Alejandro Cano

A few days ago, during a roundtable on data governance, someone in the audience asked a question that stuck with me:“What can I do about the resistance I get from my colleagues when I try to apply governance and quality policies? I feel like everyone in the office hates me.” That reflection was the main reason for this post. Because beyond the processes, the biggest challenge when implementing data governance may not lie in the data itself, but rather in how to support the people working with it. The reality is that introducing data governance probably won’t make you the most popular person in the team or across departments. And that makes sense—after all, things have worked "in their own way” for years. Decisions have been made based on questionable data quality, reports have been generated with no traceability, and then someone comes along and says everything needs to change. The problem is that, in most cases, working "in their own way” comes at a cost. It leads to poor decision-making, wasted time and money, and in the worst-case scenario, legal trouble—as we’re already seeing with new regulations like the AI Act. But often, the impact isn't clear until it’s too late. So, how do we avoid the dreaded question: Where did that data come from? Why does every effort to organise information and improve its quality seem to turn us into the enemy? I want to share what I’ve learned throughout this journey, and how, at Telefónica Tech, we’ve managed—or at least strive—to make data governance not a blocker, but a builder of smart shortcuts for everyone. 1. The problem isn’t the data—it’s the change The resistance people feel usually isn’t about data or new ideas—it’s fear.Fear that someone will question their work.Fear of not knowing how to justify the way they’ve done things for years.Fear that change will expose flaws no one wants to face. And that’s when rejection shows up. Not because of the data, but because of what it means to review it. To question it. To correct it. But within that discomfort lies a huge opportunity. Because when someone feels heard, supported, and part of the solution—not the problem—change stops being a threat and starts becoming a natural evolution. So the real question is: how do we shift from resistance to momentum? The answer may lie in how we make people feel about the change. The Data Governance Office is here to help improve what already exists, not to judge it. It’s about building, not pointing fingers. 2. You have to sell the idea—not impose it When you first started talking about data governance, you probably thought everyone would immediately see its value: more control and security, better data quality and information, and therefore better decisions. What’s not to like? But often the reality is different. It’s not that people don’t want more insight into their data and processes. It’s that they feel you’re adding more weight to their already overloaded world—new processes, new validations, new reports... It sounds like more work, not like a solution. And that’s where many of us go wrong: we try to impose change as a logical obligation, when real change needs to be an emotional choice, something people naturally want to take part in. The key isn’t in setting rules. It’s about showing the benefits. Not in demanding controls, but in telling stories: how a crisis was avoided by catching an error in time, how a client was saved from a bad decision thanks to clean data, how a process became faster because the data was reliable from the start thanks to that extra governance effort. When you pitch data governance as a tool that reduces the burden rather than adds to it, the narrative changes. It goes from being enforced to being convenient. And in that subtle shift in perception lies the difference between rejection and long-term change. What if we started telling more success stories instead of overwhelming people with rulebooks? 3. When everything is urgent, nothing is important In our experience at the Data Governance Office, we’ve learned a crucial lesson: knowing that data governance is necessary isn’t enough. What truly makes a difference is how you introduce it to the organisation—which is why methodology matters so much. At first, it’s normal for governance initiatives to be seen as roadblocks. Again, it’s not about the data itself, but the effort and change they seem to bring along. That’s why it’s essential to follow a practical, proven methodology—one that respects people’s pace, anticipates resistance, and slowly shifts initial perceptions. We don’t impose rules → We build trust.We don’t force processes → We show real benefits. It all comes down to three basic principles: Understand that change is emotional before it is technical. Work as a team from the Governance Office, giving visibility to the achievements made in collaboration with different areas. Prove—through visible results—how data governance makes daily work easier, not harder. Each phase of implementation is designed to create quick, tangible wins that people can feel and see. This makes it easier for teams to get involved and stay motivated throughout the change process. Because when data governance is seen as an ally, that’s when its true value begins to take root in an organisation’s culture. And that’s where the methodology we apply shows its real strength: it doesn’t just transform data—it transforms the way people work, decide, and trust. IA & Data Data Governance: a great ally to put limits to Artificial Intelligence October 31, 2023

As we saw in the previous post, Generative AI contributes to reducing food waste in homes, which represents about 55% of all food waste. Industry and the supply chain contribute another 45% to food waste, so investment in process optimization and food utilization is essential to reducing waste. The second installment looks at how data-driven approaches and AI systems are key to addressing and improving this issue. Solutions powered by Generative AI There are already several approaches in the use of Generative AI to help reduce the remaining 45% of food waste from food supply chains, as intuitive tools capable of contributing to the development of new solutions, as well as providing real-time predictive answers to optimize processes. A large percentage of this food waste is produced during food processing and packaging by specific machinery for these sectors. Solutions aimed at monitoring the operation of this machinery and its maintenance in a predictive way, driven by these generative AIs, have already been proposed to improve this situation. In this context, AI based on Generative Adversarial Networks (GAN) or Recurrent Neural Networks (RNN) play a crucial role in generating synthetic data sets to train new models. This synthetic data, derived mainly from historical machine behavior, would expand both the quantity and quality of data available. In food handling, synthetic data makes it possible to anticipate and avoid problems and failures by providing information about possible scenarios and critical scenarios. This would allow models to gain deeper insight into machine performance and state during various processes. Additionally, these synthetic data sets could introduce information about possible scenarios that the machinery could face, facilitating the development of models capable of foreseeing incidents and preventing critical situations or failures during food handling. Anticipating potential failures These models, which have been exploited over time by the more traditional approaches to Artificial Intelligence, maintain their relevance by continuing to be used to build predictive models focused on possible failures in the processes executed by machinery during food handling. Its continued application in this direction. The continued application of this predictive maintenance approach in addition to the great possibilities offered by the most current Generative AI-based assistants would demonstrate the effectiveness, advancement, and adaptability of Artificial Intelligence to address specific challenges in the management and optimization of processes in the food industry. The use of monitoring systems can reduce machinery breakdowns by up to 70%, produce less organic waste, and enhance the efficiency of food use. In case of process breakdowns or failures, these assistants could report on the situation, propose solutions, answer user questions, and even provide information on who to contact or how to proceed, provided that the knowledge of this assistant has been extended by providing specific functional or technical information. It is estimated that these monitoring systems could reduce the breakdowns of this type of machinery by up to 70%, which in the food sector would have a positive impact on avoiding the generation of organic waste resulting from poor food utilization, poor sealing, or simply poor organization. Image by Freepik. Regarding this last point, and in relation to the home improvement framework, companies in the food sector can also take advantage of text assistants, with a significant opportunity emerging through advanced techniques such as RAG (Retrieval-Augmented Generative). These techniques allow customizing the knowledge of the assistant or LLM so that it is able to address specific aspects at a higher level of detail, being able to focus it in a particular way to the context of food waste, configured to understand and process relevant information for real-time planning. Dynamic data-driven planning In the area of food waste reduction, these assistants could play a crucial role in performing dynamic planning based on various input data that is continuously generated. These assistants, by constantly analyzing data related to inventory, demand trends, expiration dates and other factors, could provide real-time recommendations and strategies to optimize resource management. Generative AI can be used to indicate adjustments to production based on current demand and forecast excess inventory to prevent food waste. Together with predictive analytics solutions, for example, which were already being used in "traditional AI," the assistants or LLMs could suggest adjustments in production according to current demand, foresee possible inventory excesses that could lead to waste, or even provide advice on the efficient distribution of perishable products. In addition, they could identify patterns and alert on possible areas for improvement in handling and storage processes, thus contributing to operational efficiency. Where are we headed? The use of Artificial Intelligence and idea generation technologies such as Generative AIs in the form of virtual assistants offer a promising solution to address multiple challenges, including food waste both at home and for industries. We are taking significant steps towards building a sustainable future by using LLM-based systems that are available to all and capable of creating innovative recipes, suggesting creative ways to leverage ingredients, or assisting with planning. Generative AI allows us to create innovative recipes, leverage ingredients creatively and assist with planning. Not only can we reduce the environmental impact associated with food waste, but we can also improve our processes and use of resources and have a positive economic impact by incorporating and adapting these technologies into our businesses or industry's daily practices. The collaboration, not substitution, of human creativity and adaptability with the power of generative AI can set the tone for a context where food efficiency and innovation are intertwined to build a path towards a more sustainable future. AI of Things How is digitization being addressed in the food industry and what are the benefits December 5, 2023

What problem are we facing? The latest European Commission report on food waste shows that 57 million tons of food and beverages were wasted in Europe in 2022. This is equivalent to 127 kg of waste per person per year on average, with a market value estimated at 130 billion euros according to Eurostat. The data illustrate the extent of Europe's food waste problem and the need for reduction measures. With 55% of waste coming from households and the remaining 45% distributed throughout the supply chain, this picture offers two clear areas for improvement: In the management of food by households, encouraging greater awareness and more sustainable practices In industry, which can play a crucial role through greater involvement and investment in process optimization and the recovery of food that is unfit for consumption. In the face of major challenges such as these, there are, fortunately, equally significant solutions. Data insights and Artificial Intelligence emerge as key tools to effectively address and improve these everyday issues. How does Generative AI help us reduce food waste at home? Generative AI is emerging as a resource of great value in multiple fields, becoming more and more integrated in our daily lives. This everyday use is possible thanks to the fact that tools such as Chat-GPT, Bing, or Llama 2 are available to the general public, and, in many cases, free of charge, with limited but effective functionalities to solve specific problems. We find then in these tools an innovative and democratic solution, since, in addition to being accessible, their use can be very simple if we find the right key, because we can take advantage of them using prompts or messages that guide the model to generate answers, through a daily conversation. Considering that the main cause of food waste in the domestic environment is the lack of planning and organization in the menus consumed over time, which can lead to the addition of unnecessary quantities or products in the shopping list, it is possible to use technologies that are based on text Generative AI such as GPT or Bing. Doing this kind of planning can take time that is not really necessary to invest if these conversational text assistants or LLM can do it for us. If we interact with the LLM with the following prompt: > Please create a weekly menu that avoids food waste.| We can obtain a useful but imprecise result if we have more specific needs, so it is important to add complexity to the prompt. How can we improve the above interaction? The answer, a priori, may be simple and it is none other than to add complexity to the prompt used, since the more detail is provided to the assistant about what is being asked, the more detail the response obtained will have based on those expectations resulting from more precise objectives. However, there are multiple ways to add complexity and increase the quality of the instructions provided to the LLM; it is not limited to a single way of providing instructions or just adding more text, but there are several options available that can better fit what is required: Role Prompting: gives more context to the LLM by making him/her assume a specific role, such as a specialized nutritionist, so that if you want to have a conversation with the chat, he/she will be able to contextualize all your answers: > Act as a nutritionist with 5 years of experience. Please create me a weekly vegan menu for two adults and two children, with approximate amounts. Gluten free. High in protein. No industrial or ultra-processed foods and easy to adapt to batchcooking.| Few Shot Prompting: it is the favorite method for some experts as it is able to give the most accurate answer in terms of expected format and information. It consists of providing the chat with some examples so that it learns to provide a specific answer. In this way, the LLM could be given an example of a menu with a specific format on which to make certain modifications, for example, related to some objective we hope to achieve, to diversify the options, to make it vegan or to provide us with specific recipes. > I don't have any mascarpone cheese. Can I use Philadelphia cheese to make cream for carrot cake?| These prompts can also be used to learn more about other ways to avoid food waste at home, adapting to each situation or need. From knowing what tips would be the best to prepare the weekly Tupperware in an afternoon, to designing a recipe to make the most of the last ingredients we have in the fridge, preventing them from going to waste. One of the advantages of this type of conversational assistants is that they can be used intuitively, which facilitates the resolution of small day-to-day doubts. Therefore, these prompt engineering concepts will be of great relevance in order to get the most out of LLMs, obtaining the most accurate answers possible, especially in such sensitive topics as health issues, emphasizing that this is only a support for day-to-day life and does not replace health or food professionals. ◾ In the second part, we will see how in the field of industry and the supply chain, data and artificial intelligence are also key tools for reducing food waste. Telefónica Tech AI of Things How to create realistic images using Generative AI November 23, 2023