The figure is striking because it goes beyond mere announcement effects. Ipsos plans to invest €1.2 billion over five years in artificial intelligence and acquisitions, with an explicit horizon: to “reinvent” itself by 2030.

Behind the sum lies a more uncomfortable reality for a research leader: growth has slowed, and shareholders expect a clear trajectory. The new CEO, Jean-Laurent Poitou, appointed in 2025, arrives precisely with a promise: to accelerate, industrialize, and put Ipsos back in the race for added value.

The question is therefore not “Is Ipsos doing AI?”. It’s harder: how can a company whose capital is trust adopt automation techniques without degrading the methodological quality that underpins that trust?

A transformation managed as an industrial program

The article in L’Usine Digitale describes a plan structured around two levers: AI and acquisitions, in a growth recovery logic. Ipsos already claims over a thousand data scientists and AI engineers, and wants to increase this proportion in its workforce.

The financial trajectory is explicitly assumed. Ipsos targets average annual organic growth of 3% to 4% between 2026 and 2028, then over 5% in 2029-2030.
This ramp-up also responds to a positioning objective: to become “the AI augmented Global Market Research leader”.

This vocabulary (“AI augmented”) deserves attention. It doesn’t promise human replacement. It promises enhancement. But who will define the boundary between “augmenting” and “substituting” when productivity imperatives tighten?

AI, yes—but especially for “commercial research”

Reuters provides a structural detail: Ipsos intends to focus its AI efforts on commercial research, which represents most of its revenue, rather than on political polling, which is more marginal.

This is a strategic choice that also carries risk.
In commercial research, the promise is simple: shorter cycles, faster deliverables, the ability to “produce” insight continuously. In politics, the slightest suspicion of manipulation or contamination by generative systems can destroy an institute’s credibility.

In other words: Ipsos is protecting its economic core while avoiding, at least publicly, the most explosive terrain of democratic trust.

Speed as the new market standard

In its “Horizons” communication, Ipsos announces it aims to produce insights in real-time for certain projects and “within 48 hours” for the majority.

This objective is consistent with client pressure. It also reflects an industry shift: insight is no longer a report, it’s a flow. A marketing decision, a price adjustment, a response to a reputational crisis demand immediate iteration.

But a question remains, rarely asked in press releases: what happens to statistical robustness when the market imposes the logic of “fast insight”? Reducing timelines mechanically shifts trade-offs: fewer back-and-forths, fewer checks, more dependence on automated pipelines.

Proprietary data, client data: the dividing line

Reuters specifies a crucial point: Ipsos says it mainly trains its models on data it owns, particularly from syndicated studies resold to multiple clients. For proprietary analyses commissioned by clients, data usage depends on contractual clauses regarding ownership and exploitation rights. This passage is central because it shows what AI does to research: it transforms responses and observations into learning assets. Two questions arise. Who, tomorrow, owns the value derived from research fieldwork: the client, the institute, or the model? And how can respondents be clearly informed of these uses without producing unreadable consent? Professional associations’ recommendations on survey methodology precisely emphasize transparency, data security, and adapting consent procedures when AI tools are used.

“Real” panels and synthetic data: the unstable balance

Ipsos emphasizes a competitive advantage: its proprietary panels and access to “real respondents”. The company presents this access as a reliability condition, including for relevant use of synthetic data.

This statement has an implicit scope: the industry wants the power of simulation, but knows it cannot kill its source of truth. Academic research on generating and evaluating synthetic data for surveys is progressing rapidly, with complete pipelines to produce and test these datasets.
In parallel, work explores the use of large language models as “virtual respondents”, analyzing their biases and sociodemographic coherence limitations. The risk is not theoretical. The “model collapse” phenomenon, documented in Nature, shows that training models on data generated by other models can degrade quality if the loop isn’t controlled. Operational translation for a research company: synthetic data can accelerate, test, simulate. But it cannot become the main raw material without strict control mechanisms, mixing with real data, and continuous auditing.

“Ipsos Facto”, platforms and productization: research becomes software

Ipsos investor documents describe a “self-serve” platform (Ipsos.Digital) generating approximately €140M in revenue in 2025. They also mention internal adoption of a tool (Ipsos Facto) deployed company-wide, with approximately 11,000 monthly active users (MAUs). These figures tell another transformation story: market research is no longer just a custom service. It becomes a suite of products, interfaces, industrializable modules.

This is where AI changes the economic nature of the business. It allows packaging building blocks (collection, cleaning, analysis, delivery, storytelling) and making them “scalable”. It brings Ipsos closer to a SaaS model, without saying so. But an insight SaaS is judged by two simultaneous metrics: speed and reliability. Yet these two metrics often conflict.

Mergers and acquisitions: buying what you don’t have time to build

Ipsos is not at its first move. The company finalized the acquisition of The BVA Family in June 2025, particularly strengthening expertise in customer experience, point-of-sale behaviors, and behavioral sciences.
L’Usine Digitale also recalls the (unsuccessful) attempt around Kantar Media in 2024, a sign of an expansion-through-acquisition strategy.

In a rapid technological cycle, acquisition becomes a form of shortcut: buying methodologies, teams, sometimes data assets, rather than internalizing everything.

But again, a question remains: how to integrate different methodological cultures without homogenizing downward? And how to prevent AI from becoming a tool for excessive standardization of approaches, at the expense of the customization that makes major institutes valuable?

Governance, transparency, compliance: the AI Act as backdrop

The rise of AI in European organizations now unfolds under the gaze of a regulatory framework being deployed. The European Commission continues implementing the AI Act according to the planned schedule. Obligations targeting “general purpose” AI models entered into force in August 2025, then stronger requirements for certain systems apply from 2026 onward.

Even though Ipsos is not a provider of general-purpose models “like OpenAI”, the company will have to deal with transparency and traceability requirements. These requirements will become decisive once it produces, uses, or integrates systems generating synthetic content in decision-making deliverables. European work on labeling and watermarking AI-generated content illustrates this direction.

The issue, for a research institute, is less legal than reputational: compliance won’t be enough. Trust requires concrete explainability, understandable by both a marketing director and an audit committee.

The real issue: preserving “proof” in an automation world

Ipsos promises a transformation in the way of working, with automation of certain procedures and scaling up of AI skills.

The promise is rational. The insight market is attacked on two fronts:

  • Tech players selling fast and cheap dashboards.
  • C-suites demanding direct ROI on every euro spent.

But an institute like Ipsos’s singularity lies in a simple notion: proof.
Proof is not just a result. It’s a chain: protocol, sampling, field quality, processing, interpretation, limitations, uncertainties.

AI can strengthen this chain. It can also weaken it if it becomes a black box of productivity.

The final question is therefore this: will Ipsos succeed in making AI an instrument of rigor—and not an instrument of speed?