“Layer 2”: the weapon law firms are using against AI to prove their worth

Simply providing every consultant with ChatGPT or Copilot is not enough. After a few months, the firm has mainly ended up with more subscriptions, fragmented usage and increased risks. Its methodology, however, has not become any more robust. And clients can now produce some of the analyses they previously had to pay for themselves.

This is the starting point for the first “State of Artificial Intelligence in Consulting” report published by Cigno AI, a company specialising in artificial intelligence (AI) tools for the consultancy sector. The report describes six converging forces: the commoditisation of technological capabilities, falling prices for certain models, an overall rise in AI spending, demands for sovereignty, tighter regulation and the geopolitical use of access to models.

His conclusion can be summed up in a single term: “layer 2”. This technological layer must act as an intermediary between general-purpose models – ChatGPT, Claude, Gemini or their open-source competitors – and the work of consultants. It brings together the firm’s proprietary methods, its data, its specialist staff, its access rules, source traceability and the checks applied to outputs.

“[The issue raised by] many clients of consultancy firms is: if your added value consists of copying the content from ChatGPT onto a slide using your template and charging a markup on top, you’re a bit like human packaging,” sums up Philippe Reynier, CEO of Cigno AI. It’s a blunt assessment. It cuts to the economic heart of the model: what the client is still willing to pay for when generic tools can already carry out a literature search, produce an initial summary or conduct a market comparison.

Billable work is becoming more complex

The report suggests that billable services are shifting towards the definition of ambiguous problems, sector-specific knowledge, persuasion, mediation between stakeholders and the responsibility assumed by professionals. “Market research, competitor benchmarking and some transparency-related work can now be carried out directly by the client,” explains Philippe Reynier.

Philippe Reynier, CEO, Cigno AI

This development does not necessarily mean the end of the junior consultant. It does, however, call into question the role of the generalist junior consultant, who was previously tasked with gathering information, carrying out initial analysis and preparing presentations. “It’s not an endangered species, but being a junior can no longer be limited to that,” says the CEO. According to him, knowing how to query a chatbot is no longer a distinguishing skill. New recruits will have to learn how to build agents and workflows, adapt them to a client’s specific case, and monitor their own work.

The productivity gains achieved could then be used to pursue two opposing strategies. A firm can produce the same output with less staff. It can also retain its teams and significantly increase the volume or depth of the assignments carried out. “In future, I could deliver 100 [assignments] with three people instead of ten. Or I can keep ten people and deliver 500 [assignments],” explains Philippe Reynier. This is a projection by the director, and not a productivity gain measured by the study.

Perhaps the most immediate challenge lies elsewhere. In auditing and financial consultancy, the professional remains responsible for the document they sign. Two reports cited by Cigno AI, which were withdrawn by EY and KPMG following the discovery of fabricated or disputed references, serve as a reminder that increasing the speed of production does not eliminate either errors or human responsibility. The report also draws on a ranking published in May 2026 in which the best model achieved around 73% on tasks requiring expert judgement. Whilst this result does not directly measure the quality of an audit, it highlights the continuing gap between content generation and professional judgement.

“You enter a prompt, the tool produces a result, and then all you have to do is take it at face value,” summarises Philippe Reynier. The additional layer advocated by Cigno AI is designed precisely to avoid this “black box”. Every finding must be traceable to a source, accompanied by a confidence level, and flagged when it is not based on sufficient evidence. “The ability to account for one’s sources is part of a consultant’s job.”

This traceability is also becoming a selling point. In particular, the report cites a study by Black Book Market Research conducted amongst 256 digital leaders in the UK healthcare sector: 72% would require clear audit trails when purchasing AI solutions. A Barc survey, carried out amongst 320 respondents, also indicates that 51% now consider data sovereignty to be “very important”, compared with 42% a year earlier. These findings relate to specific populations and sectors: they cannot, therefore, be directly applied to all clients of European firms.

In Luxembourg, the issue takes on particular significance when assignments involve banks, funds or other regulated entities. Locating data within Europe is not always sufficient to eliminate the risk associated with a service provider subject to extraterritorial legislation. Contracts entered into with clients may also restrict the tools that may be used or require prior approval.

Using an open-weight model hosted on infrastructure controlled by the firm can reduce some of these risks. It does not eliminate them entirely: it is still necessary to secure the hosting environment, manage access rights, isolate each client’s data and maintain an audit trail. “It is not the role of an independent consultant, nor even that of a small firm, to build all this infrastructure,” emphasises Philippe Reynier.

Cigno AI also highlights technological volatility. Based on data from Artificial Analysis as at 11 August 2026, the report notes that the cost of executing the same task can vary by a factor of up to 100 between models. It also compares the performance of the open-source Kimi K3 model with that of several proprietary models. However, rankings change rapidly and are based on different benchmarks, depending on the task. They describe the state of the market at a given date, not a lasting hierarchy.

The danger of having a single supplier

This instability has led Cigno AI to advise firms against relying on a single supplier. An architecture capable of routing each task to multiple models can reduce costs, maintain service continuity when a product is withdrawn, and reserve the most expensive tools for tasks that warrant their use. It also limits the risk of vendor lock-in.

The report identifies four categories of players: large traditional consultancies, AI-native consultancy firms, networks of independent consultants, and deployment organisations built around model providers. The former possesses the methodologies, data and client relationships, but remains tied to a hierarchical organisational structure and time-based billing. AI-native players have greater technological depth, but less extensive sector coverage. Freelancers are increasing their output capacity without necessarily having a common governance structure.

Large networks have already invested heavily in AI. But announcing thousands of licences does not mean that the operational model has been transformed. “Initially, the approach is to provide teams with Copilot, ChatGPT or Claude. Nine or 12 months later, costs have risen and the firm doesn’t necessarily see any increase in efficiency,” observes Philippe Reynier. The most advanced organisations are therefore trying to embed their methods into in-house tools rather than letting each employee work freely with a chatbot.

Cigno AI is collaborating with the European Institute of Business Administration (Insead) and the University of Oxford on various projects relating to AI and consultancy, although neither Insead nor Oxford are regarded as co-authors. Cigno AI markets precisely the type of orchestration, governance and knowledge layer that the report describes as essential.

The question posed to firms goes beyond the product promoted by Cigno AI. AI reduces the cost of generic output, but increases the cost of verification, oversight and accountability. Audit and consultancy firms will not retain their value simply because they use the same models as their clients. They will have to demonstrate the value they add – and be able to account for every result produced on their behalf.