Data Pulse by fifty-five: Your Monthly Insights on All Things Tech, Data & AI
[Decoding Tech]
Microsoft and European Digital Sovereignty
By Tiyab K.
Microsoft recently announced an unprecedented set of commitments aimed at strengthening its position in the European cloud market. This initiative comes amid stricter regulations, geopolitical tensions between the European Union and the United States, and increased competition around sovereign cloud offerings.
Here are my key highlights from the announcement:
- Microsoft commits to legally challenging any attempt by any state, including non-European ones, to suspend or shut down its cloud infrastructure in Europe.
- The company’s European cloud operations will be overseen by a board of directors composed exclusively of EU citizens and governed by European law.
- Microsoft’s offering is structured around several options: public cloud in the EU, sovereign cloud (Bleu partnerships in France and Delos in Germany), and hybrid solutions with local partners.
- Microsoft is removing costs associated with transferring data out of Azure, thus facilitating data portability – no more data exit fees.
- Azure AI platforms now welcome a wide range of open source models, particularly from the European ecosystem (Mistral, Hugging Face).
So what does this all mean from a strategic analysis viewpoint?
→The announcement aligns with major European regulations currently coming into force (DORA, CRA, NIS2, Data Act) – Microsoft seeks to position itself as a trusted provider that is "compliant by default."
→ In a tense climate between Brussels and Washington, the company adopts a moderate and cooperative stance, contrasting with the confrontational responses of other American digital players.
→ By anticipating European market expectations regarding sovereignty, Microsoft is taking the initiative against emerging offerings from competitors (AWS European Sovereign Cloud, Google/Thales S3NS, OVHcloud, T-Systems).
→ The increase in Microsoft’s datacenter capacity in Europe (target: +40% within two years) aims to meet the growing demand for AI infrastructure within a framework compliant with EU energy and regulatory policies.
Microsoft is not merely complying with European regulations: the company is transforming regulatory constraints into strategic leverage. It positions itself as a central player in Europe's digital future by integrating principles of sovereignty, local governance, and transparency as foundations of its offering. This announcement marks a significant turning point in redefining the balance between hyperscalers, European regulators, and national markets.
[Radar 2025 - which trends are we monitoring]
What GenAI reveals about your data quality (and what to do about it)
AI’s potential is huge … but only if the data can keep up.
We’ve entered a new cycle of expectations: organisations want to embed AI into marketing and customer experience. Use cases are multiplying like predictive scoring, personalization at scale, content generation, next-best actions... As LLMs grow, the promise seems closer than ever.
But the elephant in the room hasn’t moved: data quality. And more specifically what we see in the real world with fragmented sources, data discrepancies, poor taxonomy alignment, inconsistent identifiers, and low trust in existing metrics.
Trying to integrate AI on top of that is like designing a self-driving car for roads that haven’t been paved yet. AI requires more than just data: it requires structured, unified, usable data.
That’s where many businesses are realizing they may have skipped a step. They have data, but lack foundations and readiness.This disconnect leads to frustration with proof-of-concepts that never scale, hallucinating models based on unclean inputs, and marketing teams that lose trust in automation because the output doesn’t reflect the reality on the ground.
At fifty-five, we see this pattern repeatedly and we know the way forward starts with reconciling ambition and foundation. And our approach is pragmatic:
- Audit the data layer for AI readiness (identity, completeness, freshness, semantic coherence)
- Design scalable data stacks that support AI use cases (segmentation, content, decisioning)
- Orchestrate cross-functional governance between data, marketing and product teams to ensure long-term performance
Recommended by LinkedIn
In short: we make AI possible by making the data work first.
[Behind the scenes]
From Query to Insight: Building a Natural Language Web Analytics Agent with Gemini and MCP
We recently built an AI-powered web analytics agent that simplifies access to complex data by allowing users to ask questions in natural language. In practice, this means that one can ask “How many visitors did we have last week?” and receive clear, conversational answers. Our agent combines Google's Gemini AI with a universal communication standard called the Model Context Protocol (MCP) - a standardized framework that streamlines dialogue between AI models and external analytics tools like Google Analytics 4, Search Console, and PageSpeed Insights. By translating user queries into structured API calls and responding with user-friendly summaries, the agent enables non-technical teams to make decisions faster and more intuitively.
The agent is not without limitations, of course – real-time responsiveness is still constrained, and predefined data schemas are necessary. Yet this approach represents a significant step toward more accessible and intelligent analytics solutions: the agent is particularly well-suited for teams seeking to simplify reporting workflows and gain actionable insights from their web data with minimal manual effort. No need to call upon a data analyst who would have to take time out of their day to give you the data you need, or even to navigate an intricate dashboard – you can simply ask the agent and get your answer in seconds.
All the technical aspects of this project are available here – feel free to give it a read!
[Tech inspiration]
A creative prompt-writing helper for a more mindful use of Gen AI
By Julie Cazaux
As LLMs get integrated into our everyday workflows, prompting is proving to be a bit of a challenge: in many cases, users try and fail to get the result they want over and over again, patience runs thin, adoption gets delayed, and resources end up wasted overall. By “resources,” I don’t mean just time – there’s also the actual cost of using an LLM professionally and the energy required to run the model. As a reminder, and as we will explore in more detail in our upcoming study co-written with the Brandtech Group on the environmental impact of Generative AI, every request sent to an LLM represents a cost when it comes to electricity, and that cost is even higher when using models to generate visual assets.
So, to generate visual content more responsibly, we recommend refining your prompt as much as possible before sending your request. This will help both you and the LLM be more efficient and accurate. And while there are plenty of guides available to help you learn how to prompt, new tools are also coming out to help. For instance, The Brandtech Group's very own James Dow built a GPT that serves as a prompting wizard, refining any basic creative prompt into something much more useful and detailed for creative generation through AI tools. As generative AI tools become part of daily lives, we will likely see new GPTS – and even AI agents, eventually – being created to promote efficiency and adoption.
Here’s how the prompting wizard works, in practice: you can start with typing something simple like “I want a promotional image for an upcoming conference dedicated to Generative AI in retail.” And in just a few questions about your preferences, the tool will provide you with a highly detailed prompt, formatted for AI: A conceptual portrayal of next-gen retail: a stylized editorial scene where shoppers are bathed in shifting light maps that resemble neural activity, walking through an immersive space with sculptural shelves and dynamic projections of real-time analytics, surreal elements like transparent mannequins with circuit-like skin, layered environments blending physical products with floating icons and glowing connection nodes, soft-focus figurative details, lighting in metallic tones, and a pastel dream color grade for a surreal futurist finish.
[Expert POV]
Fivetran x Census: a strong signal for the Composable CDP model?
The recent acquisition of Census by Fivetran marks a notable shift in what we call the modern data stack landscape. While recent CDP acquisitions (mParticle, ActionIQ, Lytics) have largely focused on pushing some CDP capabilities closer to marketing activation, this move takes the opposite path: a vertical integration upstream, toward the data ingestion layer.
Census has been a key player in reverse ETL, enabling organizations to sync modeled customer data from the warehouse into tools like CRMs, ad platforms, and marketing automation tools, effectively powering Composable CDP architectures.Fivetran, on the other hand, automates data ingestion and normalization from hundreds of operational systems into cloud data warehouses, like BigQuery, Snowflake, Redshift, etc.
This merger brings the two ends of the data flow together aiming to control the full pipeline:
from source systems to activation, with stronger governance, lower latency, and a unified data flow across the stack.
This acquisition reinforces several market signals we’ve been seeing at fifty-five:
- The Composable CDP model is gaining traction as a scalable, modular system;
- The real value no longer lies in UIs or proprietary storage but in fluid orchestration across tools;
- Standalone reverse ETL players are under pressure to move upstream and integrate more deeply with data ingestion and transformation;
- Infrastructure vendors like Fivetran are now positioning themselves to own the full customer data lifecycle, not just pipeline hygiene.
Ultimately, this deal shows that data movement is becoming strategic and that CDP functionality is no longer confined to the marketing layer.
France Partner Manager chez Fivetran
2moWell written Philippe Kuhn !
Senior Enterprise Account Executive at Fivetran
2moYohan Busidan Virginie Brard Irina Slavitch Marshall Wilson Johann Beaud
Ryan Pestano Michal Niec