Hardly any industry feels the AI shift as directly as consulting. Large firms are building their own AI platforms, clients are asking for efficiency gains, and in almost every project team someone uses ChatGPT or Copilot. At the same time, partners keep asking us the same question: what of this is actually fit for consulting, and where does it become risky? This article offers a sober assessment.
Where AI really helps today
The effect is most noticeable in research and market overviews. AI produces a first overview of a market, its drivers and the main competitors in minutes instead of days. This helps most in the early project phase, when hypotheses are formed and the work plan is sharpened. The results are a starting point, not an end result: every number needs a verifiable source.
The benefit is similarly large when screening large volumes of documents. In a due diligence or a restructuring engagement there are often hundreds of documents. AI searches contracts, reports and minutes, flags relevant passages and summarises anomalies, such as unusual termination clauses or dependencies on individual customers. It is a fast first pass; critical passages are still reviewed by people.
AI is also excellent for structures and first drafts: issue trees, outlines, executive summaries, proposal texts. The effect is especially noticeable for proposals, which cost a lot of unpaid time. The same applies to analysing interview transcripts and workshop notes, from which patterns and open points can be worked out quickly. And finally, AI helps to rediscover existing knowledge that has been filed away in old project folders for years, provided it is accessible and well organised.
Where AI reaches its limits
AI can show options and collect arguments. Which recommendation is right for this client in this situation, however, takes experience, context knowledge and responsibility. That remains the core of the consulting service. Interpersonal work also stays human: building trust, understanding resistance in management, guiding change.
It becomes risky with numbers without a source. AI models can invent answers that sound plausible but are wrong. An invented market size in a steering committee presentation is not a small mistake but a liability and reputation risk. For confidential engagements such as M&A, due diligence or restructuring, the question of data protection comes on top: where is the data processed, is it used for training, who has access? Many freely available tools do not answer this the way clients expect.
The probably most underestimated problem is fluctuating quality. If every consultant uses AI in their own way, the result depends on how well the individual can prompt. This leads to different standards within the same firm.
| Well suited | Use with caution | Not suited |
|---|---|---|
| First market overview, forming hypotheses | Figures and market sizes (only with a source) | The final recommendation to the client |
| Screening large volumes of documents | Confidential client data (only in secure environments) | Unchecked content in the deliverable |
| Outlines, drafts, proposals | Valuations and forecasts | Client relationship and change management |
The bottleneck is the process, not the model
The large AI models are becoming more and more alike and are available to everyone. Whether a firm really becomes faster and better with AI is therefore decided elsewhere: recurring project types follow a fixed, proven approach instead of prompting by gut feeling. Every statement can be traced back to its source, and at defined points a consultant reviews before work continues. The data sits in a secure environment with EU data hosting, without training and with clear separation between engagements. And insights flow into a shared knowledge base instead of disappearing in personal chat histories.
How to get started with AI in your firm
- Pick three recurring tasks that cost a lot of time today, e.g. proposals, market overviews, document screening.
- Define one common approach instead of letting every consultant experiment on their own.
- Set rules for client data: which tools are allowed for which data?
- Introduce a source requirement: no number in the deliverable without a traceable origin.
- Measure results: time saved per project and margin development on fixed-fee projects.
Conclusion
AI does not make consultants redundant. It shifts what consultants spend their time on: away from research and formatting, towards analysis, judgment and client work. For boutique firms in particular this is an opportunity to deliver, with small teams, results that used to require considerably more people. What matters is to treat AI not as a toy for individual employees, but as a permanent, secure part of the firm’s way of working.
