At the recent TCG Retail Summit, Edouard de Mézerac set out to cut through the noise around AI. His message: the era of scattered use cases is over. What comes next is something far more structural and far more demanding.
For years, companies have experimented with AI in isolated pockets described as “dots of colour” across the organization. Useful, perhaps. Transformational? Not quite. Now, the ground is shifting. Agentic AI isn’t about adding another layer of technology. It’s about rethinking how work gets done, end to end.
What makes agentic AI different?
Before diving into transformation, Edouard grounded the audience in a simple idea: agentic AI is just software… but software with agency. Traditional AI predicts and classifies. Generative AI creates. Agentic AI goes further. It can:
- Perceive information (read emails, documents, data)
- Reason about context
- Act by triggering workflows across systems
That combination changes everything. It also changes what companies need. With predictive AI, success depended on large volumes of historical data. With agentic AI, the real challenge is semantic clarity: a shared understanding of what things mean across the organisation.
If a company can’t clearly define what a “product,” “role,” or “promotion” is, an agent won’t be able to operate effectively. As Edouard put it, without that common language, “the agent simply can’t navigate.”
From theory to reality: the Carrefour case
To prove this isn’t just theory, Edouard walked through a concrete transformation at Carrefour. The starting point was deliberately modest: a 20-person market research team responsible for evaluating new store openings. Each request triggered a familiar process repeated around 400 times a year:
- A brief from the real estate team
- One to two months of analysis
- A 20–30 page report
Rather than automate parts of the workflow, the team broke it down completely, mapping around 150 individual steps before redesigning the process from scratch. The key was shifting the agent upstream. Instead of sitting with the research team, the agent was placed directly in the hands of the people initiating the request. The impact was immediate:
- Around half of projects were filtered out upfront
- Viable cases were analysed in minutes instead of months
- The process delivered 50–70% productivity gains
But the real shift wasn’t just speed. It was structural. The role of the team changed, and so did the flow of work across the organization.
Scaling beyond a single use case
Following this success, Carrefour didn’t roll out more isolated agents. Instead, they systematically scanned the organization and evaluated every process across functions like merchandising, finance, and HR, all based on three criteria:
- Does it span multiple systems?
- Does it involve repeated handoffs between teams?
- Does it rely on repetitive, pattern-based thinking?
This led to new applications, including:
- Automating finance closing anomalies
- Reviewing promotional catalogues
- Validating supplier inputs before they enter the system
In each case, the principle was the same: don’t automate the middle—reshape the workflow.
Lessons from the trenches: A playbook for agentic transformation
Based on these deployments, several patterns are emerging for leaders.
- Set a bold North Star: Small efficiency gains won’t justify the investment. The companies seeing results are aiming for step changes,like 30% cost reductions and 50% faster processes, not incremental tweaks.
- This is a CEO-level transformation: Agentic AI cuts across functions. Left to individual departments, it creates fragmentation. Driven from the top, it becomes a unifying force.
- Don’t automate broken processes: Simply layering agents onto existing workflows risks reinforcing inefficiencies. The real value comes from redesigning how work flows before, during, and after the task.
- Focus on the right use cases: The best candidates share common traits: multi-system complexity, repeated coordination, and predictable decision patterns.
- Get the foundations right: Agentic AI doesn’t remove the need for structure, it raises the bar. Clean data, clear definitions, and aligned systems are prerequisites, not afterthoughts.
A shift in mindset, not just technology
Perhaps the most striking takeaway is how little of this is about technology alone. Edouard estimates that a successful agentic rollout is only 20% technology. The rest is 30% change management and 50% end-to-end process reinvention.
There is “zero magic” in agentic AI, assures Edouard.
Leadership must understand how functions will be drastically disrupted and be prepared to do the hard work of fixing underlying processes. Technology alone will not save the business. But for an organization willing to fundamentally rethink how work gets done, the agentic revolution offers an unprecedented opportunity to unlock productivity and growth.
The companies that succeed won’t be the ones experimenting with the most tools, but the ones willing to rethink how their organisations actually operate.
The bottom line
Agentic AI isn’t just another phase of digital transformation. It’s a shift in how decisions are made, how work is structured, and where human effort is applied. For retailers like Carrefour, the ambition is clear: less time spent on administrative work, more time focused on customers.
The question is no longer whether AI can create value. It’s whether organizations are ready to redesign themselves around it.
