Should Organisations Still Be Willing to Fund AI Investments in 2026?
AI adoption is no longer the central challenge facing most organisations. In previous years, the priority was understanding where AI could be applied and building the capability to use it. That has changed. Many organisations now recognise the need to integrate AI into their operations and have begun investing across different functions. What they have been slower to ask is how capital is being allocated across the growing number of AI tools, pilots, infrastructure investments, and transformation programmes now competing for funding.
That question is becoming more consequential because AI is moving beyond innovation and technology budgets. It now affects operating costs, workforce planning, productivity, customer strategy, risk management, and long-term competitiveness. For CEOs, CFOs, and boards, AI therefore sits directly within decisions about where capital should be committed, what outcomes that capital should produce, and how long an investment should continue before its value is reassessed. The rapid expansion of AI activity makes this allocation problem more visible. Organisations are accumulating costs across licences, cloud infrastructure, data preparation, systems integration, governance, training, and pilots running across different business units. At the same time, many still lack a consolidated view of what those investments are intended to produce, who is accountable for the outcome, and what evidence should determine whether further funding is justified.
The consequence is not simply higher AI spending. It is a growing portfolio of tools, pilots, infrastructure, and transformation programmes competing for capital without the same level of scrutiny applied to other major investments. As your organisation continues to invest, can you clearly identify which AI initiatives are creating value, which are building capabilities that require a longer investment horizon, and which are continuing because no one has decided when funding should stop?
More importantly, what should an AI investment have to prove before you commit more capital?
AI has become a capital-allocation issue
Looking across current AI investment patterns, the problem is not simply that organisations are spending more. The amount of capital being committed is increasing faster than the discipline for determining what that capital should produce.
Corporate AI investment is expected to increase from approximately 0.8% of annual revenue in 2025 to 1.7% in 2026. At the same time, 94% of organisations expect to continue investing even where AI does not produce immediate returns. AI budgets are therefore growing, but organisations are also becoming more willing to sustain those investments for longer before demanding evidence of financial value. That willingness is not necessarily misplaced. Some AI investments require time to mature, particularly where their value depends on data preparation, systems integration, process redesign, employee adoption, governance, or capabilities the organisation expects to need in the future. An investment may be strategically important before its full financial return becomes visible.
The difficulty is that the financial evidence is not developing at the same pace as the investment. More than half of CEOs report that AI has produced neither higher revenue nor lower costs, while only 12% report achieving both. McKinsey similarly found that although 88% of organisations now use AI in at least one business function, only 39% report any enterprise-level contribution to EBIT.
Recent cases show why this matters for capital allocation. In 2025, Commonwealth Bank of Australia announced that 45 customer-service roles would be removed following the introduction of an AI voice bot. Within weeks, the bank reversed the decision, apologised to the affected employees, and acknowledged that its original assessment had not considered all the relevant business factors. Reports also indicated that customer demand remained high, with staff being offered overtime and team leaders being moved onto calls. The technology may have been capable of handling some enquiries, but the evidence was not strong enough to support the wider operating and workforce decision made around it. Cando Rail & Terminals faced a different decision. The Canadian rail operator invested approximately US$300,000 in an internal chatbot intended to help employees navigate a 100-page safety rulebook. During testing, the system sometimes omitted rules and, in other cases, produced rules that did not exist. Cando paused the project rather than extending the investment into a safety-critical operation without sufficient evidence that the system could perform reliably.
The two cases do not suggest that AI investment is inherently unsuccessful. They show that organisations are committing capital to investments that may require very different decisions:
- a strategically important capability that justifies a longer investment horizon;
- a controlled pilot that is still testing a defined commercial or operational assumption; or
- an initiative that continues to consume capital even though its original case has weakened or has never been properly revisited.
The capital-allocation problem becomes harder to control when these investments are grouped together under the same language of experimentation, transformation, and future capability. Without different funding expectations for each, strategic patience can be used to protect weak projects, activity can be presented as progress, and the absence of returns can be treated as a reason to wait rather than a reason to reassess.
Understanding which AI investments should still receive funding therefore requires a closer examination of how weak projects survive, why their original assumptions are not revisited, and where accountability for their commercial outcomes begins to break down.
Does this make AI a poor investment?
Certainly not. AI can create measurable value when it is applied to a defined business problem and assessed against a clear operating baseline. In a study of professional knowledge work, access to generative AI reduced completion time by approximately 40% and improved output quality by 18%. For organisations producing large volumes of reports, research, financial analysis, software, compliance documents or technical materials, those improvements can support a credible investment case. AI can increase capacity, shorten delivery times, improve the consistency of routine work and allow experienced employees to focus on more complex decisions. It can also support revenue growth through faster product development, stronger forecasting and services previously too expensive or time-consuming to deliver. Completing a task faster does not, however, automatically create a financial return. The benefit becomes commercial value only when the organisation uses the additional capacity to increase output, improve quality, shorten delivery times, reduce external expenditure, avoid recruitment or remove costs from the wider process.
This is why workflow redesign matters. McKinsey found that the strongest AI performers were nearly three times more likely to have fundamentally redesigned workflows. Senior leaders were also more likely to own those initiatives visibly. The investment case therefore extends beyond the AI tool to the data, integration, governance, employee capability and process changes required to create value. These investments can take longer to mature, provided the organisation can explain what capability is being created and what progress should precede further funding. Problems emerge when there is no agreed standard for what an investment must prove. A pilot can begin as a cost-reduction initiative, later be described as a productivity programme and eventually be defended as capability building or organisational learning. Each objective can justify investment, but the objective should not change simply because the original result failed to appear. Without an evidence threshold, almost any activity can be presented as progress. A project can request more data, another integration, further employee training, a newer model or a longer testing period. What begins as strategic patience becomes continued expenditure without a clear basis for judging whether the investment is growing stronger or weaker.
This is the difference between funding an uncertain project and continuing to fund a weak one. An uncertain project receives limited capital to answer a defined question. A weak project continues after the evidence has reduced confidence in its original case, without its funding assumptions being formally reconsidered.
The Cost of Allocating Capital Poorly
The direct financial cost of a weak AI project may be relatively small. The wider cost can be much greater.
First, the misallocation of capital can marginalise investments that are more critical to the organisation’s long-term health: AI initiatives do not merely consume financial resources, they demand the attention of data engineers, cybersecurity specialists, legal teams, and senior executives, while testing the capacity of the workforce to manage further transformation. When these limited resources are tethered to a specific project, they are effectively withheld from other strategic priorities. This creates a hidden opportunity cost, where essential infrastructure upgrades are deferred, legacy systems remain unaddressed, or more compelling AI applications lack the technical support required to reach maturity. An organisation may find itself sustaining an underperforming pilot while the very investments that could drive productivity are left starved of funding. Consequently, poor capital allocation does not just exhaust budgets, it can leave an enterprise strategically vulnerable in areas where that capital would have yielded a more substantial return.
Second, the cost of an AI investment can multiply between pilot and scale: A limited pilot can appear to include only the software or model being tested. Wider deployment introduces data preparation, integration, access controls, security testing, monitoring, employee training, human validation and continuing model-usage costs. Where pricing depends on tokens, processing volume or automated actions, expenditure can also change continuously as usage increases. Amazon provides a recent example. An internal project using Anthropic's Claude Sonnet to match author details with product listings reportedly accumulated a US$1.8 million bill, exceeded its budget by 860% and continued for five months before the overrun was identified. The system was not launched. Other internal projects involving financial auditing and logistics also generated substantial unexpected costs. The lesson is not simply that AI projects can become expensive. Usage-based costs can accumulate without the controls normally applied to major technology expenditure. An affordable trial can become an uneconomic operating commitment when the organisation has not modelled how integration, infrastructure, oversight and usage costs will change at scale.
Third, AI investments can create dependence that makes future decisions more expensive: Once employees, organisational data and operating processes become connected to a vendor or platform, stopping the investment may require new integrations, retraining and migration costs. Capital allocation therefore affects not only current expenditure but the organisation’s ability to change direction later. These risks will become more material between now and 2029 because enterprise AI is moving beyond tools that assist individual employees. Gartner expects that by 2028 more than half of enterprises will stop paying primarily for assistive tools such as copilots and smart advisers and will instead favour systems expected to deliver complete workflow outcomes. That shift could produce greater value because the system is assessed against a completed business outcome rather than a single task performed within the process. Reconciling an invoice creates more value than extracting information from it. Identifying, validating and routing a compliance issue is more commercially significant than summarising the relevant regulation. Detecting a production fault and initiating the appropriate response produces a clearer operating result than merely flagging an anomaly.
Gartner also expects task-specific AI agents to be integrated into 40% of enterprise applications by the end of 2026. Yet it also forecasts that more than 40% of agentic AI projects will be cancelled by the end of 2027 because of rising costs, unclear business value or inadequate controls. Organisations will be under pressure to increase investment in systems with greater operational capability at the same time, a significant number of those projects may fail to establish sufficient value or control. The cost of poor capital allocation will therefore extend beyond a wasted licence or unsuccessful pilot. It may include disrupted workflows, incorrect automated actions, security exposure, regulatory intervention and dependence on infrastructure that the organisation can no longer easily replace.
Regulation will reinforce that pressure. Transparency requirements under the EU AI Act begin applying on 2 August 2026. Requirements covering certain high-risk uses, including employment, education, essential services and critical infrastructure, are scheduled to apply from 2 December 2027, with rules for AI embedded in regulated products applying from 2 August 2028 which changes the economics of AI investment. Documentation, risk classification, human oversight, monitoring and auditability cannot be treated as costs to address after a project succeeds. They affect whether the investment can be deployed, whether it can scale into more consequential processes and whether the organisation can continue using it as regulatory expectations increase. Organisations that allocate capital on the assumption that AI value will come primarily from removing employees may therefore invest against an incomplete operating model. By 2029, the capital question will no longer be limited to whether employees are using AI or whether a model can perform a task. Organisations will need to determine whether each system can be trusted with enough data, authority and operating responsibility to justify its complete financial, regulatory and commercial exposure. Poor allocation today can therefore create a much larger constraint tomorrow. Organisations are not simply purchasing tools.
The next stage of AI investment requires organisations to move from funding activity to funding proof.
This begins with separating the different reasons organisations invest.
- Financial return: the investment is expected to improve revenue, margin, costs or cash flow.
- Operational value: it should improve speed, quality, capacity, resilience or risk management.
- Capability building: it develops the data, systems, governance or workforce capability required for future priorities.
- Strategic learning: limited capital is committed to testing a specific assumption before a larger investment decision is made.
Each can justify investment, but they cannot be assessed against the same evidence threshold, time horizon or funding expectation. A cost-reduction initiative should not become a capability programme after the expected savings fail to appear, and a pilot should not continue indefinitely because it is still generating useful learning. At the same time, a foundational investment should not receive open-ended funding without showing which future capabilities it is enabling. The intended value must remain clear throughout the investment period. The same discipline applies beyond corporate portfolios. Governments and development institutions funding AI-enabled public services must also distinguish strategic capability from uncontrolled experimentation. Their returns can appear through faster services, reduced leakage, stronger compliance or improved programme outcomes rather than profit, but the intended value, accountable owner and evidence threshold must remain explicit.
The Pacepoint SCALE Test
Before an AI initiative receives significant additional funding, executives should assess it across five areas: strategic relevance, commercial value, accountable ownership, launch and operating readiness, and evidence of impact.
S — Strategic relevance
The business problem must come before technology. An organisation should be able to explain what needs to change, why it matters commercially or operationally and why AI is the appropriate response. Executives should identify which organisational priority the investment supports, what the cost of inaction would be and whether the same outcome could be achieved more effectively through process improvement, conventional automation or technology already available. A competitor announcement, vendor demonstration or pressure to show visible AI activity is not a strategic case.Strategic relevance also determines how long an investment should be allowed to mature. An initiative supporting a near-term cost target should produce evidence sooner than one building a long-term data or governance capability. A clear strategic connection can justify patience. A weak connection allows time to become a substitute for accountability.
C — Commercial value
Commercial value requires a credible connection between the investment and an identifiable financial or operational result. The intended value could come through higher revenue, stronger margins, increased capacity, faster delivery, reduced losses, lower operating costs, avoided future recruitment, improved risk management or better capital utilisation. It does not need to appear as immediate profit, but it must be explicit enough for executives to understand what the organisation expects to gain. The investment case must also reflect the complete cost of implementation. And, the cost of a licence or initial pilot provides only a partial view of the financial commitment.
A small trial often operates with limited users, clean data and close support from the implementation team. Wider deployment introduces larger volumes, more exceptions, greater infrastructure requirements and stronger control obligations. The economics must therefore be tested against the cost of reaching and maintaining operational scale, not only the cost of demonstrating technical feasibility. The value calculation requires the same discipline. Hours saved should not automatically be recorded as financial savings. Released capacity becomes commercial value when it produces more output, reduces external expenditure, avoids future hiring, shortens delivery times or improves the performance of the wider operation. Executives need to understand both what changes and how the organisation will capture the benefit.
A — Accountable ownership
Every AI investment needs a named business executive who owns the result. Technology teams can select, develop and integrate the capability, but they cannot independently change operating processes, workforce responsibilities, customer journeys or business-unit performance targets. Accountability should therefore sit with the executive responsible for the function, process or financial outcome affected by the investment. An AI investment in procurement should have a procurement executive accountable for the commercial outcome, while a forecasting tool should have a finance leader responsible for how it changes planning and decision-making. The technology leader remains accountable for architecture, integration, security and technical performance, but business value mu
st be owned by the part of the organisation expected to capture it.
This ownership should extend from the original business case through adoption, workflow redesign, benefit realisation and the eventual recommendation to scale, pause or stop. Where no business executive is prepared to own the outcome, the initiative is not ready for significant additional capital.
L — Launch and operating readiness
A system can perform well in a controlled test and still be unfit for wider deployment. Launch and operating readiness considers whether the organisation can integrate the capability into real workflows, manage its risks and support it under normal operating conditions. This includes data quality, compatibility with existing systems, employee capability, process redesign, cybersecurity, governance, regulatory requirements, vendor capacity and the organisation’s ability to absorb further change. Many pilots succeed because the implementation team provides clean data, close supervision and manual support. Wider deployment exposes inconsistent processes, unexpected exceptions and costs excluded from the original estimate. These dependencies need to be understood before the organisation purchases additional licences, infrastructure or long-term vendor commitments.
Readiness also requires clarity about the role of human judgement. Executives should know which actions the system can perform, where human approval remains necessary, how exceptions will be handled and who remains accountable when an output is wrong. As AI systems gain more operational authority, these considerations become part of the investment decision rather than governance tasks to address after implementation.
E — Evidence of impact
The final test is what the investment has proved under real operating conditions.
A successful technical demonstration, positive employee feedback, high usage and increased activity all provide useful information, but none of them independently proves commercial value. Evidence of impact requires a baseline, a defined outcome, actual adoption, the full cost incurred and a credible measurement period. The evidence threshold should reflect the purpose of the investment. A financial-return initiative should show a credible path to revenue, margin, cost or cash-flow improvement. An operational investment should demonstrate measurable gains in speed, quality, capacity, resilience or risk. A capability investment should show what foundation is being built and which future priorities depend on it. A strategic learning investment should answer the specific uncertainty it was funded to test.
The evidence does not need to remove every uncertainty before more capital is released, but it should make the case stronger. Where additional spending produces more activity without increasing confidence in the expected outcome, the investment is not progressing.
Turning Assessment into a Funding Decision
The SCALE Test assesses the strength of the investment case. Organisations seeking to escape the AI Spending Trap need a structured approach to deciding what to fund, what to extend and what to stop. A Review places every investment in one of three categories: Scale, Pause or Stop.
Scale
An investment earns the right to scale when it has produced a measurable contribution to revenue, margin, cost reduction, operating performance or risk mitigation, and when the contribution can be traced to a defined business outcome. It also requires a named business owner, not only a technology owner, as well as a clear integration path showing how the capability will become part of the operating model rather than remain a standalone initiative. Scaling should mean more than purchasing additional licences or increasing user numbers. It should include workflow redesign, adoption expectations, performance measures, stronger controls and accountability for capturing the benefit. AI applications consistently meeting these conditions include fraud detection, supply-chain optimisation, targeted software-development acceleration and forms of automation where the outcome is specific, measurement is direct and the business case can be defended at board level. Before releasing more capital, executives should be able to explain what the initiative has already proved, what additional value scale is expected to create, which new costs and risks will emerge, who owns the result and when the investment will be reviewed again.
Pause
An investment warrants a structured pause when the strategic objective remains relevant but the financial case, evidence or operating readiness has not yet been established. A pause is not an indefinite extension of experimentation. It requires a specific question the investment is expected to answer, a limit on the additional capital available and a clear date for the next decision. It also needs success measures capable of supporting either a move to scale or a decision to stop. The unresolved issue could involve data reliability, employee adoption, model accuracy, integration costs, regulatory exposure, security, operating performance or the organisation’s ability to convert a productivity gain into measurable financial value. Without these conditions, a pause becomes continued spending under a different name. The initiative remains active, but the organisation is no closer to understanding whether it deserves a place in the portfolio.
Stop
An investment should be stopped when costs are compounding without measurable output, when no business executive owns the result, or when the original commercial objective has been revised repeatedly without being achieved.
The decision also applies where AI adds unnecessary complexity to a simpler solution, where the investment duplicates an existing capability, where employees continue to avoid or bypass the system, or where the full cost has moved beyond the value the initiative could reasonably produce. The difficulty with stopping AI projects is often institutional. Stopping an AI project should not automatically be treated as failure. Where an organisation has tested an assumption, identified the limits of the investment and prevented a larger loss, stopping demonstrates disciplined capital management.
Organisations should protect investments linked to material priorities, give strategic capabilities sufficient time to mature and continue testing opportunities capable of creating value. At the same time, they should limit funding where the evidence remains incomplete and withdraw it where the original case has weakened. The organisations that will benefit from AI will not be those with the largest portfolio of pilots or the highest number of licences, but those able to distinguish between strategic patience and delayed accountability, between useful experimentation and prolonged spending, and between a promising capability and an expense whose case no longer justifies more capital.
The Capital Discipline AI Now Requires
The AI spending trap does not begin with one unsuccessful pilot or an expensive software licence. It begins when organisations continue adding tools, infrastructure and transformation programmes without becoming more precise about what each investment is expected to produce. This becomes more costly as AI moves deeper into the operating model. Between now and 2029, organisations will be funding systems with greater access to data, stronger integration with core processes and more authority over day-to-day decisions. Poor investment choices made today can therefore become tomorrow’s recurring cloud costs, vendor dependence, compliance exposure, security obligations and operating processes that are expensive to reverse.
Our position is that organisations should continue investing in AI. The commercial opportunity remains significant, and withdrawing simply because returns have taken longer than expected would leave valuable productivity, growth and capability gains unrealised. But continued investment should not mean continued funding for every initiative already in the portfolio. Capital should move towards AI investments that solve material business problems, have accountable business ownership, show a credible path to value and can be integrated into the organisation without creating costs greater than the benefit. Projects still testing an important assumption should receive limited time and funding to prove it. Projects whose original case has weakened should stop consuming capital simply because the technology works or the organisation has already spent too much to feel comfortable walking away.
This is the discipline the next phase of AI investment requires. Not less ambition, but greater precision.
References
- Boston Consulting Group. (2026). As AI Investments Surge, CEOs Take the Lead.
- PwC. (2026). PwC 2026 Global CEO Survey: CEOs Confident About Revenue Growth in 2026 as Most Struggle to Turn AI Investment into Tangible Returns.
- McKinsey & Company. (2025). The State of AI: Global Survey 2025.
- Reuters. (2025). Australian Lender CBA to Cut 45 Jobs in AI Shift, Draws Union Backlash.
- Chalmers, S., ABC News. (2025). Commonwealth Bank Backtracks on AI Job Cuts, Apologises for ‘Error’ as Call Volumes Rise.
- Hu, K., Reuters. (2025). AI Companies Need Business Customers. Why That Could Take a While.
- Noy, S. and Zhang, W. (2023). Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence. Science, 381(6654), 187–192.
- Financial Times. (2026). Amazon Finds Cases of AI Causing Runaway Spending on Tech Projects.
- Gartner. (2026). Gartner Expects Most Enterprises to Abandon Assistive AI for Outcome-Focused Workflow by 2028.
- Gartner. (2025). Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up from Less Than 5% in 2025.
- Gartner. (2025). Gartner Predicts Over 40% of Agentic AI Projects Will Be Cancelled by End of 2027.
- European Commission. (2026). Navigating the AI Act.
