'AI Blueprint Deploying GenAI for Enterprise Transformation' | RAISE Summit 2024 | Paris

Panel Context and Scope

• The roundtable, moderated by Jérémy Arosh (Capgemini), examines the transformative impact of Generative AI (Gen AI) on enterprise operations, featuring representatives from Dust, Engie, Orange, Oracle, and Accenture.
• Participants define Gen AI as a democratizing force that shifts technology from a domain for specialists ("geeks") to a universal tool accessible to the general workforce.

Startup Perspective: Dust (Gabriel Hubert)

• Core Capability: Dust enables real-time reformulation of internal information, such as automatically translating code changes into marketing notifications without manual intervention.
• Adoption Metrics: At customer Alan (600 employees), over 50% of the workforce uses Dust daily, and 80% uses it weekly.
• Technology Architecture: The company utilizes Retrieval Augmented Generation (RAG), defined as retrieving specific information before generating an answer to ensure accuracy.
• Product Strategy: Interfaces must manage user expectations regarding latency; complex workflows may require asynchronous processing rather than instant conversational responses.
• Feedback Loops: The industry is shifting from binary "thumbs up/down" ratings to granular feedback systems (e.g., star ratings) to distinguish between model errors and outdated knowledge bases.
• Cultural Requirement: Successful adoption correlates with an "iterative and exploratory mindset" rather than high technical proficiency.

Enterprise Transformation: Engie (Biljana Kajtovic)

• Use Cases: The company applies Gen AI for predictive maintenance of wind turbines and optimizes employee productivity through specialized co-pilots.
• Safety Innovation: Gen AI analyzes decades of safety incident reports to identify patterns humans miss, proactively alerting workers to specific risks during industrial operations.
• Workforce Scale: Managing adoption across 100,000 employees requires addressing "foggy fear" among non-tech-savvy staff alongside the enthusiasm of digital-native innovators.
• Training Approach: A dual strategy involves top-down training on model limitations (e.g., non-deterministic outputs) and empowering exploratory users to lead peer-to-peer adoption.
• Regulatory Compliance: Engie is preparing for the EU AI Act by categorizing models by risk level and establishing clear governance for data usage and decision-making.

Infrastructure and Network Evolution: Orange (Bruno Zerbib)

• Platform Deployment: Orange launched "DINO2," a unified interface connecting employees to multiple Large Language Models (LLMs) including Mistral, OpenAI, and Meta.
• Usage Statistics: The platform trained 25,000 employees in Europe with approximately one million weekly requests.
• Productivity Gains: Developer productivity has increased by 30-40% through the use of Gen AI for code generation and content creation.
• Network Modernization: AI is used to build smarter predictive models for 5G infrastructure and to allow non-technical staff to troubleshoot networks via natural language interaction.
• Strategic Shift: The company is transitioning from a "state of innocence" to a "state of anxiety," acknowledging that all job roles will be disrupted and committing to retraining and co-creation with business units.
• Sustainability Goal: Orange aims for a 40% CO2 reduction by 2030, addressing the high energy consumption of LLMs to ensure sustainable AI growth.

Cloud and Hardware Strategy: Oracle (Philippe Battelle)

• Deployment Models: Enterprises must choose between public APIs (e.g., Mistral, Microsoft) or building custom LLMs on private infrastructure (e.g., Oracle Cloud Infrastructure/OCI).
• Regulatory Solutions: Oracle offers "DRCC," a private cloud deployment allowing on-premise or sovereign cloud operations to comply with regulations like the EU AI Act.
• Hardware Partnership: Oracle maintains a unique deep partnership with NVIDIA, hosting DGX Cloud and Blueprints on OCI to provide tens of thousands of GPUs for model training.
• Industry Example: Oracle supports the construction of "The Line" in NEOM, a 170km city, using digital twins for massive cost savings before physical construction begins.
• Future Outlook: The next generation (e.g., "Agent AI") is rendering manual prompt engineering obsolete, with young professionals naturally adapting to autonomous agents.

Consulting and Systemic Blueprint: Accenture (Laetitia Cahito)

• Internal Disruption: Accenture uses Gen AI to disrupt its own software delivery lifecycle, covering requirement gathering, testing, and coding, rather than just using co-pilots for writing code.
• Global Presence: The firm operates 12 Generative AI studios across Europe, tailored to specific industries or functions.
• Medical Innovation: A partnership with Stanford University uses multimodal Gen AI to synthesize siloed healthcare data (e.g., lung, brain, heart records) for holistic cancer pattern recognition.
• Blueprint Framework: Successful transformation requires five fundamentals: value definition, business blueprint (challenging orthodoxy), technical blueprint, responsible AI blueprint (compliance/debiasing), and governance.
• Statistical Forecast: A World Economic Forum study cited by Accenture indicates 47% of working hours will be impacted in the short term, while 95% of workers see opportunities but distrust their organizations' execution.

Cross-Cutting Themes and Risks

• Accuracy Challenges: Gen AI models often exhibit ~65% accuracy without hard constraints, necessitating strict "source of truth" integration to prevent hallucinations.
• Latency and Cost: There is a critical trade-off between model size, inference speed, and cost; B2C applications prioritize speed, while B2B applications may prioritize accuracy and asynchronous processing.
• Cultural Barriers: Organizations face a "fear of missing out" alongside anxiety about job displacement; leadership must provide clear visions to mitigate the "age of anxiety."
• Philosophical Shift: The transition is described as moving from deterministic tools (calculators) to stochastic systems ("maybe it is okay"), requiring a fundamental shift in trust and verification processes.
• Societal Impact: The panel concludes that Gen AI is a tool for societal betterment, requiring collaboration across technology, business, and ethics to address climate change and social dysfunctions.