Three Data & AI Trends for 2025
Paul PETON is currently the Operations Director at MDW France, leading the Data & AI offering based on a dual partnership with Microsoft and Databricks. With 20 years of experience in the data field, he has worked on and now designs data platform architectures, keeping pace with technological advancements. A Microsoft MVP since 2018, he also provides expertise in the field of Artificial Intelligence.
The world of information technology is characterized by cycles and disruptions — essentially, a revolution in every sense of the word. Understanding the evolution of this sector, as well as its impact on organizations, requires staying grounded while keeping an eye on the emerging stars of the Data & AI galaxy.
In the closely interconnected fields of Data and AI, 2024 solidified the unified cloud platform strategy represented by Databricks, Snowflake, and the latest player, Microsoft Fabric. Simultaneously, generative AI models continued to push boundaries, offering increasingly powerful capabilities and embracing multimodality — moving beyond text to include images and videos. However, these models are still fundamentally based on Transformer mechanisms, with lengthy inference processes traversing entire neural networks to generate each token. The global scale adoption of generative AI may face constraints due to the production of essential GPUs and their significant energy costs.
December 2024 marked China’s entry into the global generative AI market with DeepSeek v3, a conventionally designed LLM featuring intriguing optimizations and offered as open source. Yann LeCun continues to critique the limitations of LLMs, as highlighted in his article on objective-driven AI and the release of a model like V-JEPA in 2024. However, these developments reflect a longer-term vision.
Now, let us explore possible trends for 2025.
The Keyboard, This Archaic Relic
Today, multiple generations coexist: those who discovered microcomputing (but wrote with fountain pens in school and perhaps used the optical pen of the Thomson MO5), the “Thumbelina” generation described by Michel SERRES, typing with two thumbs on a smartphone screen, and a new generation that naturally adopts touch-screen swiping from early childhood (always follow the recommendations regarding children and screen time!).
Except for the almost extinct stenographer-typist profession, each of these generations inevitably faces the challenge of writing efficiently with a keyboard. With practice, habits form, and while some nostalgics may cling to keyboard shortcuts, typing speed will never match the efficiency of oral expression.
In 2025, technology has advanced to the point where we can now talk to our devices and, more importantly, receive responses with latency close to that of human conversation. This new approach aligns with Satya NADELLA’s prediction that “SaaS is dead.”
GPT-4o Realtime
Azure’s cognitive services, along with other providers, have offered Speech-to-Text and Text-to-Speech models for years, allowing for simulated oral conversations. However, achieving satisfactory latency times had been elusive — until now.
With its realtime API, OpenAI finally delivers an efficient Speech-to-Speech solution. Imagine being able to interrupt a conversational robot or reference elements already discussed in the conversation. On its side, the bot can incorporate Retrieval Augmented Generation (RAG) to enrich its responses with external information. Moreover, synthetic voices now feature increasingly realistic and diverse intonations, shedding their overly robotic tone.
The GPT-4o realtime API is available through the Azure AI Foundry portal.
Dynamics 365 Contact Center
Dynamics 365 Contact Center is an integrated solution designed to streamline customer interaction management across multiple channels, including phone calls, emails, live chat, and social media. With artificial intelligence, real-time analytics, and native integration with other Dynamics 365 applications, Contact Center helps businesses personalize interactions, enhance customer satisfaction, and optimize team productivity.
Beyond Microsoft’s CRM, Contact Center is versatile and can connect to any CRM system. It incorporates a voice-enabled conversational chatbot that leverages data from connected sources. Conversations are stored and summarized through Copilot actions, offering operational insights that become invaluable when a human agent takes over customer-related tasks.
Everyday tools like Outlook and Teams also interact seamlessly with Dynamics 365, enabling features such as email drafting based on prior exchanges or meeting summaries to continuously enrich customer knowledge bases.
The complete set of features for Dynamics 365 Contact Center is available on this webpage.
Agents Are Reluctant Talkers
Five years after the first wave of conversational bots in 2017, ChatGPT redefined our relationship with chatbots by offering a smooth, guided, and non-blocking user experience. Whether in private or professional contexts, we have massively adopted these assistants for translation, summarization, information retrieval, and other tasks through intuitive chat interfaces.
While useful, these chat interactions remain user-focused. True productivity gains in the enterprise require task automation. This is the domain of Robotic Process Automation (RPA). However, traditional RPA tools struggle with scenarios requiring “thinking,” which has so far been the domain of humans. The power of language models combined with connectors or APIs has expanded the horizons of automation.
Microsoft Research’s AutoGen framework describes an agent as an “entity capable of sending messages, receiving messages, and generating responses using models, tools, human input, or a combination of these elements.” This abstraction allows agents to model real or abstract entities (people, algorithms, etc.).
The emergence of Large Language Models (LLMs) initially challenged developers due to their lack of repeatability, as traditional testing processes favor predictable outcomes. However, advanced models like OpenAI’s o1 and o3 simulate reasoning in complex contexts, enabling tools to tackle analysis and decision-making processes requiring high levels of human expertise. This paves the way for automating not just simple, repetitive tasks but also ambitious, high-value scenarios. Of course, new safeguards will be essential.
At the Ignite 2024 conference, Microsoft outlined its agent strategy by introducing no-code/low-code approaches via Power Platform’s Copilot Studio, complemented by pro-code capabilities through Software Development Kits (SDKs). The vision is to democratize the power of agents with the simplicity of the Power Platform while supporting more complex scenarios with advanced tools.
This dual approach reflects a future where automation and conversational AI transcend routine tasks to deliver meaningful innovation and business value.
Azure AI Agent Service
In 2024, the Azure AI Studio portal was rebranded as Azure AI Foundry, with expanded features, particularly in the field of AI agents. Microsoft’s official documentation describes an AI agent as an “intelligent microservice” capable of answering questions (using Retrieval-Augmented Generation, or RAG), performing actions, or fully automating workflows. These agents combine the power of generative AI models with tools that allow them to access real-world data sources and interact with them effectively.
Tools are a critical component of agents, and the platform manages tool calls and responses seamlessly. These tools can be user-defined or pre-integrated options such as Bing Search or Azure Functions. Execution steps are transparent, allowing users to understand how the agent arrives at its outcomes, which adds a layer of interpretability and trust to the process.
This screenshot is from Microsoft’s official video introducing the Azure AI Agent Service.
Metadata: The New Gold Mine to Explore
The Big Data movement, once revolutionary, now feels like ancient history given the rapid evolution of the tech world. Collecting and leveraging data has become a common goal for most organizations, even those without fully implemented data-driven strategies. This was followed by the Data Mesh movement, aiming to better distribute data ownership and exploitation responsibilities through data products. This governance model also incorporates observability to raise alerts about data quality issues.
However, a persistent challenge remains: finding the right data at the right time (and ensuring it’s up-to-date) while understanding the underlying definitions (calculation rules, interpretations, associated KPIs). Tools like Master Data Management (MDM) systems and Data Catalogs, which include discovery processes for scanning connected data sources, aim to address these challenges.
The Chief Data Officer’s (CDO) daunting task is to document data sources, tables, or fields in a manner understandable to business users — in natural language. This is where Large Language Models (LLMs) can act as accelerators. LLMs can handle repetitive tasks such as describing data sources or even performing initial content verification by identifying inconsistencies in textual fields or values.
Once metadata quality improves, ambitious projects like natural language to SQL become feasible. Such tools allow non-developers to translate their business needs into SQL queries, significantly reducing the workload on IT departments. This shift would enable faster, more democratized access to data insights while freeing up IT resources for more complex tasks.
The combination of high-quality metadata and LLM capabilities marks a transformative step forward, enabling organizations to unlock the full potential of their data ecosystems.
Microsoft Purview for Data Quality
Microsoft Purview is a comprehensive data governance and risk management solution designed to help organizations protect, manage, and leverage their data effectively. Purview provides tools to catalog, classify, and track sensitive data across hybrid and multi-cloud environments. With AI-driven and automated features, it ensures regulatory compliance, mitigates data-related risks, and fosters a culture of governance within enterprises.
The Microsoft Purview Data Profiler generates insights by analyzing a random sample of one million rows from a table. It computes key metrics such as measures of central tendency and dispersion for numerical data, as well as frequency distributions for categorical values. This profiling is an essential precursor to defining robust data quality rules.
Microsoft Purview Data Quality evaluates the quality of data using native rules, which can include regular expressions or even AI-generated rules. These rules generate quality scores based on the entirety of the data rather than a sample. If the scores fall below expected thresholds, alerts are triggered, leading to actionable steps. These actions can be assigned to designated Data Quality Stewards within Microsoft Purview.
Demonstrations of these capabilities are available in a video on the Data Chouette channel.
Looking Ahead
As we approach the end of 2025, we will see whether these trends hold true. The IT world evolves at an exponential pace, and unexpected developments are always possible. However, staying grounded is essential amidst the hype surrounding certain technologies. The focus should remain on creating value for organizations, whether by reducing costs or enabling new products or services. As Rahul Vohra, CEO of Superhuman, aptly puts it:
“When AI works, people stop calling it AI.”
