Why knowing when to stop an AI investment matters as much as knowing what to automate.
Organisations are investing heavily in artificial intelligence, yet many still cannot demonstrate what that investment is worth.
This week’s research provides a particularly revealing picture.
Gartner reports that only 13% of CIOs believe their organisations have realised significant or transformational value from AI. Meanwhile, a separate CEO survey finds that business leaders are continuing to increase technology investment despite difficulty demonstrating financial returns.
There is nothing inherently wrong with investing ahead of measurable benefits. Transformation often requires that.
The problem arises when organisations have no credible mechanism for determining whether those benefits are materialising, whether the original assumptions remain valid or whether investment should continue.
The next discipline in enterprise AI transformation may be knowing when to stop.
Not because the technology has failed. Because the business case has.
The hidden cost of keeping AI initiatives alive
Consider a typical enterprise AI programme.
An initial use case demonstrates promising results. Employees complete certain activities faster. The pilot receives positive feedback, and the technology team confirms that the solution performs within agreed technical parameters.
The initiative progresses into production.
Over time, additional licences are purchased, integrations become more complex and operational support requirements increase.
Employees continue using the solution, although some of the original productivity assumptions prove difficult to verify.
The business may still be deriving value. However, nobody can confidently explain how much.
Meanwhile, the initiative has become an established part of the technology portfolio.
Discontinuing it now requires a decision that nobody particularly wants to own.
This is where economic leakage begins.
The organisation continues funding an activity because it exists, rather than because its current contribution justifies the investment.
The cost is not limited to technology expenditure.
Management attention, employee time, implementation capacity and organisational change resources remain committed to an initiative that may no longer represent the best available opportunity.
Every pound allocated to an underperforming initiative is a pound unavailable for something with greater potential.
Why the economics become difficult to see
Traditional technology investment models generally distinguish between implementation expenditure and ongoing operating costs.
Enterprise AI introduces additional variables.
Consumption-based pricing can fluctuate with usage. Agentic workflows may require repeated model interactions, monitoring, exception handling and human review.
New controls, integrations and operational responsibilities can increase the total cost of ownership.
At the same time, the benefits are often distributed across departments.
A faster task in one team might create more work elsewhere. An automated response may still require verification before a consequential decision. A reduction in processing time might release employee capacity without reducing expenditure or increasing revenue.
These outcomes are not necessarily failures.
They become failures of investment governance when the organisation records them as financial benefits without demonstrating how that value was realised.
Time saved is a potential source of value. It is not automatically a financial return.
The economic mechanism must be explicit.
Will the organisation process more transactions with the same resources? Reduce avoidable expenditure? Improve customer retention? Increase revenue-generating capacity? Avoid future recruitment? Reduce operational losses?
Each mechanism requires different evidence.
Without that distinction, AI investment decisions can become increasingly detached from business performance.
The missing discipline: Stop, redesign or scale
Gartner’s latest CIO research identifies a particularly important weakness.
Only 31% of CIOs reported having guidelines for preventing unsuccessful AI pilots from progressing into production.
Just 17% reported criteria for discontinuing low-value AI initiatives already operating in production.
These findings matter because investment governance should not end when deployment begins.
In fact, production is often when the organisation first obtains reliable evidence of operational value.
Pilot environments rarely reproduce every exception, dependency, approval requirement or workforce consequence of real enterprise operations.
The business case should therefore remain open to challenge.
A well-governed initiative needs three possible outcomes.
Scale: The evidence demonstrates sustainable value, acceptable risk and sufficient organisational readiness to expand.
Redesign: The underlying opportunity remains attractive, but the workflow, responsibilities, controls or adoption model need to change.
Stop: The evidence no longer supports continued investment, or a better opportunity exists for the resources involved.
Stopping an initiative should not automatically be interpreted as failure.
It can be evidence that the organisation is allocating capital intelligently.
The greater failure is continuing to invest because nobody established the conditions under which the organisation would walk away.
The operating model determines whether value survives
A separate Gartner CEO survey published this week found that 63% of respondents identified radical operating-model redesign as a leading move for accelerating tangible financial returns from AI.
That is an important signal.
It does not prove that operating-model redesign will produce a particular return. It does indicate that senior executives increasingly recognise the limitations of adding AI to existing processes.
The distinction is fundamental.
An organisation can automate part of a workflow while leaving the surrounding approvals, hand-offs, responsibilities and performance measures unchanged.
Individual activities become faster. The overall process may not.
A more productive employee may simply accumulate additional tasks. A faster analysis may wait in the same approval queue. An automated customer interaction may create more exceptions for a human team to resolve.
In each case, technology has improved a component without necessarily improving the business outcome.
The operating model determines how those components work together.
That includes decision rights, information flows, workforce responsibilities, governance, incentives and the mechanisms used to measure performance.
The question is not simply what AI can do. It is what the organisation must change to benefit from it.
Workforce adoption cannot be separated from investment decisions
This week’s Gartner CIO findings reveal another substantial gap.
Only 4% of surveyed CIOs reported having explicit plans to redesign roles for a workforce combining people and AI agents.
This is difficult to reconcile with the scale of investment now taking place.
If AI changes how work is performed, the workforce implications cannot remain an afterthought.
New research from Charter reinforces the point.
Its AI Pulse study found that employees benefit more when they have influence over how AI fits into their work, receive practical support and see managers demonstrate effective use.
The research also identified a gap between the support managers believe they provide and what employees actually experience.
These findings are observational rather than proof of causation, but their operational implications are substantial.
Workforce adoption depends on more than access to technology.
Employees need to understand which activities are changing, where their judgement remains essential, how performance will be assessed and what happens when AI produces an uncertain or incorrect result.
Managers need to understand how to redesign responsibilities rather than simply demand greater output.
Organisations need to decide how released capacity will be used.
Without those decisions, AI can increase activity without improving outcomes.
The TFx operating-model perspective
This is consistent with the enterprise transformation experience informing the TFx approach.
In an anonymised Fortune 500 environment supporting approximately 15,000 users, AI adoption increased from 15% to 68%.
The wider transformation programme delivered measured improvements including 38% greater efficiency and 63% lower compliance effort, contributing to an estimated $10–12 million in annualised ROI.
These are engagement-specific outcomes, not benchmarks or guarantees.
Their relevance lies in how the transformation was approached.
Technology adoption was integrated with workflows, governance, employee enablement, operating responsibilities and business-value measurement.
The objective was not simply to increase usage. It was to improve how the organisation operated.
That distinction becomes even more important as AI capabilities become more autonomous.
An AI agent may be technically capable of completing a process, but the organisation still needs to establish whether that process should be automated, who owns the resulting decisions and how performance will be measured.
Governance-by-design is essential, although technical compliance alone does not establish economic value.
A safe, reliable and widely adopted AI system can still be a poor investment.
From use-case selection to continuous value realisation
The conventional AI investment process often concentrates its greatest analytical effort at the beginning.
Organisations identify opportunities, assess feasibility, approve funding and measure implementation progress.
A stronger approach treats value realisation as an ongoing management responsibility.
The original business case should identify the workflow being changed, the outcome being targeted and the mechanism through which improvement becomes economic value.
It should establish the evidence required to justify progression, the conditions that trigger redesign and the circumstances under which investment should stop.
Once deployed, the initiative should be assessed against actual operating performance.
That means examining the complete workflow, not merely the AI component.
The measurement should include direct costs, human oversight, exception rates, quality, risk, workforce effects and realised business outcomes.
Crucially, someone must have authority to act on those findings.
Finance should be able to challenge the economic assumptions. Business owners should be accountable for operational outcomes. Technology teams should provide reliable evidence of cost and performance. Risk and governance functions should ensure that increasing autonomy does not introduce unacceptable exposure. Workforce leaders should help redesign roles and embed the new ways of working.
These responsibilities belong within the operating model rather than being delegated entirely to an AI programme team.
The real test of AI investment maturity
The next stage of enterprise AI maturity will not be determined by how many use cases organisations deploy.
It will be determined by how consistently they convert those capabilities into measurable outcomes.
Some initiatives will deserve further investment. Others will require substantial workflow redesign. Some should be discontinued.
Organisations capable of making all three decisions objectively will be better positioned to allocate capital, manage risk and build sustainable organisational capability.
The objective is not to minimise AI investment. It is to maximise the value created by the resources committed to it.
If every AI initiative in your organisation had to justify its continued funding against actual business outcomes tomorrow, how many would survive?
Sources
- Gartner, 2027 CIO and Technology Executive Agenda findings, 19 October 2026.
- Gartner, CEO survey on AI financial returns and operating-model redesign, 20 October 2026.
- Charter, AI Pulse workplace research, 21 October 2026.
- Gartner, AI value and risk analysis, 19 October 2026.
