Artificial intelligence is rapidly becoming one of the largest investment themes in professional services.
Firms are announcing AI platforms, building internal copilots, launching agent initiatives, expanding data capabilities, hiring AI specialists, and signing increasingly large technology agreements. Deloitte plans to invest billions into AI capabilities. PwC committed US$1 billion to its AI transformation. EY, KPMG, BDO, Grant Thornton, Baker Tilly, and Forvis Mazars are making similar investments in different forms. Across the industry, the direction appears remarkably consistent. AI is no longer treated as an experiment. It is increasingly treated as infrastructure.
The assumption underneath much of this activity feels relatively straightforward. AI will make professional-services firms more productive. Work will be completed faster. Fewer hours will be required. Margins will improve. Some organizations may ultimately require fewer people. The logic appears intuitive because professional services has historically been a labor-intensive industry. If AI reduces labor effort, costs should fall. Much of the current discussion around AI business cases, utilization improvements, workforce planning, and productivity gains is built on exactly that assumption.
The difficulty is that professional-services firms may increasingly be asking the wrong question. The industry talks about AI primarily as a labor-reduction technology. In practice, AI often behaves more like infrastructure. And infrastructure has never been free. Railways did not eliminate transportation costs. They changed where transportation costs occurred. Cloud computing did not eliminate technology costs. It shifted costs away from physical infrastructure and toward platform consumption. Global delivery centers did not eliminate delivery costs. They redistributed costs into governance, coordination, workflows, and integration. AI increasingly appears to be following the same pattern. The visible effort required to perform individual tasks may decline substantially while the infrastructure required to support those tasks becomes larger, more sophisticated, and more expensive. The question is therefore not simply how much work AI removes. The question is where the economics go once that work disappears.
The Industry Keeps Looking at the Top of the Stack
One of the reasons AI economics are becoming increasingly difficult to understand is that visibility and cost often sit in different places. Most executives experience AI through the layer closest to the client. They see consultants producing presentations faster. Auditors completing testing procedures more efficiently. Tax specialists generating analyses in minutes rather than hours. Lawyers reviewing contracts with AI support. These interactions are tangible, measurable, and easy to understand because they happen directly inside client-facing work. Naturally, this becomes where most organizations focus their attention.
The problem is that the visible layer often represents only a small portion of the underlying economic system. Every AI-generated insight depends on models, workflows, integrations, governance structures, security controls, data environments, and infrastructure investments sitting underneath the user experience itself. The consultant entering a prompt may see only a few seconds of interaction. Behind that interaction sits a much larger environment involving cloud platforms, model providers, orchestration layers, compliance controls, cybersecurity frameworks, and institutional investment. Most of that environment remains largely invisible because it sits outside the immediate delivery process.
This creates a pattern that should feel familiar to anyone studying the economics of professional-services firms. The visible benefits appear near the top of the organization. The hidden costs accumulate further down. Productivity gains are highly visible. Infrastructure requirements are not. Faster delivery becomes measurable immediately. Governance obligations emerge gradually. The result is that many firms may be evaluating AI through the narrow lens of labor productivity while substantial new costs are quietly accumulating somewhere else entirely. The AI Cost Stack provides a way of understanding those hidden layers because AI costs do not occur in one place. They increasingly occur across an interconnected system.
The AI Cost Stack
The AI Cost Stack provides a way of understanding where AI economics increasingly sit inside professional-services firms. While most discussions focus on the visible productivity gains created by human and machine cognition, a growing share of the economic reality sits deeper within the institution: orchestration, governance, and capital. As AI adoption expands, costs increasingly flow down the stack even while value remains concentrated at the top.
Human Cognition: The Layer Closest to the Client
For most of the history of professional services, expertise was the business model. Firms hired talented people, organized them into pyramids, sold access to their knowledge and judgment, and measured success through utilization, leverage, and partner profit. Whether the service involved audit, consulting, tax, legal advice, or engineering, the underlying economic engine remained remarkably consistent. Human cognition created value. Human cognition delivered value. Human cognition generated revenue.
This reality shaped almost every institutional mechanism inside the industry. The partnership model emerged because expertise lived inside individuals. Promotion systems existed to develop expertise. Utilization mattered because people represented the primary productive asset of the firm. Even growth followed this logic. More professionals created more capacity. More capacity created more revenue. The economic center of gravity sat largely where the work happened.
Artificial intelligence does not change that reality as quickly as many observers assume. In some ways it may increase its importance. Clients rarely pay premium fees for document summarization or data extraction. They pay for judgment. They pay for confidence in uncertain situations. They pay for trusted advisors capable of helping them navigate complex decisions with significant financial, operational, or regulatory consequences. As routine cognitive activities become increasingly automated, human expertise may move upward rather than disappear. The advisor remains the face of the service. The difference is that the advisor increasingly depends on capabilities sitting deeper within the stack. Human cognition remains the layer closest to the client. Increasingly, however, it is no longer the only cognitive system participating in production.
Machine Cognition: Where Expertise Starts Behaving Like Infrastructure
Beneath human expertise sits the layer most people associate with artificial intelligence itself: machine cognition. This is where models transform data into outputs. Research, summarization, classification, pattern recognition, forecasting, content generation, knowledge retrieval, and increasingly decision support all sit within this layer. The promise appears compelling because machine cognition can perform certain cognitive activities at a scale, speed, and consistency that humans cannot match. Tasks that previously required hours of manual effort can often be completed in seconds. The productivity gains are real, which is precisely why so many firms are racing to deploy AI capabilities across audit, tax, consulting, legal, and advisory environments.
The mistake many organizations make is assuming that AI primarily makes expertise cheaper. A more accurate description may be that AI changes the nature of expertise itself. Historically, expertise lived inside people. Knowledge accumulated through experience, training, mentorship, and professional judgment. Artificial intelligence begins moving parts of that expertise into systems. Research can be reused. Analysis can be replicated. Recommendations can be generated repeatedly at near-zero marginal cost. What previously required an individual increasingly becomes available through a platform. Expertise starts behaving less like labor and more like infrastructure.
That distinction matters because infrastructure follows very different economics than labor. Labor scales largely through people. Infrastructure scales through investment. Labor sits inside engagements. Infrastructure sits beneath them. Once expertise becomes embedded in systems, competitive advantage increasingly shifts toward ownership of the underlying capability rather than execution of the individual task. The conversation therefore starts moving away from prompts and productivity toward platforms and control. The question gradually becomes less about who performs the work and more about who owns the cognitive infrastructure performing it. That shift may ultimately prove far more important than any immediate reduction in delivery effort.
Orchestration: Where Most of the Hidden Economics Sit
Most AI demonstrations create a misleading impression. A user enters a prompt. A model generates an answer. Productivity appears almost instantaneous. The impression created is that artificial intelligence removes complexity from the organization. Increasingly, the opposite may be true. The model itself may be relatively easy to access. Building a production environment around it rarely is.
Professional-services firms increasingly discover that AI creates an entirely new layer of operational work. Models must be connected to data sources. Outputs must be integrated into workflows. Systems must interact with document repositories, ERP environments, CRM platforms, audit applications, knowledge-management systems, and client-facing tools. Someone must determine where human review is required, how decisions flow through processes, which outputs can be trusted, and where exceptions are handled. The AI may generate the answer, but the institution still needs a mechanism capable of using that answer safely, consistently, and repeatedly.
This pattern should feel familiar because the industry has encountered it before. Global delivery centers promised lower delivery costs but created entirely new coordination requirements. Shared-service organizations reduced duplication while increasing dependency on governance and process management. AI increasingly follows the same trajectory. The activity itself becomes cheaper. The system required to coordinate the activity becomes more complex. Productivity gains at the task level often create orchestration costs at the organizational level. What disappears from delivery frequently reappears somewhere else inside the operating model. Much of the economic value of AI may ultimately sit inside orchestration. Unfortunately, much of the hidden cost often sits there as well.
Governance: Why Defensibility Expands Faster Than Productivity
Every major technology wave eventually encounters governance. Artificial intelligence is reaching that point far faster than many firms expected.
The first demonstrations focused on capability. Models could summarize documents, generate code, review contracts, identify anomalies, and draft reports in seconds. The obvious conclusion was that large parts of professional work were becoming dramatically more efficient. Yet capability and acceptability are not the same thing. Professional-services firms do not merely produce outputs. They produce outputs that must be trusted. Audit opinions must withstand regulatory scrutiny. Tax advice must remain defensible. Legal interpretations must survive challenge. Consulting recommendations must be explainable. In these environments, the question is rarely whether the answer is useful. The question is whether the institution can defend how the answer was produced.
This is where the economics begin shifting again. Every increase in automation creates a corresponding demand for control. Firms require audit trails, validation frameworks, model governance, documentation standards, human oversight, data lineage controls, risk reviews, and regulatory safeguards. The more important the decision, the larger the governance apparatus surrounding it becomes. What appears initially as a productivity initiative gradually evolves into a governance challenge. The technology may reduce execution effort, but it simultaneously creates new obligations around transparency, accountability, and institutional trust. In highly regulated environments, the cost of generating an answer often falls while the cost of proving that answer can be trusted rises. Defensibility increasingly expands faster than productivity.
Capital: The Layer Nobody Sees but Everybody Depends On
Most discussions about AI are ultimately discussions about technology. Increasingly, they may be discussions about capital.
Building a chatbot is relatively inexpensive. Building institutional AI capability is not. Large-scale AI environments require integrated data architectures, cloud infrastructure, cybersecurity controls, orchestration platforms, governance frameworks, workflow redesign, organizational change programs, and specialized expertise. The visible interaction between a professional and an AI assistant represents only a tiny fraction of the broader investment required to support it. Clients rarely see these investments. Many professionals using the tools never see them either. Yet the entire system depends upon them.
This is where the economics of AI begin colliding directly with the economics of the traditional partnership model. Historically, professional-services firms scaled primarily through people. Growth required hiring talent, developing expertise, and expanding client relationships. Capital requirements remained comparatively modest because much of the value creation process remained human rather than infrastructural. AI increasingly changes that equation. Competitive advantage depends less on access to individual tools and more on the ability to fund the broader environment supporting them. Data platforms, knowledge systems, governance environments, orchestration capabilities, and shared infrastructure all require sustained investment long before the benefits become fully visible.
This creates a structural tension already emerging across the industry. The organizations most likely to benefit from AI often require larger and longer-term investments than traditional partnership economics were designed to support comfortably. Capital therefore starts becoming strategically important in ways that resemble platform businesses far more than traditional professional-services firms. The conversation gradually shifts from tools toward infrastructure. From productivity toward capability. From annual cost savings toward long-term investment capacity. This is also why AI increasingly intersects with broader themes such as private equity, platform economics, delivery industrialization, and centralized operating models. Once expertise starts behaving like infrastructure, firms require the ability to fund infrastructure.
Why Costs Flow Down the Stack
One of the most important observations within the AI Cost Stack is that value and visibility tend to move in the opposite direction of cost accumulation.
The highest visibility typically sits at the top. Clients see advisors. Partners see engagement teams. Boards see productivity gains. Success stories focus on faster delivery, improved analysis, and enhanced client outcomes. These benefits are real, which is precisely why AI adoption continues accelerating across the industry. Yet visibility declines rapidly as organizations move deeper into the stack. Few clients ask about orchestration architectures. Few partners spend significant time discussing governance frameworks. Even fewer conversations focus on the capital investments sitting underneath the entire system. The deeper the layer, the less visible it often becomes.
The economics increasingly flow in the opposite direction. Human cognition remains expensive, but much of the new investment associated with AI accumulates further down the stack. Machine cognition requires continuous capability development. Orchestration introduces workflow and integration complexity. Governance expands as firms seek to preserve trust and manage risk. Capital funds the infrastructure supporting everything else. As organizations scale AI adoption, these layers become increasingly interconnected. A weakness in one layer creates pressure elsewhere. Better models without orchestration create fragmentation. Greater automation without governance creates risk. Increased adoption without capital creates bottlenecks.
This is ultimately why AI behaves differently from many traditional productivity technologies. Most organizations instinctively focus on individual tasks because tasks are visible. The real economics increasingly sit at the system level. The firms generating sustainable value from AI are unlikely to be those optimizing individual layers in isolation. They will be the firms capable of managing the entire stack as an integrated operating system. The question is no longer simply how much productivity AI creates. The question increasingly becomes whether the institution can align all five layers effectively enough to capture that productivity at scale.
Closing Thoughts
The biggest misconception about AI in professional services is that it is primarily a labor story.
The narrative is appealing because it fits the industry’s historical economic model. Professional-services firms have traditionally scaled through people. More work required more professionals. More professionals required more management. More management required more structure. If AI can perform part of the work previously carried out by humans, it seems logical to assume that costs should fall accordingly. Much of the current discussion around productivity, utilization, staffing models, and workforce planning is built on exactly that assumption.
The AI Cost Stack suggests a different interpretation. AI increasingly behaves less like labor and more like infrastructure. Infrastructure rarely eliminates cost. It changes where costs occur. Railways shifted transportation costs into networks, maintenance, and coordination. Cloud computing shifted technology costs into platforms, consumption models, and operating environments. Global delivery centers shifted labor costs into governance, workflows, coordination, and integration. AI appears to be following a similar pattern. The visible effort required to perform individual tasks may decline substantially while the infrastructure required to support those tasks becomes increasingly sophisticated and capital intensive. The economics move. They do not disappear.
This does not mean AI will fail to improve productivity. It almost certainly will. Nor does it mean firms should avoid investing aggressively. Quite the opposite. The firms that build effective AI capabilities may gain significant advantages in quality, speed, scalability, and client service. The challenge is that many organizations still evaluate AI through a cost lens designed for a labor-based industry while increasingly operating inside a platform-based environment. As AI becomes embedded into professional-services firms, the strategic question shifts away from how many hours can be removed and toward how effectively the institution can manage the infrastructure underneath them. The future economics of professional services may therefore depend less on who adopts AI first and more on who understands the full stack supporting it.
What This Means for Boards
Many board discussions around AI remain focused on tools, pilots, productivity gains, and implementation roadmaps. These topics matter, but they often concentrate attention on the visible layer of the stack while overlooking the broader institutional implications underneath.
Boards should increasingly ask different questions. Which layer of the AI Cost Stack represents our primary source of competitive advantage? Which layers are we building internally, and which layers are we depending on external providers to deliver? Where are governance requirements likely to expand faster than expected? Which capabilities require long-term investment regardless of short-term productivity gains? And perhaps most importantly, who controls the infrastructure that increasingly determines how work is performed across the organization?
The firms most likely to benefit from AI may not be those deploying the largest number of tools or running the largest number of pilots. They may be the firms that understand AI as a system rather than a technology. A system spanning human expertise, machine intelligence, orchestration, governance, and capital. A system where visibility decreases as dependency increases. And a system where the most important strategic decisions often sit furthest away from the user interface. Understanding that system may become one of the defining governance challenges for professional-services boards over the next decade.
I work with boards and executive teams on independent perspectives related to professional-services transformation, governance, operating models, platform economics, and the changing economics of professional-services firms.
If your leadership team is working through similar questions around ownership structures, governance alignment, investment pressure, or operating-model evolution, you may find my Future of Professional Services board sessions and AI Economics Review valuable. Feel free to reach out.