this issue for years, but the emergence of GenAI has shone a harsh and unforgiving light on this problem. One of the core use cases of GenAI is easing access to information, so any data siloes have a hugely detrimental impact on the value that the technology can deliver. Breaking down these siloes is, of course, complex – largely because each domain has its own protocols, standards, and eco-system of vendors: acquiring expertise in even one domain is challenging, but acquiring sufficient expertise in every domain that must be integrated is doubly so.  

Unique in the IT industry, Orange Business has acquired a detailed understanding of each of these domains, so it has an unrivalled capability to bring all the data in these systems together and break down the silos that inhibit the creation of value from GenAI. It also empowers organizations to unlock the full potential of Generative AI by helping clients to prepare, secure, and govern their data, making sure it is reliable, compliant, and ready for GenAI innovation. This comprehensive support enables businesses to confidently address regulatory requirements and maximize the value of their data for competitive advantage. 

Additionally, Orange Business provides robust, scalable, and flexible digital infrastructure alongside its expert data management services. Our cloud-based solutions (delivered on a sovereign basis if required) deliver the high-performance computing power required for AI workloads, while our vendor-agnostic approach ensures seamless integration across different technologies.

Implementing GenAI for employee experience is a strategic endeavor fraught with significant data-related challenges. Ensuring data quality, security, and seamless integration, while addressing bias, privacy, and ethical considerations, is essential for realizing the true promise of GenAI. Organizations that invest in robust data governance create a foundation for responsible and effective GenAI adoption—one where data is not just an asset, but a resource that powers future innovation.  By proactively tackling these challenges, business leaders can harness the power of GenAI to create more engaging, equitable, and effective workplaces.