Databricks Data+AI World Tour 2025: How Data Becomes a Strategic Competitive Advantage

On December 2, 2025, Databricks brought together France’s data and AI ecosystem for a day focused on transformation. The program included over 40 sessions, client testimonials from Decathlon and STMicroelectronics, and the latest product innovations on the platform, from Lakehouse to Lakebase, including Agent Bricks and generative AI.

Between governance strategies, AI democratisation, and field testimonials, Data+AI World Tour2025 confirmed a trend: data is no longer a technical topic but a strategic competitive lever that radically transforms organisations.

Devoteam and Databricks: Recognised Expertise Serving Data and AI Transformation

Devoteam has established itself as a leading Databricks partner, with a team of experts 100% dedicated to the platform. Supporting around fifteen major French accounts from industrial and service sectors, Devoteam intervenes throughout the entire lifecycle of Databricks projects: from initial data platform integration and cloud migration, to large-scale deployment of advanced analytics and AI use cases, including data governance via Unity Catalog.

This high-level technical expertise is embedded in Devoteam’s DNA: combining sharp technological skills with customised support capabilities to transform companies’ data ambitions into operational solutions. This dual competency, deep Databricks mastery and sector-specific business knowledge, enables Devoteam to develop innovative solutions like Cyberlake, which leverages Databricks’ power to address critical cybersecurity challenges for organisations.

Cyberlake: An Innovative Cybersecurity Solution on Databricks

Cyberlake is a solution developed by Devoteam that combines Devoteam’s cybersecurity expertise with the power of the Databricks platform. The objective is twofold:

• Improve clients’ security posture
• Drastically reduce the total cost of ownership (TCO) of cybersecurity solutions

Facing the exponential increase in companies’ cloud attack surface and intensifying geopolitical risks, Cyberlake offers an economical alternative to traditional proprietary solutions (Splunk, Palo Alto, Checkpoint). Storage on Databricks costs 50 to 100 times less than on Splunk, while offering AI and machine learning capabilities impossible to deploy on traditional cyber platforms.

An Intelligent Pipeline Leveraging the Entire Databricks Stack

Cyberlake uses all components of the Databricks platform:

• Log ingestion and storage (notably via the Splunk connector)
• Workflow transformation and orchestration
• Machine learning to train client-specific models
• Generative AI for vectorisation and semantic analysis
• MLflow to manage model lifecycle, GenAI SQL for natural language queries
• AI/BI for reporting

The process unfolds in three stages:

1. Classification of suspicious logs via machine learning
2. Vectorisation and comparison with referenced attack scenarios (MITRE ATT&CK, Sigma, VirusTotal)
3. Automatic generation of action plans using LLM models fine-tuned for cybersecurity

This approach detects known attacks but also anticipates zero-day attacks through semantic pattern matching with generative AI. Devoteam positions itself as a key Databricks partner in this area, being the only integrator with dual cybersecurity and data competency, with ongoing projects at major French industrial clients.

Databricks Data+AI World Tour 2025

Guillaume Brandebourg, Country Manager France at Databricks, opened Data+AI World Tour by highlighting the company’s commitment to France with over 250 direct jobs created and a dynamic ecosystem bringing together large enterprises, startups, and scale-ups.

Placed under the theme of curiosity, this day offered over 40 sessions and client testimonials, notably featuring presentations by Didier Mama (Chief Digital Officer at Decathlon), Jean-Luc Decaux (CIO at STMicroelectronics), and Yannick Nyanga, a former international rugby player, on the connections between sports and data.

Strong Growth in Europe

Samuel Bonamigo, General Manager EMEA, shared figures illustrating Databricks’ momentum: over 2,000 employees in Europe, including Product and R&D teams based in Amsterdam, Berlin, and Belgrade, and over 300,000 monthly active users on the platform.

Databricks is heavily investing in its partner ecosystem (system integrators, ISVs, data partners) and training via the Databricks Academy and partnerships with European engineering schools.

Major Strategic Partnerships

Several announcements were highlighted:

• Partnership with London Stock Exchange to improve financial data quality
• Strategic collaboration with SAP resulting in Business Data Cloud
• Agreement with OpenAI enabling access to the best technologies (GPT-5, Anthropic, Google Gemini) directly from the platform

Governance and Security at the Heart of the Offering

Samuel Bonamigo emphasised the importance of data governance and security in the AI era:

“No governed data, no secure data, no AI.“

Unity Catalog, Databricks’ 100% open-source solution, addresses these challenges and positions the company as a leader according to Gartner’s Magic Quadrant.

Product Vision: From Lakehouse to Data Intelligence Platform

David Meyer, Senior Vice President of Product at Databricks, presented the company’s strategic vision and the evolution of its platform since 2019.

The Challenge of Data Fragmentation

David Meyer highlighted a reality shared by most organisations: despite AI’s transformative potential, putting it into production remains difficult. Companies juggle multiple information silos, on-premise and cloud data warehouses, separate machine learning systems, SaaS applications, each storing data in proprietary formats with their own security policies.

Lakehouse Architecture: A Conceptual Breakthrough

Facing this complexity, Databricks introduced the Lakehouse concept in 2019, initially met with industry scepticism. The principle: centralise all data in low-cost object storage, in open formats (Delta Lake, Iceberg), and unify governance in a single layer. This approach enables data use without copying between different modalities (analytics, AI, data science) while maintaining consistent governance.

Unity Catalog: Beyond Traditional Governance

Unity Catalog, presented as Databricks’ most strategic investment, goes far beyond traditional data catalogues. While the latter are limited to structured data (tables, columns), Unity Catalog governs the entire lifecycle: data, machine learning models built from that data, and agents using those models. This approach enables complete lineage tracking: a data change can impact a model, which in turn affects an agent’s performance, while applying cost controls and quality monitoring across functions.

Agent Bricks: Simplifying AI Agent Production

Agent Bricks, the platform’s new composable agent layer, includes:

• Multi-Agent Supervisor: orchestrates agents from different systems and protocols (MCP)
• Knowledge Assistant: queries all enterprise documentation in natural language
• Genie: an agent dedicated to data analysts for querying structured data
• Data Science Agent: automatically generates analysis notebooks (forecasting, fraud detection)
• Document Processing: reliable data extraction from PDFs and other documents

LLM Judges: Continuous Self-Improvement

David Meyer introduced the concept of LLM judges, the cornerstone of Agent Bricks. Based on the principle that it’s easier to evaluate a quality response than to produce one (like recognising a masterpiece is easier than painting it), the system uses inexpensive LLMs to automatically score each agent response.

This continuous evaluation enables automatic system optimisation: re-segmentation of vector indexes, selection of the best models for each task, quality improvement while reducing costs. The results are compelling: at AstraZeneca, on 400,000 clinical trial documents, Agent Bricks achieved quality superior to GPT-4o Mini at significantly lower cost.

Agent Bricks’ central principle is separating business definition, controlled by users in simple business language, from underlying technical AI complexity, managed automatically by the platform. Agents thus improve with each use, without additional intervention from technical teams.

STMicroelectronics: Transforming Data into Competitive Advantage

Jean-Luc Decolle, CIO of STMicroelectronics, illustrated how the semiconductor giant made data a strategic differentiation lever.

A European Semiconductor Leader

STMicroelectronics employs 50,000 people worldwide, including 9,000 R&D engineers. It has 200,000 clients with the world’s largest semiconductor product portfolio (€13 billion revenue in 2023). As an IDM (Integrated Design Manufacturer), the company controls the entire value chain with 14 production sites, including two cutting-edge factories in Crolles and Agrate.

ST operates in four major markets: automotive, industrial/IoT, personal computing, and satellite communication. The company notably manufactures iPhone imaging sensors in Crolles.

From Automation Approach to Data-Driven Culture

Initially, data served to automate and control an industry operating 24/7. But after this automation phase, ST realised that the last competitive gains were hidden in the data itself.

The assessment is clear: “If you have siloed data, you cannot succeed,” he explains. The company shifted from an “Automation Driven” approach to a “Fact Database Driven Decision Making” culture.

Concrete Business Impacts

Operations complexity illustrates the challenge: creating a one-square-millimetre chip with 600 million transistors requires 40 weeks. By de-siloing Manufacturing, R&D, sales, and market analysis data, ST obtains tangible results:

• Real-time yield improvement in factories
• Dynamic tolerance adjustment during manufacturing
• Daily productivity gains through widespread prompt usage across all departments
• Innovation acceleration: freeing 10% of a Design engineer’s time enables introducing one additional product per year

Radical, Not Incremental Transformation

For Jean-Luc Decolle, the traditional approach of successive ROI-focused projects is unsuited to the current technological evolution. ST operated a complete “Greenfield” transformation, abandoning silos and scattered POCs for central AI governance with Unity Catalog and democratized data access.

The message is clear: “There are no more IT projects. There’s an enterprise-wide project.” This centralised transformation relies on a single Demand Management Process arbitrating all projects at the enterprise level.

Four Levels of Data Maturity

On this high-fidelity “Single Data Hub,” ST structured data exploitation according to four progressive levels:

1. Basic reporting with a single source of truth
2. Advanced reporting with correlations
3. Autonomous data analytics
4. “AI Driven” deep analysis with LLMs and conversational agents

A Strategic Partner Ecosystem

ST relies on a tight ecosystem, comprising Microsoft, Confluent, ServiceNow, and Databricks, to build an architecture based on Product Data. This approach enables the progressive decommissioning of legacy applications and a focus on high-value processes.

This transformation requires time: two years to establish the operating model, create convictions, and overcome resistance.

In conclusion, Jean-Luc Decolle used a maritime metaphor, encouraging the audience to cross the “Cape of Storms” of the data revolution to reach the “Cape of Good Hope” of competitiveness. His advice: favour a transformational leap over a sequential approach, too slow given market evolution.

Lakeflow, Lakebase and AI Democratisation

David Meyer continued his presentation detailing Databricks platform evolution and its most recent innovations to simplify data and AI access.

Lakeflow: From Spark to Simplified ETL

David Meyer recalled Databricks’ origins: “For the first years, until 2017-2018, we were primarily an ETL company.” The company built itself around Spark, created in Berkeley’s AMPLab by Databricks’ founders. Today, Spark is one of the three largest open-source projects worldwide (with Linux and Kubernetes).

Despite this success, Spark required significant technical expertise and wasn’t optimised for easily extracting data from databases and SaaS applications. Databricks recently contributed its declarative pipeline framework (formerly Delta Live Tables) to Apache Spark, making it fully open source and accessible to all.

Lakeflow now groups four key components:

• Connect: connection to all data systems (SaaS, files, databases)
• Jobs: large-scale ETL orchestration (Databricks launches over 40 million machines per day in the cloud)
• Spark Declarative Pipelines: simplified transformation pipelines
• Designer: visual interface with integrated AI

Designer notably enables describing desired transformations in natural language, or even providing an image (GIF or PNG) of the desired table structure, with the system automatically generating corresponding SQL code.

From Lakehouse to Lakebase: A Revolution for AI

David Meyer introduced a new concept he considers potentially even more transformative than Lakehouse: Lakebase. “If a Lakehouse is a data warehouse residing in your lake without copying, then a Lakebase is a database residing in your lake without copying.“

This innovation addresses a significant issue: traditional databases, although they have undergone considerable evolution since the 1970s, still retain a on-premise mentality that is unsuited to modern AI development. David Meyer states bluntly: “Your developers hate your databases. They don’t tell you because they don’t want to upset you, but cloning a database at scale takes hours, not minutes.“

Lakebase is based on Postgres, completely re-architected to separate compute and storage. Data is stored directly in the lake as open-source files, enabling:

• Instant creation: any size database available in under one second
• Instant cloning: copies available in under one second
• Blob storage cost when the database isn’t used
• Instant hydration upon first query

Copies for Vibecoding

This architecture radically changes AI application development. David Meyer cites Neon’s example, a company recently acquired by Databricks: over 80% of its databases are created by AI agents. “When you’re Vibecoding, code is sometimes bad and can corrupt your data. You must work on a copy.” With Lakebase, creating a thousand clones becomes possible at negligible cost. A thousand agents can work in parallel, and then successful branches are merged like code.

Data now flows without friction. Any Unity Catalog schema can be exposed via a Postgres interface with 9-millisecond reads. Conversely, operational data is available in real-time in Unity Catalog, with the same governance and access policies.

Databricks Apps: From Idea to Production

To democratize access to developed AI agents, Databricks launched Databricks Apps. Apps enable building custom applications directly in the platform. David Meyer highlights a recurring problem: “You develop your app with Vibecoding, you’re happy, then IT tells you it needs SSO integration, multifactor authentication, VPC policies… and your genius never goes into production.”

Databricks Apps solves this by natively integrating security, governance, audit, and compliance. Developers can use any Python or Node framework. And Databricks works with major Vibecoding platforms (Lovable, Replit, GitHub Copilot) to enable one-click deployment.

It’s Databricks’ fastest-growing product ever launched, proving the demand for a simplified AI application deployment approach.

AI/BI and Genie: Analysis for Everyone

Databricks’ Business Intelligence vision: a tool available to everyone without user license fees, with AI at the heart of the experience. Users can start with a phrase in natural language (English, French, Portuguese…) that will automatically generate a dashboard.

The major difference from traditional BI solutions with AI lies in data access depth. “If your BI is on a separate layer, the integrated AI is limited to what’s in that BI layer. With AI deeply integrated into the engine, you can access billions of rows and ask any question.“

Genie, Databricks’ conversational agent, is fully integrated via API to all business systems.

Decathlon: Data Serving Sports Democratisation

Didier Mama, Data and AI Manager at Decathlon, shared the transformation journey of France’s favourite sporting goods retailer, which for 50 years has made sports accessible to all.

A Reimagined Architecture for the AI Era

Decathlon has used Databricks since 2020, as part of a project launched by the CEO and CIO called “Edge of Access.” The objective: create value with data and accelerate AI.

At the time, the technical landscape was complex: a Redshift Data Warehouse, an emerging S3 Data Lake, and business applications on Google Cloud. “Use cases first required breaking down silos,” explains Didier Mama. “The decoupling between storage and compute was complex on the Redshift side, and we had anticipated AI acceleration.“

Data Factory teams already used Python and Spark for use cases going beyond classic Business Intelligence. Databricks was chosen to break down silos, streamline data access, and materialise an ahead-of-its-time strategy: building “Data as a Product” as early as 2022.

Use Cases Across the Entire Value Chain

Decathlon has a distinctive feature: the company is vertically integrated, covering design, R&D, production, retail, and supply chain. “In all these playing fields, we have use cases,” emphasises Didier Mama. “Both for building dashboards and for data analysis, and especially increasingly for automation through AI.“

Concrete examples include:

• Pricing and KVI: Facing inflation issues, Decathlon worked on Key Value Items, combining A/B tests and regular analyses to find the balance between passing on inflation and maintaining competitiveness.
• Supply Chain: AI significantly increased forecasting precision. “For us, gaining a few percentage points has absolutely enormous impacts,” Mama specifies.
• Value Measurement: Decathlon implemented a lightweight methodology based on Business Canvas to quantify each ad-hoc analysis’s impact. The result is eloquent: in 2023, all data analyses generated €300 million in additional revenue through better decisions.

Large-Scale Structured Data Organization

Decathlon’s data team is organised as a central platform (150-200 people) supporting four major business domains (Supply Chain, Sport, Corporate, Omni) totalling about 400 people. In total, between 500 and 600 people work on data, not counting country teams.

Usage intensity is impressive: over 20,000 people daily use platform resources, whether consulting dashboards, doing Machine Learning, or starting to use Agentic AI. “We’ve built Product as a Service to reduce Time to Market,” Mama explains. “It’s really important for us to respond as quickly as possible to business expectations.“

Governance: Rights and Responsibilities

With the intensifying use of generative AI, data quality becomes increasingly critical. “While there’s always a human behind a dashboard who can be alerted,” Mama observes, “tomorrow we’ll connect RAGs to agents that will execute tasks in a completely automated manner. There, the quality level becomes absolutely key.“

Decathlon is rethinking its governance around the concept of “Governance by Design,” with a clear ambition: “If we succeed, we won’t need governance anymore.” The approach relies on several principles:

• Use of Data Contracts and Unity Catalog
• Producer accountability: each data creator is responsible for what they produce, with traceability and auditability
• For AI agents: no data source can be connected without an identified Owner

“We’ve given you power, you can do many things: with rights come responsibilities,” Mama summarises.

From Reporting to “Data Talk”

With over 6,000 dashboards in production, Decathlon faces a new challenge:

“It’s become so easy that everyone creates their dashboard, their KPIs, which harms result consistency. Relevant KPIs should be determined by the data.“

The “Data Talk” vision consists of leveraging Large Language Models and solutions like Genie to eliminate a large portion of dashboards in favour of conversational interactions. The objective: shift from “pull” to “push,” with real-time alerts on performance deviations rather than a constant flow of information.

Three Strategic Axes for AI

Didier Mama detailed Decathlon’s AI roadmap around three axes:

1. AI for Many: generalise reduction of low-value tasks for all employees, within a well-defined and governed framework.
2. Optimisation: significantly improve complex processes without transforming them, relying on business experts (HR, finance, design).
3. Transformation: with multidisciplinary teams, fundamentally rethink professions. “We’re not just enablers, we’re contributors,” Didier Mama insists. “If I let the business lead, they ask me for faster horses. My job is to make cars,” paraphrasing Henry Ford. “Data defines an event’s behaviour, a process. We must be more than enablers, co-creators in transforming these professions.”

Sports and Data: Balancing Measurement and Intuition

To conclude the keynote, Yannick Nyanga, former international rugby player (46 caps) and author of a book on data in sports co-written with Aurélie Jean, shared his nuanced vision of data use in sports performance.

Having gone from a difficult relationship with measurement, his first physical test was a failure, to strategic adoption during his “wilderness years” (5 years without selection), Yannick Nyanga used a systematic collection of his training data to identify high-performance conditions and avoid injuries and overtraining.

Today, all clubs measure training intensity, player feelings, and match metrics via GPS and sensors. But he insists on the limitations: “Data only shows what we measure,” and can bias behaviours or create misleading correlations. He cites Ireland’s example, eliminated in the World Cup after banning certain risky but game-creating moves, or Finn Russell’s comeback from 28-3 by ignoring data-driven instructions.

His conviction: “Data is only a means serving a vision. The best teams remain those that put this data at the service of intuition and humanity.”