Why AI’s economic potential extends beyond labour substitution — and why governance may be a market enabler rather than merely a cost.
Aswath Damodaran recently published a fascinating analysis, AI’s Bar Mitzvah Moment: From Hype & Hope to Business Questions!, arguing that AI has reached the point where technological promise must give way to business fundamentals.
I couldn’t agree more.
After years of extraordinary technological progress, soaring valuations and unprecedented infrastructure investments, it is time to ask some rather ordinary questions: Who will pay for AI? How much? At what cost can these services be delivered? And which companies will ultimately capture the economic value?
Damodaran structures his analysis around three fundamental drivers of business value: market size, industry economics and competitive advantage.
He reminds us that technological revolutions create winners and losers, often in unexpected places. The companies building the infrastructure are not necessarily those capturing the greatest share of the economic benefits.
I find his framework compelling, particularly because it moves the discussion away from technological capabilities towards economic reality.
However, approaching the same questions from the perspective of enterprise AI transformation and governance, I would like to offer two complementary considerations.
The first concerns how we define and measure the economic value AI can create.
The second concerns whether regulation, governance and assurance should be viewed primarily as costs and constraints — or whether, under certain conditions, they may actually expand the commercially addressable market.
1. Is labour substitution the right starting point for measuring AI’s economic potential?
In estimating AI’s total addressable market, Damodaran uses employee compensation as an important starting point.
He estimates total US employee compensation at approximately $12.96 trillion in 2025, then considers how much of this expenditure could potentially be addressed by AI products, depending on their ability to replace workers, pricing, applicability across industries and geographic reach.
This is a useful framework for understanding the economics of labour substitution.
Importantly, Damodaran also acknowledges that AI may enable entirely new businesses. In his August 21 update, he explicitly distinguishes genuinely incremental economic activity from revenues redistributed from existing businesses.
My question, therefore, is not whether he recognises these additional effects. He does.
It is whether labour substitution should occupy such a central position in estimating AI’s economic opportunity, given the different mechanisms through which AI may create value.
I would distinguish three.
Substitution: doing existing work with fewer resources
This is the most straightforward case.
An AI system automates tasks previously performed by employees, reducing the human capacity required to deliver a given output.
Here, labour costs provide a useful reference point.
The economic case depends on the difference between the current cost of delivering the work and the total cost of the new operating model, including AI services, integration, oversight, maintenance and transition costs.
But even here, technically automatable work should not be confused with economically replaceable labour.
A task may be automated without eliminating a position, particularly when employees perform multiple interconnected activities.
Augmentation: creating more value with existing resources
Consider an AI-supported financial analyst who can evaluate more investment scenarios, identify additional risks or support better management decisions.
The economic benefit is not necessarily a reduction in the analyst’s salary.
It may be higher revenues, avoided losses, better capital allocation or increased output.
Of course, productivity improvements do not automatically translate into financial results.
If an employee completes a task in four hours instead of eight, the business has not necessarily realised four hours’ worth of savings.
The economic value depends on what happens to the released capacity and whether it improves the company’s cash flows.
Nevertheless, the potential benefit is not necessarily bounded by the cost of the labour involved.
Creation: enabling new products, services and markets
AI may make highly personalised services commercially feasible, enable new business models or allow companies to serve customer segments that were previously uneconomical to reach.
These opportunities cannot be fully estimated by examining today’s payroll.
This does not mean that every AI-enabled revenue stream represents new economic value.
Some new businesses will replace existing ones. Others will generate additional economic activity.
Damodaran is right to insist on distinguishing the two.
However, I believe an additional distinction is essential.
Value creation is not the same as value capture — and neither is the same as AI-provider revenue
Consider a company that pays an AI provider $100,000 annually for a solution that enables it to generate $1 million in incremental operating cash flow before the AI fee and other implementation costs.
The AI provider records $100,000 in revenue.
The adopting company may capture a substantially larger economic benefit.
And the net benefit to the wider economy will depend on how much of that value is genuinely additional rather than transferred from competitors, customers or workers, as well as the resources consumed in producing it.
These are three different questions:
AI-provider revenues: How much can technology providers charge for their products and services?
Enterprise value capture: How much incremental economic value can adopting organisations retain?
Net economic value creation: How much additional value does AI generate across the economy, after accounting for costs and displacement effects?
The three are related, but they are not interchangeable.
This distinction is particularly important when using the size of the labour market to estimate AI’s potential.
Employee compensation can provide a useful benchmark for a labour-substitution scenario.
But the potential economic value created by AI — especially through augmentation and entirely new services — cannot be reduced to the amount currently spent on human labour.
Conversely, even very substantial economic benefits for adopting companies do not automatically translate into equivalent revenues for AI providers.
The critical question is not simply how much work AI can replace, but where economic value is created, who captures it, and how much of that value is genuinely incremental.
And this brings me to a second challenge: even when the potential value is clear, realising it is another matter entirely.
2. From AI adoption to economic value: Three gaps
Geoffrey Moore’s Crossing the Chasm offers a useful perspective.
Moore argued that technology adoption is not a smooth progression from innovators to mainstream customers. Early adopters and the early majority have fundamentally different expectations.
Visionaries are willing to experiment with promising technologies. Pragmatic customers want proven solutions, reliable integration, references and predictable outcomes.
Moore’s concept of the whole product captures this distinction: successful mainstream adoption requires more than an impressive technology. It requires a complete solution that enables customers to realise the promised value.
I believe this idea is particularly relevant to enterprise AI.
We tend to measure adoption through user numbers, licences, deployment statistics or reported productivity improvements.
Yet these indicators tell us surprisingly little about whether sustainable economic value has been realised.
I would distinguish three gaps:
The adoption gap: Between access to AI technology and its meaningful, sustained use.
The value gap: Between using AI and translating its capabilities into measurable business outcomes.
The transformation gap: Between achieving productivity improvements in individual tasks or processes and redesigning the organisation’s operating model to capture their full potential.
These are not stages in Moore’s original framework, but an extension inspired by his work.
An organisation can close the adoption gap while leaving the other two almost entirely untouched.
Employees may use AI extensively and become faster at individual tasks, while the company continues to operate through essentially unchanged processes, organisational structures and decision-making systems.
The resulting productivity gains may remain fragmented or economically unrealised.
For enterprise AI, the whole product therefore extends beyond the underlying model. It includes process redesign, data infrastructure, workforce capabilities, system integration, decision rights, operational controls and performance measurement.
The technology may be ready for the enterprise before the enterprise is ready to capture its value.
And this leads to my second consideration: the economics of AI governance.
3. What if governance helps AI cross the chasm?
Damodaran rightly identifies regulation and social constraints as factors that can influence AI economics.
Compliance requirements, privacy restrictions, infrastructure limitations and the political consequences of workforce displacement may increase costs or restrict market opportunities.
These are legitimate considerations in business valuation.
But does a cost-focused assessment capture the full economic effect of governance and assurance?
Consider the difference between an early adopter experimenting with an AI agent and a regulated financial institution considering deploying that technology in a critical business process.
The early adopter may tolerate uncertainty and work around limitations.
The financial institution needs credible evidence of reliability, clear accountability, effective controls, ongoing monitoring and the ability to intervene when something goes wrong.
Without these elements, the technology may be technically impressive but commercially unusable.
This is precisely where Moore’s whole product becomes relevant.
For many mainstream enterprise customers, assurance, governance and operational reliability are not optional additions. They are part of what makes the product commercially viable.
And if these capabilities enable adoption that would otherwise not occur, they have economic value.
From compliance cost to market access
Consider a simplified, hypothetical example.
An AI provider has developed a solution that could technically serve 1,000 enterprise customers.
However, only 200 organisations are currently willing to deploy it. The remaining companies are concerned about reliability, accountability, security or their ability to demonstrate compliance.
Assume that each active customer generates $100,000 in annual revenue for the provider, with a contribution margin of 40%.
The current business generates:
200 customers × $100,000 × 40% = $8 million in annual contribution.
Now suppose the provider invests in stronger governance capabilities, credible assurance, monitoring and operational safeguards.
As a result, another 400 customers become willing to adopt the solution.
The provider’s per-customer contribution margin declines from 40% to 35% because of additional operating costs.
The expanded business generates:
600 customers × $100,000 × 35% = $21 million in annual contribution.
That represents an increase of $13 million in annual contribution before accounting for the upfront investment and other incremental costs.
These figures are illustrative, not empirical estimates.
But they demonstrate an important mechanism.
Governance can increase unit costs while improving the economics of the business by expanding the commercially accessible market.
The question is therefore not simply how much compliance costs.
It is whether additional market access, stronger customer retention, reduced risk exposure and potentially greater pricing power justify those costs.
The valuation question
In a discounted cash flow framework, we could compare two scenarios.
Scenario A — Limited governance and assurance
Lower initial investment and operating costs, but potentially slower enterprise adoption, a smaller accessible market, greater exposure to risk-related losses and weaker customer retention.
Scenario B — Stronger governance and assurance
Higher upfront and ongoing costs, but potentially faster adoption, broader market access, lower expected risk losses and more durable customer relationships.
The difference in value would depend on the incremental future free cash flows generated by the second scenario, discounted appropriately for their timing and risk.
Conceptually:
Incremental value from governance = Present value of incremental free cash flows attributable to governance.
Those incremental cash flows would reflect additional revenues and any margin improvements, less operating costs, required reinvestment and taxes, together with changes in expected risk-related losses.
Benefits already reflected in projected cash flows must not be counted again through separate valuation adjustments.
The outcome is not predetermined.
Poorly designed governance may create substantial costs without improving adoption or reducing risk.
However, proportionate and credible governance may generate economic benefits that exceed its costs.
And the effects will likely differ between consumer AI applications and enterprise systems operating in highly regulated or sensitive environments.
Reducing information asymmetry
There is another economic mechanism worth considering.
AI systems can be difficult for enterprise customers to evaluate.
Buyers may struggle to assess reliability, failure modes, model limitations, data protection practices or performance in their own operating environment.
This creates information asymmetry between providers and customers.
Economist George Akerlof famously demonstrated how information asymmetry can damage markets when buyers cannot reliably distinguish between higher- and lower-quality products.
The same underlying problem may arise in enterprise AI.
If buyers cannot assess the quality and risk of competing AI solutions, they may delay purchasing decisions, demand significant price discounts or avoid the market altogether.
Credible assurance can help reduce that uncertainty.
Independent assessments, transparent documentation, meaningful performance evidence and ongoing monitoring may allow buyers to make better-informed decisions.
In this sense, assurance is not merely a compliance activity.
It can function as market infrastructure: helping buyers distinguish between solutions and enabling transactions that might otherwise not take place.
Of course, assurance only creates this benefit when its evidence is credible, relevant and sufficiently informative. A certificate alone does not guarantee that a system will perform reliably in a particular deployment.
Governance is not the same as regulation
An important distinction is necessary here.
Regulation establishes legal requirements and constraints.
Governance defines how organisations direct, control and remain accountable for AI systems.
Assurance provides evidence that systems meet relevant requirements and perform as intended.
Their economic implications differ.
Regulation can establish minimum standards, but it can also impose disproportionate obligations.
Governance can improve operational reliability, but excessive bureaucracy may obstruct innovation.
Assurance can reduce information asymmetry, but only if it provides information that buyers find credible and useful.
None of these mechanisms automatically creates economic value.
The question is under what conditions they improve market access, lower transaction costs, reduce risks or support sustainable competitive advantage.
This is not an argument that more regulation necessarily produces better economic outcomes.
It is an argument for evaluating regulatory and governance mechanisms according to their actual economic effects, including benefits as well as costs.
Damodaran himself recognises trust as a potential competitive differentiator, particularly when customers grant AI systems access to sensitive information.
I would take that insight one step further.
Trust may influence not only which provider a customer chooses.
It may determine whether the customer enters the market at all.
And that brings us back to Moore.
What if governance is not merely the cost of AI adoption, but one of the conditions for crossing the chasm?
If so, its economic contribution deserves a place alongside the familiar drivers of revenue growth, operating margins, reinvestment and risk.
4. AI’s economic adulthood: From technological possibility to realisable value
I share Damodaran’s conviction that AI has reached a stage where business fundamentals must take centre stage.
Extraordinary technological capabilities cannot indefinitely substitute for evidence of revenues, profitability, sustainable competitive advantage and returns on invested capital.
But understanding AI’s economic maturity requires us to distinguish several questions.
How much revenue can AI providers generate?
How much economic value can enterprises create by adopting AI?
How much of that value is genuinely incremental rather than redistributed?
What prevents technically feasible AI applications from becoming commercially viable?
And how do organisational readiness, governance and assurance influence that transition?
These questions are related, but their answers may differ significantly.
An AI application may create substantial value for an adopting business while generating relatively little revenue for its technology provider.
A technically impressive product may fail commercially because customers cannot deploy it reliably or responsibly.
And a governance investment that appears costly when viewed in isolation may become economically attractive if it expands the accessible market or reduces future losses.
I particularly appreciate Damodaran’s invitation to challenge and extend his framework.
My two suggestions are therefore not arguments against applying business fundamentals to AI.
They are arguments for examining both the economic value AI makes possible and the conditions required to realise it.
If AI is entering economic adulthood, perhaps we should judge its success not only by how much human labour it can replace, but by how much sustainable economic value it can create — and how effectively technological possibility can be converted into commercially viable, widely trusted solutions.
This article is a response to Aswath Damodaran’s “AI’s Bar Mitzvah Moment: From Hype & Hope to Business Questions!” (August 20, 2026, updated August 21), drawing also on Geoffrey A. Moore’s “Crossing the Chasm” and George Akerlof’s work on information asymmetry.
References
Aswath Damodaran (2026). AI’s Bar Mitzvah Moment: From Hype & Hope to Business Questions!, Musings on Markets. https://aswathdamodaran.blogspot.com/2026/08/ais-bar-mitzvah-moment-from-hype-hope.html
Geoffrey A. Moore (1991; revised editions 1999, 2014). Crossing the Chasm: Marketing and Selling High-Tech Products to Mainstream Customers.
George A. Akerlof (1970). The Market for “Lemons”: Quality Uncertainty and the Market Mechanism. Quarterly Journal of Economics, 84(3), 488–500.
Governance is what lets AI value survive scrutiny. Start with a decision trail you can defend.