For the past few years, the corporate conversation about artificial intelligence has centred on adoption. Organisations have been asking which tools to use, where to automate and how quickly they can deploy. Those were reasonable questions while businesses were discovering what the technology could do. As AI becomes more widely available, however, the conversation needs to move towards what its use actually delivers.
A recent Financial Times discussion of the economics of AI enablers, alongside LGT’s distinction between today’s infrastructure providers and tomorrow’s adopters, brought this into focus for me. The investment question is where sustainable returns will accrue across the AI value chain. For boards and management teams, the same question applies within their own businesses: how much of the value created by AI will they capture?
An organisation can adopt AI successfully, improve productivity and reduce the cost of individual activities without materially increasing enterprise value. From the perspective of strategy, transformation and capital allocation, that distinction matters. The investment case now needs to explain how those improvements will translate into stronger revenue, margins, returns on capital or competitive position.
The gap between productivity and financial returns
The emerging evidence suggests that many businesses are struggling to make that connection. PwC’s 29th Global CEO Survey, covering 4,454 CEOs across 95 countries and territories, found that 56% reported neither higher revenues nor lower costs from AI. Only 12% reported both. McKinsey’s 2026 State of AI survey points to a similar gap: while 80% of respondents said AI had improved their individual productivity, only 37% reported a positive contribution to their organisation’s EBIT. Just 6% qualified as high performers, attributing at least 5% of EBIT to AI and describing its contribution as significant.
These figures suggest that the benefits of AI are easier to observe at the level of an individual task than in the economics of the whole organisation. Employees can produce documents more quickly, write code faster, retrieve information more efficiently and analyse larger volumes of data. Whether those improvements generate a financial return depends on what the business does with them.
Time saved might allow a company to serve more customers, bring a product to market sooner or operate with a lower cost base. It might also leave employees with additional capacity that is never put to productive use. The productivity improvement is real in either case, but the financial outcome is different.
Even when a gain reaches the bottom line, there is a further question about how long the company can retain it. Suppliers may raise their charges, customers may expect lower prices and competitors may adopt the same tools. Productivity, financial value and competitive advantage are therefore connected, but each requires something more than the one before it.
That progression is often missing from corporate AI discussions. A useful application is described as transformation before anyone has established how it changes the business. Deployment becomes the measure of progress, while the harder questions about economics and ownership remain unresolved.
Where does the value go?
To understand why, it helps to follow the value beyond the initial improvement.
A company introducing an AI platform may save substantial staff time while taking on new software licences, cloud expenditure, integration costs and ongoing support requirements. The investment can still make sense, but its attractiveness depends on the benefit remaining after those costs are included. Where a large share of the gain becomes revenue for technology providers, the adopting business may retain less value than the initial productivity figures suggest.
This is also where the build-versus-buy discussion becomes more useful. Most companies have little reason to develop their own foundational models or AI infrastructure. They do, however, need to understand which parts of the resulting capability they must own or control. Buying the technology can be entirely sensible when the company retains proprietary data, specialist knowledge, customer relationships or distribution that makes its application distinctive. The position is less secure when competitors can rent substantially the same capability on similar terms.
Some of the remaining benefit may then flow to customers. If AI lowers the cost of delivering a service, competitive pressure may force prices down or service expectations up. That can help protect market share and strengthen customer relationships, but it changes the return available to the business. A company that reduces delivery costs by 15% and passes most of the saving to customers has become more efficient without achieving an equivalent improvement in profitability.
As competitors adopt similar tools, that pressure can intensify. An early mover may enjoy a temporary lead, only to find that its improved performance soon becomes the industry standard. General productivity applications are particularly exposed to this pattern because they are widely available and relatively easy to replicate. They may be necessary investments, but management should understand whether they help the company catch up, keep pace or establish a position that others will find difficult to match.
Why the organisation has to change too
Yet some of the most significant obstacles to value capture arise before suppliers, customers or competitors have responded. They sit within the organisation itself.
A report may be produced in minutes rather than hours, while still passing through the same approval layers. Analysis may be completed faster, while the decision it supports remains delayed. A customer request may move more efficiently through one activity, then wait at an unnecessary handover. In each case, AI improves part of the work without materially changing the performance of the process.
The same problem arises when employees save time but the organisation neither reduces costs nor redirects their capacity towards more valuable activity. The business has created an opportunity to improve its economics, but has yet to act on it. Capturing that opportunity often requires changes to responsibilities, approvals, staffing, incentives and the sequence in which work is performed. This is why workflow redesign is so important: the financial benefit depends as much on how the organisation changes as on what the technology can do.
Different investments, different expectations
Consider a company introducing an off-the-shelf AI system into its finance function. The system processes invoices, classifies expenditure, supports reconciliation and identifies anomalies. Processing times fall, errors decline and fewer staff hours are required. With the right changes to the surrounding process, the investment may deliver a clear and worthwhile return.
Its strategic effect is likely to be more limited. Competitors can purchase similar software, and lower processing costs may eventually become the expected standard. The company is better run, although its position relative to competitors may be largely unchanged.
Now consider an industrial equipment company with thousands of machines operating at customer sites. It begins using data from that installed base to predict failures and improve maintenance. As its understanding of equipment performance develops, it starts offering contracts linked to availability or output, creating a recurring service relationship alongside equipment sales and replacement parts.
Here, the investment has the potential to change both the customer proposition and the revenue model. The installed base becomes a source of operating knowledge, that knowledge improves the service, and the service deepens the customer relationship. Historical data and integration into customer operations may also make the offering harder for a new entrant to reproduce.
Both investments can be attractive. What matters is that management understands their different purposes. The finance application should be assessed primarily on efficiency, reliability and cost. The equipment application also needs to be assessed on its potential to strengthen customer retention, recurring revenue and competitive position. Expecting every AI initiative to transform the business is as unhelpful as assuming that every productivity improvement creates a lasting advantage.
What boards should ask before committing more capital
For boards, this means bringing greater precision to the investment conversation. The starting point should be the business constraint an initiative is intended to address. If the problem is small, even an impressive technical result may have little economic significance. If it is material, management should be able to explain how the proposed investment changes the work required to solve it and where the resulting benefit will appear.
That explanation should connect operating measures to financial outcomes. Usage rates, adoption levels and hours saved can show that a tool is being used effectively. They need to lead to a credible account of additional revenue, lower costs, better capital utilisation or reduced risk. Without that connection, it is difficult to judge whether an initiative deserves further investment.
The discussion should then turn to how much of the benefit the company can retain. Proprietary data, specialist expertise, distribution, customer relationships and integration into customer operations can all help protect a return. Where the benefit rests mainly on access to widely available software, the business case should reflect the likelihood that competitors will achieve similar improvements.
Clear accountability is needed to carry these assumptions through to delivery. Someone must own the economic result, alongside performance and risk, and have responsibility for recommending whether the investment should continue, change or stop. That responsibility becomes particularly important as an initiative moves beyond a pilot. Integration requirements, vendor charges, human oversight and organisational complexity can alter the economics considerably at scale.
Scale, strengthen, reconfigure or stop
Taken together, these questions allow boards to manage AI as a portfolio of investments with different purposes, levels of uncertainty and expected returns. Some initiatives will justify scaling because the economics are proven and the benefits can be retained. Others will need strengthening through better data, clearer ownership or more effective processes. Some will require a more fundamental reconfiguration because the technology works but the surrounding workflow or business model prevents the value from reaching the company. Others should stop when the likely return no longer supports further expenditure.
Making those decisions is part of the normal discipline of capital allocation. Early experimentation helped businesses understand AI’s capabilities. As experience grows, that learning needs to inform firmer choices about where to commit resources and what to expect in return.
The companies that benefit most will be those that can follow the value from a faster task to a better process, from a better process to a financial result, and from that result to a return they can sustain. For boards and senior executives, the central question is how much of that value will remain in the business.
