Telecom BusinessOrangeAIOrange Business ServicesGenerative AI
Orange banks 'hundreds of millions' in AI value — without layoffs
Orange's enterprise unit hit its €50M AI value target for 2025 and raised it to €70M, with 30 use cases in production and 100,000-plus active users — no layoffs required.
Telecom BusinessWhy it matters
- OBS hit its €50 million ($55.5M) AI value-creation target for 2025 and raised it to €70 million ($77.7M)
- 30 AI use cases are in production at OBS, 20 more are being industrialized, and active users exceed 100,000
- Orange's workforce fell about 10,000 since end-2022 to 126,000, but shrank at a similar rate pre-ChatGPT
- Bank of England puts AI's annual UK productivity contribution at just 0.1–0.2 percentage points since 2024
- Alvarez says open-weight models trail frontier models by roughly six months
The story
Orange has set itself a group-level AI value-creation target running to hundreds of millions of euros, and its enterprise unit Orange Business Services already hit a €50 million ($55.5 million) goal for 2025. This year, the OBS bar has been raised to €70 million ($77.7 million). The man steering that effort, OBS chief data and AI officer Miguel Alvarez, says the gains came from speed, not headcount cuts.
Alvarez laid out the numbers on the sidelines of the Techritory event in Latvia last month. Orange is exploring several hundred AI-based use cases. Thirty have been running in production at OBS for a while, and another 20 are being "industrialized" in the current expansion.
Workforce adoption has outrun the plan. Orange originally hoped for 10,000 to 20,000 users within a couple of years of ChatGPT's arrival. Today, more than 100,000 people across the organization are active users, according to Alvarez.
Is AI actually costing jobs at Orange?
The evidence says no. Orange has cut roughly 10,000 jobs since the end of 2022, when ChatGPT arrived, falling to around 126,000 employees by September. But it shed a similar number in the preceding three years, before generative AI existed. "The teams within Orange that are getting the most value out of AI have not done it by reducing headcount," said Alvarez.
The macro data leans the same way. The Bank of England's November Monetary Policy Report found the evidence on AI's impact on UK productivity growth "modest" — an annual contribution of just 0.1 to 0.2 percentage points since 2024. Its survey shows "close to zero impact of AI on employment so far." In the US, unemployment stood at 4.2% in October, down from 4.5% in November last year.
How does Orange measure the value?
Alvarez uses what he calls "value creation," which bundles cost savings, cost avoidance, revenue protection and additional revenue. His poster case: a new product that reached a working version in a couple of months, and a market-ready version a few months after that. Without AI, he reckons the process would have taken four to five times longer.
The coding side has shifted fastest. A year ago, OBS developers used AI like a teenager's predictive text — basic autocomplete. Now some teams hand whole requirement-driven development lifecycles to AI agents. "We still have some teams using it as autocomplete, we have some teams that are optimizing some aspects of the development, and we have some teams doing requirement-driven development that are automating most steps of the software development lifecycle," said Alvarez.
Who stays in charge — the human or the machine?
That question is where Cory Doctorow enters. The AI critic coined "enshittification" for decaying online platforms, and lately he has warned about the "reverse centaur": a human steered by the AI rather than steering it. "I think it's a very good metaphor," said Alvarez. "You consider it as another tool, and it's not the human at the service of the machine. That's not the way that we see the world."
He expects agentic systems, not humans, to write most future code. People will still learn core coding concepts, he argues, for the same reason his children still learn mental arithmetic — "because it's important for the brain." Some will need to improve the systems, control outcomes and build the tests.
On guardrails, he sees a genuine trade-off. "You are giving them more leeway so they find better solutions, but as a result they could do more wrong," he said. Non-deterministic models with big capabilities demand more care about what they're allowed to do.
Why not just buy the best frontier models?
Because dependency bites. Earlier this year, Trump ordered Anthropic to block non-Americans from its most advanced models. Orange's response is a mix of commercial frontier models and open-weight alternatives deployed on its own infrastructure, including on-premises.
Three principles guide the approach. Orange chooses where the AI runs, "so you don't create a public-cloud dependency from the beginning." It demands multimodal AI, since relying only on the most expensive models "will automatically create a scaling limitation." And agents must stay portable, built on an open-source framework from LangChain, a San Francisco nonprofit. "They are not bundled into some black box or platform where it's hard to make an export," said Alvarez.
Open-weight models trail the frontier by about six months, in his view, but they've improved enough for work that needed frontier technology a year ago. OBS now runs an on-premises contract-management system that a cloud model couldn't touch for commercial reasons.
The real risk, Alvarez believes, sits with companies whose transformation can't keep pace with the technology. "Then you're going to have the cost without the benefit," he said. An AI that costs more than it saves holds very little value — a verdict that's hard to argue with, and one Orange intends to keep proving wrong with a higher target next year.
Also reported
Original: theguardian.com
More from Amara Osei
Show full bio
News editor covering technology and product news in Carrier business, 5G and 6G, fiber and broadband at Telecom Gazette.
527 articles