Sipa Ouest-France’s AI lessons: Keep the rules simple and trust your people
The French publisher transformed Ouest-France, the country’s biggest daily newspaper by circulation, into one of France’s leading online news platforms before expanding into video and television.
Founded 80 years ago, the group today comprises more than 100 media brands, employs over 5,000 people, including 1,200 journalists, and reaches 28 million readers every month. It is now turning its attention to using AI to improve efficiency and personalise content.
But Bakhouche, speaking at our World News Media Congress in Marseille, was candid: the organisation “has made a lot of mistakes”, changed course, and is still figuring it out.
Successful AI adoption depends on clear principles, leadership and trust, rather than complex policies, according to Bakhouche.
“We never stopped adapting ourselves to changes in technology,” he said.
Four principles instead of endless AI rules
The first lesson from Sipa Ouest-France’s AI journey is about governance.
Complexity is better managed through a few robust principles than through exhaustive procedures that no one truly owns, Bakhouche said.
Trust remains the group’s most valuable asset, and poorly governed AI has the potential to undermine years of work building credibility with readers, according to Bakhouche.
The publisher now uses a simple framework built around four core principles, instead of detailed policies for every use case:
- Human oversight.
- Transparency with readers.
- Comprehensive assessment before deployment.
- Editorial responsibility.
Why restricting AI use failed
One of Sipa’s biggest lessons came after an early attempt to tightly control AI adoption.
Initially, only a limited number of employees were allowed to use the company’s in-house LLM, Muse. This was done to protect their editorial content.
“But it didn’t work,” Bakhouche said.
Employees were already using consumer AI tools such as ChatGPT and Gemini in their personal devices, making it difficult to prohibit their use at work.
The result was widespread “shadow AI” (employees using AI tools outside the company’s approved systems and governance).
An internal survey found that around 80 percent of employees were already using AI, but largely through external tools on their personal devices rather than the company’s own platform, he said.
That prompted a complete change in strategy.
Instead of restricting experimentation, the publisher encouraged employees to explore AI while introducing only a handful of clear “red lights.”
The most important rule was: editorial content must only be entered into Muse.
“We don’t want to feed AI platforms with our content. It’s clear, it’s simple and it’s easy to understand,” Backhouche said.
The strategy produced results. Across the organisation, teams began developing AI-powered projects voluntarily.
One example came from the group’s advertising business, Additi, where a four-person team built a classified platform for warehouses in just one month. The platform is already attracting customers, generating revenue, and still growing, he said.
AI adoption depends on training, not age
Encouraging experimentation also introduced another challenge: not everyone felt equally prepared to use AI.
Within the group, levels of AI maturity vary widely, with early adopters spread across different age groups while others feel overwhelmed by the pace of change.
For those employees, awareness alone is insufficient, according to Bakhouche.
“We need to train them on a regular basis to increase not only their awareness but their ability to use these tools. The risk, if we don’t do that, is that it discourages them and slows down implementation,” he said.
Another decision that proved valuable was collaboration with the unions.
“We decided not to hide and say, ‘we will develop things without talking to you, and when we decide to industrialise, we will come and explain.’ We embarked them at the very beginning of the journey, and to be honest, they were very collaborative,” he said.
While collaboration does not necessarily guarantee agreement on every project, involving employee representatives early creates a stronger foundation for future change, Bakhouche added.
For publishers introducing AI at scale, that early engagement may prove as important as the technology itself.
WAN-IFRA Members can access Fabrice Bakhouche’s presentation on our Knowledge Hub.
The biggest organisational impact is still ahead
Despite numerous successful experiments, the most significant organisational changes have yet to arrive.
The group has already tested numerous AI use cases and launched innovative commercial projects while seeing gains in efficiency.
However, the company still lacks a clear understanding of how widespread deployment will reshape workflows, newsroom processes and organisational structures once AI becomes embedded in day-to-day operations, according to Bakhouche.
“The role of the CEO and the top managers will be key in showing the way, giving meaning to those transformations and convincing our unions and our employees to embark on the journey,” he said.
Success will require both vision and empathy, he added. Leaders must communicate not only with enthusiastic early adopters but also with employees who remain uncertain about the changes ahead.
Why publishers must work together
Even for a large publisher such as Sipa Ouest-France, understanding AI and investing in new technologies remains expensive and resource-intensive.
Productive collaboration is key, and “it’s very useful and crucial to join forces,” Bakhouche said.
He pointed to two collaborative projects:
- Participation with 11 other publishers in WAN-IFRA’s Advanced Newsroom AI Catalyst.
- A joint project with three other regional French publishers – alongside AI company Mistral – to build chatbots for journalists and readers, and to develop the data infrastructure needed to connect publisher content to LLMs effectively.
He also said publishers need a collective approach in negotiations with AI companies, citing France’s neighbouring rights talks with Google as an example of effective industry cooperation.
“We managed to find agreements that were satisfying for both parties,” he said.
Publishers create significant value through trusted local and hyperlocal journalism whether audiences access it directly, through the open web, or via AI assistants.
“I see no reason why we should do that for free with zero visibility into who our readers are or what sort of questions they are asking chatbots,” he said.
“It’s not about begging for glass beads or for a couple of euros,” Bakhouche said, adding, “It’s just asking for a fair and balanced deal.”
