The rise of ‘shadow productivity’ in the age of AI: Commentary

Much of today’s debate about artificial intelligence at work focuses on speed. Is AI being adopted too quickly? Will it displace workers? Are companies and their employees prepared for the disruption?

However, a quieter and potentially more consequential development may already be taking place inside organisations. AI is being used widely – just not always in ways companies can track, measure, or scale.

Many working professionals are already experimenting with, and benefiting from, generative AI tools in their daily work, often using their own personal accounts. A human resource recruiter summarises resumes in minutes. A communications executive drafts press releases faster. A finance analyst asks a chatbot to explain anomalies in a spreadsheet. A consultant generates a first version of presentation slides before refining them.

These are not speculative observations. Microsoft’s 2024 Work Trend Index, based on a survey of 31,000 workers across 31 countries, found that 75 per cent of knowledge workers globally report using generative AI at work. Interestingly, 78 per cent say they bring their own AI tools to work rather than relying on systems provided by their employers.

Other surveys paint a similar picture. Gallup in end 2025 estimates that 45 per cent of employees have used AI at work at least occasionally, with usage even higher among those working remotely.

Yet, at the organisational level, integration of such technology use remains uneven. Recent global surveys suggest that while experimentation with AI is widespread, relatively few firms have embedded it deeply into their workflows.

McKinsey’s latest 2025 State of AI survey finds that 78 per cent of organisations now report using AI in at least one business function. Yet, many remain in the early stages of scaling the technology across the enterprise. In fact, nearly two-thirds of organisations say they have not yet begun deploying AI at scale across multiple functions.

A similar pattern exists in Singapore. According to the Singapore Digital Economy Report by the Infocomm Media Development Authority, three in four workers here report using AI tools regularly, but adoption at the firm level remains uneven: only about 14.5 per cent of small and medium-sized enterprises (SMEs) reported adopting AI in 2024, compared with roughly 62.5 per cent of larger companies.

These figures highlight a central challenge – while individual workers may already be experimenting with AI, integrating it meaningfully into organisational systems is a much more complex and long-term undertaking.

As a result, individual adoption of AI may be outpacing enterprise adoption. Enterprise AI differs from consumer AI as it relies on up-to-date business data and is trained to deliver outputs relevant to a particular enterprise. It involves integrating those tools and software into existing large-scale operations.

Meanwhile, individuals’ quiet AI use is creating “shadow productivity” – efficiency gains generated informally that remain largely invisible to organisational systems designed for a pre-AI era. Technology researchers already speak of “shadow AI”, referring to employees using AI tools outside officially sanctioned company systems.

When workers quietly complete tasks faster or produce higher-quality drafts with AI assistance, those gains may not be captured in performance metrics, recognised in reward systems, or shared across teams.

And the bigger concern, then, may not be that AI is moving too fast. It may be that it is moving in the wrong layer of the organisation.

Personal AI versus enterprise AI

To understand why this matters, it helps to distinguish between personal AI use and enterprise level AI integration.

Consider a public relations executive using a personal AI tool to draft a media release. The tool helps refine language, suggest headlines and improve clarity. That is useful.

But imagine an enterprise-integrated AI system connected to the company’s internal databases. It could retrieve past press releases, analyse previous media coverage, flag compliance requirements and suggest journalists based on historical engagement patterns. Such a system would draw not only on general internet knowledge, but also on the firm’s institutional memory.

The difference is thus not simply speed. It is the ability to draw on models trained on proprietary data to improve organisational performance and drive consistent productivity gains, while integrating such capabilities into critical business systems compliant with corporate governance requirements.

Similarly, an HR hiring manager using a chatbot privately might summarise job applications more efficiently. An enterprise system integrated with the firm’s recruitment platform could compare candidates against historical hiring outcomes, flag diversity metrics and generate structured interview guides aligned with competency frameworks.

In finance, where an analyst might use AI to summarise a report, an integrated system could reconcile data across departments, detect anomalies against past trends and feed insights directly into forecasting dashboards.

Personal AI improves task efficiency. Enterprise AI reshapes workflows. This distinction matters because the most meaningful productivity gains from AI are system-level rather than individual-level.

AI creates sustained value when it reduces coordination costs, embeds organisational knowledge into processes and changes how work and risk is allocated across roles. It becomes part of how the organisation operates, rather than a private efficiency tool.

The invisible productivity problem

When AI use remains informal and individualised, several tensions emerge.

First, performance management systems struggle to keep up. If an employee completes work in half the time with AI assistance, what exactly is being evaluated – effort, output, judgment or oversight? Traditional appraisal frameworks often reward visible activity and task completion within established role boundaries and deadlines. They were not designed for AI-augmented work.

Second, rewards may not align with actual contribution. One employee who openly uses AI tools may be seen as more efficient, while another who uses them discreetly may gain no recognition at all. Managers may not even know which tasks are AI-assisted.

Third, organisations cannot easily institutionalise these gains. When AI use remains personal and ad hoc, best practices stay fragmented. There is no systematic way to embed learning across teams or redesign workflows collectively. Productivity gains remain individual rather than cumulative.

Over time, this creates a subtle but important misalignment. Workers adapt quickly. Organisational systems adapt much more slowly.

The SME capability gap

The implications are particularly significant for SMEs.

Large enterprises typically operate with enterprise resource planning systems, customer relationship management platforms and centralised data infrastructure. They employ IT teams and data specialists who can experiment with integrating AI into workflows. They also have the resources and scale to justify pilot projects and process redesign.

Many SMEs operate on more fragmented systems. Data may sit in spreadsheets, siloed applications or legacy software. There may be limited in-house capability to organise and connect this data in ways that allow AI tools to generate meaningful insights.

This does not mean SME employees are less capable of using AI. The challenge for them is organisational capability.

As earlier figures suggest, AI adoption among smaller firms remains far lower than among large companies. While adoption is rising quickly, the disparity highlights the complexity involved in integrating AI effectively.

Enterprise AI is not simply giving employees access to a chatbot. In practical terms, it requires three things.

Company information must be organised so it can be searched and retrieved reliably. This often means cleaning data, standardising records and ensuring information is stored in accessible formats.

AI systems must be connected to internal platforms – such as HR, finance or operations systems – so they can draw on real company data rather than generic external knowledge.

And workflows and oversight structures must be redesigned so AI outputs are embedded into daily processes with clear accountability, governance and performance metrics.

In short, enterprise AI is about making AI part of how the organisation works – not just a personal productivity tool.

Capability-building over panic

The prevailing narrative often frames AI as an imminent threat to the labour market. But the more pressing issue may be one of organisational readiness.

Digital payments and remote work did not transform overnight. They required infrastructure, regulatory adaptation and behavioural change. In Singapore, it took sustained policy support – and even a pandemic – to accelerate digitalisation.

AI integration is likely to follow a similar trajectory. It touches decision-making, knowledge work and organisational design. Meaningful transformation should unfold over several years, not in a single budget cycle.

This suggests that the conversation should move beyond fears of job displacement towards questions of capability-building. How can firms, especially SMEs, improve data infrastructure, upgrade systems and redesign performance management frameworks? How can managers learn to evaluate and reward AI-augmented work fairly and effectively?

Individual upskilling remains important. Workers should definitely still work towards understanding how to use AI tools responsibly and effectively. However, individual skills alone will not determine who benefits most from AI.

The real divide may emerge between organisations that redesign their systems and those that rely on informal, ad hoc use.

If shadow productivity continues to rise quietly inside firms, its gains may remain invisible, unevenly rewarded and difficult to scale. The challenge for businesses and policymakers alike is not simply to accelerate AI adoption, but to ensure that adoption occurs where it matters most: in the systems that shape how organisations work in the form of coordination, collaboration, and decision-making.

In the long run, AI’s impact will depend less on how quickly workers experiment with it, and more on whether firms build the capability to integrate it meaningfully. That is a slower, less dramatic story – but it may ultimately prove the more decisive one.

Kang Yang Trevor Yu is associate professor at Nanyang Business School, and co-director of the NTU Centre for Research and Development in Learning, Nanyang Technological University.

Source: The Straits Times © SPH Media Limited. Permission required for reproduction.

ADV: LawNet Training Workshop Couple sues ERA, agent and law firm for S$731,212 over alleged misrepresentation in ‘99-1’ property deal