AI adoption is no longer the challenge.The question is which investments deserve more capital.
AI adoption is no longer the central challenge facing most organizations. The more pressing issue is making better investment decisions. Many organizations have moved beyond asking where AI can be applied and have begun investing in tools, infrastructure, data, skills, and transformation programs across multiple functions. What remains less developed is a disciplined approach to deciding where capital should be allocated, what outcomes those investments should produce, and when continued funding is no longer justified.
This matters because AI spending now extends well beyond innovation and technology budgets. It affects operating costs, workforce planning, productivity, customer strategy, risk management, and long-term competitiveness. Yet many organizations still lack a consolidated view of their AI portfolio, clear accountability for results, or agreed evidence for determining whether an initiative should be scaled, redesigned, maintained as a longer-term capability investment, or stopped.
The central leadership question is therefore no longer whether to invest in AI. It is what an AI investment should have to prove before the organization commits more capital.
AI Investment Is Growing Faster Than the Evidence
Looking across current AI investment patterns, the problem is not simply that organizations are spending more. It is that investment is growing faster than organizations’ ability to determine what that capital should produce, how success should be measured, and when continued funding remains justified.
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 organizations expect to continue investing even where AI does not deliver immediate returns. That is not necessarily irrational. Many AI investments require time to mature, particularly where value depends on data preparation, systems integration, process redesign, governance, workforce adoption, or capabilities expected to create future competitive advantage.
The challenge is that evidence of value is not keeping pace with the investment. More than half of CEOs report that AI has produced neither higher revenue nor lower costs, while only according to PwC 2026 12% report achieving both. McKinsey Global Survey 2025,. found that although 88% of organisations now use AI in at least one business function, only 39% report any enterprise-level contribution to Earnings Before Interest and Taxes (EBIT).
Recent examples illustrate why these matters. According to ABC News Australia, the news division of Australia’s public broadcaster, in 2025, Commonwealth Bank of Australia reversed plans to remove 45 customer service roles after introducing an AI voice bot, acknowledging that its original assessment had not fully reflected operational realities, including continued customer demand. In such a case, the organization reassessed investment decisions when the available evidence no longer supported the original business case.
This example does not suggest that AI investment is failing. It shows that organizations are funding very different types of initiatives that require different investment expectations:
Strategic capabilities that justify a longer investment horizon. Controlled pilots testing defined commercial or operational assumptions. Initiatives that continue consuming capital even though their original business case has weakened or has never been revisited.
The challenge of allocating capital arises when these different categories are managed in the same way. The expectation that AI investments need more time can be used to protect weak projects, activity can be mistaken for progress, and the absence of measurable results can become a reason to continue investing rather than to reassess.
The next question is therefore not whether organizations should invest more in AI, but how leaders distinguish investments that deserve continued funding from those that no longer do.
AI Is Not the Problem. Poor Investments Decisions Are.
AI can create real, measurable value when it is applied to a defined business problem and assessed against a clear operating baseline. In one study of professional knowledge work, access to generative AI reduced completion time by approximately 40% and improved output quality by 18%. (Noy & Zhang 2023). For organisations producing large volumes of reports, research, financial analysis, software, compliance documents, or technical materials, improvements of this scale 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 that were previously too expensive or time-consuming to deliver.
Completing a task faster does not, however, automatically create financial value. Productivity gains become commercially meaningful only when the organisation uses the additional capacity to increase output, improve quality, accelerate delivery, reduce external expenditure, avoid recruitment, or remove costs from the wider process.
This is why workflow redesign matters. McKinsey found that organisations reporting the greatest value from AI were significantly more likely to redesign workflows.”. Senior leaders were also more likely to take visible ownership of those initiatives. The investment case therefore extends beyond the AI tool itself to the data, systems integration, governance, employee capability, and process changes required to create value.
These investments may take time to mature, provided the organization can explain what capability is being created, what progress should be expected, and what evidence will justify further funding. Problems emerge when there is no agreed standard for what an investment must prove.
A pilot may begin as a cost-reduction initiative, later be described as a productivity program, and eventually be defended as capability building or organizational learning. Each objective can justify investment, but the objective should not change simply because the original result failed to materialize.
Without a clear evidence threshold, almost any activity can be presented as progress. A project can request more data, another systems integration, further employee training, a newer model, or a longer testing period. The expectation that AI investments need more time can then be used to justify continued expenditure without a clear basis for determining whether the investment case is becoming 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 the organization formally reconsidering its assumptions, objectives, or funding.
The Hidden Cost of Poor AI Investment Decisions
The direct cost of a weak AI project may be relatively modest. The wider organizational cost can be far greater because poor investment decisions compound over time.
Poor investment decisions do more than waste money. They divert scarce technical expertise, management attention, and organizational capacity away from higher-value priorities. AI initiatives require data engineers, cybersecurity specialists, legal teams, operational leaders, and senior executives. They also draw on the workforce’s limited capacity to absorb further transformation. When these resources remain tied to underperforming projects, infrastructure upgrades are delayed, legacy systems remain unresolved, and stronger AI opportunities struggle to secure the support needed to succeed.
Costs also increase sharply between pilot and scale. A limited proof of concept may appear to involve only the software or model being evaluated. Wider deployment introduces additional requirements for data preparation, systems integration, access controls, security testing, monitoring, employee training, human oversight, and ongoing usage. Where pricing depends on tokens, processing volume, or automated actions, expenditure can continue rising as adoption expands.
According to a Financial Times report, Amazon provides a recent example. An internal project using Anthropic’s Claude Sonnet to match author information with product listings reportedly accumulated costs of approximately US$1.8 million, exceeded its budget by 860%, and continued for five months before the overrun was identified. The system was never launched. The lesson is not simply that AI projects can become expensive. It is that usage-based costs can accumulate without the financial controls normally applied to major technology investments. An affordable pilot can become an uneconomic operating commitment when organizations fail to model how infrastructure, integration, oversight, and usage costs will change at scale.
The financial consequences will become more significant as enterprise AI moves beyond tools that assist individual employees and towards systems expected to deliver complete business outcomes. Technology research and advisory firm Gartner forecasts that by 2028 more than half of enterprises will move away from paying primarily for assistive tools, such as copilots and smart advisers, and towards systems expected to deliver complete workflow outcomes. Completing an end-to-end business process creates far greater value than automating a single task, but it also creates much greater operational and financial exposure.
As AI becomes embedded in core business processes, investment decisions also become harder to reverse. Once organizational data, workflows, and employees become integrated with a platform, changing direction may require data migration, new integrations, retraining, and operational redesign. Gartner also forecasts that more than 40% of agentic AI projects will be canceled by the end of 2027 because of rising costs, unclear business value, or inadequate controls. Organizations will therefore be expected to invest in systems with greater operational capability while recognizing that many of those investments may never establish sufficient value or control.
Regulation will reinforce these pressures. Transparency requirements under the EU AI Act begin applying on 2 August 2026. Requirements covering certain high-risk applications, including employment, education, essential services, and critical infrastructure, are scheduled to apply from December 2027, with further obligations for AI embedded in regulated products following in August 2028. Governance, documentation, human oversight, monitoring, and auditability are no longer implementation considerations. They are prerequisites for deploying, scaling, and sustaining enterprise AI.
Poor investment decisions therefore become increasingly expensive over time. What begins as an underperforming pilot can evolve into a long-term commitment of capital, organizational capability, operational dependence, and regulatory exposure that becomes progressively more difficult and costly to reverse.
Four Reasons to Invest. Four Different Ways to Measure Success.
The next stage of AI investment requires organizations to move beyond funding AI activity and towards funding evidence. That begins by recognizing that organizations invest in AI for different reasons. Not every investment is expected to deliver an immediate financial return, and not every investment should be judged against the same criteria.
Broadly, AI investments fall into four categories:
- Financial return: The investment is expected to improve revenue, margins, costs, or cash flow within an agreed timeframe.
- Operational value: The investment should improve speed, quality, capacity, resilience, compliance, or risk management.
- Capability building: The investment develops the data, systems, governance, or workforce capabilities needed to support future priorities. Its value may emerge over a longer period but should still be assessed against defined capability milestones.
- Strategic learning: Limited capital is committed to test a clearly defined assumption before deciding whether a larger investment is justified.
The mistake is not investing across all four categories. The mistake is evaluating them as though they were the same investment. Each requires a different investment horizon, different evidence of progress, and different criteria for deciding whether additional capital should be committed.
The question for leaders is therefore no longer whether to invest in AI. It is whether each investment is producing the evidence appropriate to the reason it was funded in the first place. That is the purpose of the SCALE Test.
The Pacepoint SCALE Test
The question is no longer whether to invest in AI. It is whether an investment has earned the right to receive more capital.
The Pacepoint SCALE Test provides a simple framework for making that decision. Before releasing significant additional funding, executives should assess every AI investment against five questions.
S — Strategic relevance
Does the investment solve a business problem that matters? The technology should support a defined strategic objective, not exist because competitors are investing or vendors are promoting new capabilities. The expected investment horizon should also reflect the purpose of the initiative. Long-term capability investments require different expectations from short-term efficiency programs.
C — Commercial value
How will the organization capture value? Productivity gains alone are not enough. Faster work becomes commercial value only when it increases output, reduces costs, improves quality, strengthens resilience, or avoids future expenditure. The investment case should also reflect the full cost of operating at scale, not simply the cost of a successful pilot.
A — Accountable ownership
Who owns the business outcome? Every AI investment requires a named business executive accountable for delivering value. Technology teams own technical performance, but only business leaders can redesign processes, change operating models, and realize financial or operational benefits.
L — Launch and operating readiness
Can the organization deploy the capability successfully? Readiness extends beyond technical performance to include data quality, systems integration, governance, cybersecurity, regulatory compliance, workforce capability, and the organization’s capacity to absorb change.
E — Evidence of impact
What has the investment proved? Evidence should reflect the original purpose of the investment. Financial initiatives should demonstrate a credible path to improved commercial performance. Operational investments should show measurable improvements in speed, quality, resilience, or risk. Capability investments should demonstrate tangible progress towards future strategic objectives. Learning investments should answer the specific question they were funded to test.
From Assessment to Investment Decisions
The SCALE Test assesses the strength of the investment case. It should then lead to one of three funding decisions.
Scale
Increase investment where the initiative has demonstrated measurable value, has clear business ownership, and can be integrated into the operating model with confidence.
Pause
Continue only where the strategic case remains strong but important questions around value, readiness, adoption, or operating performance still require evidence. A pause should have defined objectives, limited additional funding, and a clear review date.
Stop
Withdraw funding where the original business case has weakened, evidence is no longer improving, ownership is unclear, or the investment has become more expensive than the value it can realistically deliver. Stopping a project after testing an important assumption is not failure. It prevents further capital from being committed to a case the evidence no longer supports.
Organizations that benefit most from AI will not necessarily be those running the largest number of pilots or purchasing the most licenses. They will be those that consistently distinguish between strategic investment and prolonged experimentation, between capability building and weak business cases, and between projects that deserve more capital and those that do not.
The Investment Approach AI Now Requires
The AI investment trap does not begin with one unsuccessful pilot or an expensive software license. It begins when organizations continue adding tools, infrastructure, and transformation programs without becoming more precise about what each investment is expected to produce.
The consequences will become more significant as AI moves deeper into the operating model. Between now and 2029, organizations will increasingly fund 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 difficult and expensive to reverse.
Organizations should continue investing in AI. The commercial opportunity remains significant, and withdrawing simply because returns have taken longer than expected could leave valuable productivity, growth, and capability gains unrealized. But continued investment should not mean continued funding for every initiative already in the portfolio.
Capital should move towards investments that address material business problems, have accountable business ownership, show a credible path to value, and can be integrated into the organization without creating costs greater than the benefits. Projects still testing an important assumption should receive limited time and funding to produce the required evidence. Projects whose original case has weakened should stop consuming capital simply because the technology works or the organization has already invested too much to feel comfortable walking away.
This is the investment approach the next phase of AI requires: not less ambition, but greater precision about what to fund, what to extend, and what to stop.
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.
