The question of what artificial intelligence does to professional services has typically been framed as a binary: will machines replace humans, or merely assist them? This framing misses the more consequential shift already under way. The relevant question is not whether AI compresses the time spent on technical mechanics—it plainly does—but whether this compression fundamentally reorders what clients are willing to pay for.
Consider the economics of a typical deal mandate. A significant portion of junior analyst hours has historically been consumed by financial model construction: linking balance sheets, building scenario analyses, formatting outputs. These hours have always been ancillary to the core value proposition, yet they’ve been billable because they were unavoidable. AI doesn’t eliminate this work entirely, but it does make it trivially cheap. What was once three days of analyst time becomes three hours. The client still needs the output, but the scarcity—and therefore the pricing power—has evaporated, or will do in coming months.
This creates a challenging dynamic. The mechanics that once justified a certain quantum of fees are now effectively free in relation to the cost of senior analyst especially when factoring in the consistency and integrity of the produced work. Meanwhile, the work that remains—judgement about valuation parameters, synthesis of market conditions, navigation of stakeholder dynamics—has always been valuable but was previously bundled with the technical labour. The question becomes whether clients will pay the same total fee when the bundle is unbundled — this is a burning question for firms who solely provide financial modelling services.
There are two plausible trajectories. The optimistic case holds that as technical barriers fall, differentiation migrates entirely to higher-order skills: strategic judgement, pattern recognition across deals, the ability to synthesise fragmentary information into actionable advice. In this world, AI liberates bankers from drudgery and allows them to focus on what they’ve always claimed to be best at.
The profession becomes more advisory, less executory. Accordingly margins might even expand as firms shed junior headcount but maintain fee levels.
The pessimistic case is that much of what has been called “judgement” was actually pattern-matching that AI can replicate at scale. Valuation bands for comparable transactions? A language model trained on decades of tombstones can produce credible ranges. Market conditions synthesis? Any analyst with access to Bloomberg can now ask an AI to digest overnight flow and surface relevant themes. Even softer elements—reading a room, understanding client anxieties—may prove more mechanistic than we care to admit. Tools such as formuly are listening to a meeting and working out the action plan for the analyst to execute efficiently afterwards.
The truth likely lies between these poles, but with important nuances. Judgement itself is not a monolith. Some forms of judgement—applying historical precedent to novel situations, identifying second-order consequences—remain genuinely difficult for AI. Other forms—choosing between three defensible valuation approaches, calibrating a fairness opinion within established parameters—are increasingly algorithmic, even if they’ve been dressed up as art.
This suggests that value creation in banking will bifurcate.
At the top end, a smaller cadre of professionals who genuinely excel at non-replicable judgement—those who can convince a board to pursue a transformative deal or navigate a hostile situation—will command premium fees.
At the mid-tier, competition will intensify as the technical moat disappears and clients become more discerning about what constitutes real advice versus well-formatted analysis.
The parallel to other industries is instructive. When Bloomberg terminals democratised market data, sell-side research didn’t disappear, but it did stratify. The most insightful analysts became more valuable; the median analyst became less so. Similarly, as AI commoditises financial modelling, the ability to tell clients something they couldn’t learn from their own AI-augmented finance team becomes the new hurdle for relevance.
There’s also a compositional effect worth noting. The soft skills that increasingly matter—building trust, managing relationships, synthesising qualitative factors—tend to accumulate with experience. Yet the* traditional path to developing these skills has run through years of building models and preparing presentations*. If AI short-circuits this apprenticeship, how does the next generation develop the intangible expertise that will define their later careers? This is not an idle concern and one banks must strategise carefully over both in recruitment and culture.
The human capital formation model of professional services assumes that junior grunt work is also junior training. Remove the grunt work and you may inadvertently remove the training. This requires the right platform that does not one-shot a solution but gives the analyst to learn whilst working like a superhuman.
One can imagine various adaptations. Firms might formalise mentorship and client exposure earlier. They might hire differently, prioritising communication and commercial instincts over pure technical facility. The structure of teams might flatten as the leverage model—one senior banker supervising five analysts—loses its economic logic. But these are significant organisational shifts, and incumbent firms are not known for their organisational agility.
The broader point is this: AI’s impact on investment banking is less about replacing human effort and more about redefining which human efforts command economic rents. The mechanical work that once justified the billable hour is becoming a commodity. What remains is work that requires genuine synthesis, contextual understanding, and interpersonal navigation. These have always been the real source of value, but they were previously cross-subsidised by the mechanics. Now they must justify the fee on their own.
Whether the profession can maintain its economics through this transition depends on an uncomfortable question: how much of what clients have been paying for was actually the judgment, and how much was simply the unavoidable cost of getting the mechanics done? If it was mostly the former, bankers have little to worry about. If it was partly the latter, there will be a reckoning.
First published on Nick Crawley’s Substack.
