Is This the End-game for Boutique Consultancies: Automated Expert Solutions?

As the AI revolution kicks-off, one of the most promising frontiers lies in multi-agentic systems – coordinated AI bots that replicate and enhance human workflows. While much attention has focused on large language models (LLMs) like GPT, boutique consultancies are well-positioned to lead the development of more capable systems.

This seems perhaps like an odd thing to say. After all, ChatGPT4 outperforms most specially trained LLMs. When pitted against Bloomberg’s financially trained AI model, or several medically trained models, ChatGPT4 outperformed them.

The difference with what I am recommending for (some) consultancies is three-fold: first, they are not should not be building their own LLMs, but fine tuning existing smaller models; second, their IP should lie with the interactions and training of combinations of LLM – what we call agents; finally, unlike LLMs, consultancies power does not lie in vast datasets or abstract simulations, but in their ability to engage in real-life experimentation through live client projects.

This capability to train, test, and refine multi-agentic systems in dynamic, high-stakes environments offers a competitive advantage that is unparalleled and fundamentally difficult to replicate.

Why Large Language Models Cannot Compete with Real World Simulations

Large language models are impressive generalists, trained on enormous datasets to perform tasks across a wide range of domains. However, their inherent limitation is that this training occurs in static, controlled environments, disconnected from the real-world complexity of consulting workflows.

LLMs excel at answering prompts or analysing text in a vacuum, but they struggle with the nuance, unpredictability, and ever-changing variables that characterise human decision-making in live projects.

In consulting, workflows are rarely linear. They involve constant adaptation to shifting client priorities, competing stakeholder interests, and unexpected risks. These are not anomalies but the very fabric of the work. Static models, trained on general data-sets, are not great at generating advice for niche instances.

ChatGPT4o will generate a great sounding process for a general M&A approach, but when dealing with a precise case (e.g. an Indian textile manufacturer taking over a dye distributor in Munich) much of its output will fail the expert eye test.

Fortunately, boutique consultancies have exactly the type of experience and data to fine-tune these models to make them much better at useful advice.

Live Experimentation: A Unique Opportunity for Boutique Consultancies

Boutique consultancies operate at the intersection of human expertise and complex, context-specific problem-solving. This makes them uniquely positioned to use live client engagements as laboratories for training and improving multi-agentic systems.

Every project offers an opportunity to embed AI agents into workflows, allowing them to learn, adapt, and improve in real time.

This live experimentation provides three critical benefits:

1. Contextual Learning: AI agents embedded in consulting projects are exposed to the realities of messy, ambiguous data, unpredictable dynamics, and shifting goals. They learn not just from outcomes but from the decision-making processes consultants undertake to navigate these challenges. This contextual learning is far more valuable than training in static datasets.
2. Iterative Refinement: Through continuous interaction with consultants and clients, AI systems can be refined with each engagement. As they encounter edge cases and exceptions, they adapt and improve. Over time, these systems become not only more autonomous but also increasingly aligned with the nuanced needs of consulting workflows.
3. Client-Centric Development: AI agents trained in live settings are shaped by direct interaction with client-specific challenges. This ensures that they are tailored to meet the actual needs of clients rather than being designed in isolation. The result is a level of relevance and precision that generic AI solutions cannot achieve.

Building Proprietary, Adaptive Systems

Boutique consultancies have the opportunity to develop proprietary AI systems that embody their niche expertise.

For example:

• A firm specialising in organisational change could create AI agents that analyse cultural dynamics and recommend tailored interventions.
• A consultancy focused on financial performance could deploy agents that autonomously assess key financial metrics and flag areas of concern or opportunity.

These systems are not generic tools—they are strategic assets, deeply integrated with the firm’s specific capabilities. Crucially, as they are deployed in live client projects, they continuously improve, becoming more sophisticated and irreplaceable over time. This creates a strong competitive differentiator that sets boutique consultancies apart.

Start With Assistance and Move to Autonomy

We are not yet ready for this model of working. However, agents can start at the assistant level now – helping capture and analyse data, making recommendations and so on.

However, crucially, these need a process by which they can learn from errors. There’s absolutely no asset improvement if you copy AI generated analyses into a Word document and make edits yourself.

Instead, imagine an AI bot that works with, and learns from, the everyday activities of a specific role in your firm. For example, your M&A analyst bot might be fine-tuned on your M&A analysis instructions and on previous examples of M&A analysis output.

It might then be instructed to learn from interactions from the human AI analyst. When undertaking the analysis, the AI initially supports the human, ingesting data and performing analysis. This saves the human time. But as the human corrects and edits the analysis, the M&A analyst bot improves through learning.

Eventually, as AI capabilities improve, the AI is likely to be better than the human at analysis, so we move from a position of assistance to one of autonomy. Now, replicate this across your M&A advisory project: competitor analysis, valuation, project management, risk analysis and mitigation, and so on. These roles, or activities, eventually become AI agents which can run the process for you.

To reiterate, this capability is ‘almost’ there. AI is now performing at human levels in most knowledge based tasks, but processes are tougher and will require more agentic capability than we have in 2024. But in 2025, this seem eminently possible.

Self-Funded Research and Development

Unlike tech firms that require substantial upfront investment to develop AI systems, boutique consultancies can fund their AI innovation through their existing client work.

Every project will become a live training environment, where AI agents gain experience and refine their capabilities. This approach ensures that AI development will be both cost-effective and closely aligned with real-world needs.

For instance, an agent designed to assist with strategy development might start by automating background research and data analysis. Over time, as it observes consultants making decisions, it learns to prioritise insights and even suggest initial recommendations.

This iterative development is funded by client fees, making it a sustainable and scalable model for innovation.

The Long-Term Strategic Advantage

Live experimentation with AI agents creates a powerful feedback loop: as agents improve through real-world interaction, they will enhance the firm’s ability to deliver value to clients.

This, in turn, generates more data and insights, further refining the systems. Over time, this cycle builds AI solutions that are not only highly effective but also deeply aligned with the firm’s unique expertise.

Moreover, this model is both cheaper and more scalable than the human-centric approach to consulting. However, it does have flaws. First, as I’ve said, it’s only in draft mode. Second, it won’t be able to do sales or innovation because these come from human interactions with clients.

Sure, conversations can be captured and analysed by AI, but it’s the interaction with the trusted advisor when clients truly open up. Finally, there are significant cultural and behavioural challenges here – not least the reluctance of consulting leaders to accept that AI will change their firms fundamentally in the next five years.

Either through the firm implementing AI and agentic models, or by losing market share to those that have done this already.