Few industrial case studies cover eight years of transformation in such detail. At the Tech For Industry Show, Maxence Tilliette, Managing Director, Global Industry Lead Industrials — Supply Chain & Engineering (Accenture), Ludovic Doudard, General Manager Process & Digital Supply Chain (Renault Group) and Laurent Sidler, IT Leader Mirror Your Industry (Renault Group) looked back at the journey that took the automaker from data-driven operations to digital twins powered by agentic artificial intelligence.

Ludovic Doudard opens the session by outlining the automotive industry's structural challenges: rapidly accelerating global competition (the volume of vehicles sold in Europe by Chinese manufacturers is said to have doubled in six months), energy cost volatility that's especially pronounced for the sector, supply chain instability (Renault monitors 40,000 rank-N suppliers, with at least one major alert a week requiring a crisis unit), and constant reconfiguration of production tools — illustrated by the Douai plant, which produces more than 1,000 vehicles a day on a single line assembling several different models (Scénic, Mégane, R5, the small Alpine A390, Micra).

These constraints are part of Renault's Future Ready plan, built around cost reduction, increasing plant capacity, and a target of cutting customer-facing quality incidents by 50% — an effort that has helped Renault become the world's number 2 on warranty costs and customer quality perception, just behind Toyota and ahead of the German manufacturers.

Laurent Sidler details the first phase of the transformation, launched around 2017-2018 and widely deployed since 2023.

Renault chose to build its own data collection solutions rather than relying solely on the cloud, seen as "too late, too costly, even impossible" for structuring data. The company thus developed a library of reusable process data models: the screwdriving model, for example, applies identically to the 2,000 to 3,000 screwdrivers of different brands deployed across the group's plants, capturing the operation as a whole (equipment, operator, workstation, applied torque, OK/not-OK status) rather than just the equipment's raw data. Renault now has around 80 of these models, with around fifteen more being created.

In practice, these models are deployed directly at the equipment level, via a "drag and drop" tool developed with a partner, with configuration taking anywhere from a few minutes to 1-2 hours. The data then flows up to the plant level and, entirely, to the cloud. Edge computing is still used for genuinely real-time cases (energy meters, AGV control), but most industrial use cases are satisfied with a "relevant timeframe" of a few minutes to a few hours, processed at the cloud level.

Across the group's 25 plants, 25,000 pieces of equipment are now connected, generating around 5 billion data points daily (images, time series, tabular data). This foundation feeds, among other things, control towers where screens can be reconfigured on the fly depending on the subject of the ongoing meeting — quality one moment, supply chain the next — avoiding debates over the reliability of the numbers to focus directly on problem-solving.

From the start, Renault wanted to give direct access to data and AI tools to operational teams — maintenance, quality, line supervisors — rather than to data scientists alone. Around 7,000 users now benefit from this self-service access. One striking example: AI-powered automated visual inspection, with a target of 500 checks per plant by the end of 2025 and 1,000 by the end of 2027, built on an internal platform (relying mainly on standard models) rather than on solutions purchased site by site — too costly and not scalable. Across 20 plants each producing around 1,000 vehicles a day, this represents several million automated quality checks.

Renault estimates it has generated more than €120 million in value through this data-driven approach, spread across energy, maintenance, quality, supply chain, plant yield, and above all time to market — the latter being presented as the main remaining source of value, since assembly yield is already close to 98%.

Two concrete examples illustrate this shift: cutting the ramp-up time for a new model from 12 to 6 months (knowing that a new body-in-white model requires re-coordinating around 800 robots); and reducing quality defect detection time from 2 months to less than 2 days, thanks to a partnership with towing companies that makes it possible to know within 24 hours that a Renault vehicle has broken down, combined with a per-vehicle digital passport that precisely identifies at-risk vehicles rather than recalling a million units.

Once the data-driven foundation was in place, Renault ran into a limit: most manufacturing relies on "legacy" systems more than 30 years old, complex and slow to evolve — a direct obstacle to new needs such as laser cladding (a new manufacturing process required by European regulation) or managing the software of increasingly connected vehicles.

For three years, Renault has been working with Accenture to rebuild its systems around real-time digital twins, driven by an industrial ontology — a structured data model linking equipment, operators and operations. These new systems are designed to be resilient (continuous operation), built with the help of AI, and above all open to agentic AI to directly query and modify production systems.

To illustrate the approach, the speakers detail the rollout of a new line dedicated to depositing a coating layer on brake discs (a regulatory requirement tied to durability). Rather than starting from a poorly suited traditional MES template, the platform built with Accenture mobilizes several specialized AI agents:

The result: a process of this kind, which would have taken several months with legacy systems, was developed in five days.

During a simulated demo, the speakers show a central agent called "Ask the Line," embedded at the heart of the system rather than on its periphery, queried in natural language: the average coating thickness measured at the quality station, or identifying the line's current bottleneck (the grinding station, in this example). A specialized agent then proposes solutions (adding an operator, reducing the cycle time of the previous station), which a human operator validates — with that validation then dynamically reconfiguring the system, without going through manual configuration screens.

Asked about how this transformation was organized, Laurent Sidler and Ludovic Doudard stress three principles: a group-wide strategy rather than an isolated initiative, requiring alignment across all functions (not just production); dual IT-and-business governance, to reconcile technical requirements with business vision; and a resolute focus on business value, prioritizing rollouts where business demand is strongest, to ensure the program's long-term viability.

Asked about a possible next step — physical automation via humanoid robots — the speakers confirm a public partnership with Wandercraft, with humanoid robots already operational at the Douai site. The stated approach is pragmatic and industrial rather than a showcase: a humanoid robot in a plant must succeed in at least 999 out of 1,000 cycles and be able to carry heavy loads to be truly industrializable — against the backdrop of French industrial sovereignty in this field.

Asked how long it would take to retrace this path today, the speakers estimate that eight years would no longer be necessary, thanks to the accelerators developed along the way — easier equipment connectivity, self-service tools, operational control tools. They also highlight an often underestimated point: data-driven transformation also upends the role of middle management, caught between frontline teams with real-time data access and leadership that also has a complete view — hence the importance of continuing to value frontline know-how, which future expert agents will precisely seek to capture for predictive maintenance.

Watch the full session, along with the other Tech For Industry Show replays, on our dedicated page.