The report “AI in 2025: current initiatives and challenges in large enterprises”, produced by Wavestone with French Tech Grand Paris, surveys the adoption of artificial intelligence in large organisations in 2025. Here are its main findings, and what they imply for a company that wants to put AI to work on its procurement and finance processes.
Governance: between compliance and performance
The rise of generative AI has forced companies to rethink their governance. According to a Gartner study cited in the report, more than 80% of organisations have had to adapt their governance frameworks to the specifics of generative AI. The shift runs from a purely technological approach towards a vision aligned with business objectives.
Integrating AI with existing practices (data governance, cybersecurity, human resources) becomes central. The EU AI Act is driving the definition of standards, despite criticism of its complexity. For a finance department, the governance question is concrete: who validates what the AI proposes, and where is the record? That is the point of the mandate Zylio gives its agents: investigate, never decide.
From experimentation to deployment
So-called traditional AI technologies, machine learning and computer vision, are established in large enterprises. According to a McKinsey study quoted in the report, more than 60% of organisations use them regularly. Generative AI opens a new phase, with promises of productivity and personalisation.
The move from proof of concept to deployment at scale is the tipping point. It succeeds when the use case answers a precise, verifiable business question (for instance, does every invoiced line respect the negotiated price?) rather than a wish to try the technology.
Traditional and generative AI: finding the balance
Companies have to combine both approaches. According to a Deloitte analysis, the combination often gives the best results. Centralised platforms connect enterprise data, while dedicated pipelines address specific needs. MLOps and LLMOps are emerging as the methods for industrialising these systems.
That matches what we observe: a language model reads and normalises a document, but confronting an invoice line with a contract clause is done by deterministic rules, so the same case always yields the same conclusion.
Security and trust
AI intensifies existing risks and creates new ones. According to an IBM report cited in the study, AI-related incidents rose by 40% in 2024. Emerging frameworks, the EU AI Act and the NIST AI RMF, guide organisations on cybersecurity and privacy. Managing incidents involving models is becoming so complex that many organisations outsource it.
For procurement data (contracts, prices, suppliers), the requirements are clear: controlled hosting, certifications, and the commitment that a customer’s data is never used to train a model. Our security and compliance page sets this out.
Ethical and environmental responsibility
The democratisation of AI raises ethical questions, particularly with generative AI. According to the MIT Technology Review, the environmental impact of generative AI is significant, in energy and water consumption. Companies are looking for frugal practices and lean engineering, including the use of small language models (SLMs) sized to the task. The lack of vendor transparency and of standardised metrics remains a problem.
Workforce transformation and the race for talent
Traditional AI has already transformed business processes over the past decade. According to the World Economic Forum’s Future of Jobs Report, generative AI will have a broader impact on the workforce. Companies are deploying “Secure GPT” environments and organisation-wide AI solutions; HR has to anticipate, redefine objectives and support teams.
This transformation feeds a global competition for talent, with wide differences in salaries and availability across regions. According to LinkedIn, demand for AI professionals rose by 74% in 2024. The pace of innovation outstrips training programmes: upskilling and reskilling become strategies in their own right.
Strategic autonomy
Global competition in AI is dominated by the United States and China. According to the Stanford AI Index, Europe lags significantly and risks technological dependence. The report argues for strategic autonomy that aligns public and private interests, and for a clear industrial policy that supports local ecosystems in key sectors.
Outlook and recommendations
The report identifies several trends: the emergence of multimodal AI and multi-agent systems, more frugal models, stronger global AI governance, and the growing weight of ethics and responsibility. On that basis, it recommends that organisations:
- define a clear AI strategy aligned with their objectives;
- invest in training and talent development;
- strengthen governance and security;
- favour an ethical and responsible approach;
- preserve their technological autonomy.
What we take from it
2025 marks a turning point in AI adoption by large enterprises. The challenges (governance, security, ethics, talent) are real, but they share one trait: they are solved use case by use case, with a precise question, real data and a decision that stays human. That is the transition we describe in our article on moving from generative AI to agentic AI.
The OECD’s outlook on AI adds a complementary view of the macroeconomic effects of this transformation. To see what an AI use case yields on your procurement data, book a demo.
