Study shows that AI is improving efficiency, but not contributing to growth

Artificial intelligence (AI) is moving from being a technological hype to a more pragmatic phase of being integrated into workflows.

While the private sector has led adoption in previous tech cycles, the global public sector is embracing AI in equal measure to improve efficiency and citizen service delivery.

Government organisations have rolled out chatbots, copilots, and various automation tools across the whole-of-government.

But there remain significant gaps between adoption and realised value, according to a research report titled “What Drives AI Value: Why Modernisation and Workflow Integration Matter” by Harvard Business Review Analytic Services and Appian.

While 59 per cent of organisations have AI in production, the majority are still focused on incremental gains that prioritise efficiency and productivity over top-line growth, the report notes.

The survey of 385 respondents, all AI decision-makers, shows that while businesses are moving beyond early experimentation, they are discovering that simply "having AI" is not enough to drive meaningful business outcomes.

The conclusion is that AI has had a strong impact on bolstering productivity, but less on enabling growth.

Most respondents indicate the most impact of AI on productivity improvements (64 per cent) and operational efficiency (58 per cent), while new revenue streams (30 per cent) and ROI (35 per cent) are among the least likely to have improved.

This points to a significant opportunity for organisations to use AI to deliver broader business outcomes and growth.

Need to move from standalone AI systems

The next phase of AI maturity requires a shift from viewing AI as a standalone tool to embedding it directly into core operations.

Only 18 per cent of respondents report that AI was primarily integrated within workflows, while a larger share (34 per cent) continue to use AI as a standalone tool, with another 34 per cent reporting a mix of both approaches and 12 per cent not yet using AI at all.

Appian has previously noted that when AI operates outside the existing workflows, it lacks the context and connectivity required to drive significant business impact.

Appian’s CEO, Matt Calkins, notes that the true potential of AI “can only be realised when it moves from a standalone tool to an embedded worker that drives revenue”.

This transition is proving complex, as organisations grapple with legacy infrastructure, fragmented data, and the need for fundamental process redesign.

To get there, leaders must prioritise the foundational orchestration and rules-based guardrails required to safely apply AI to high-impact work, Calkins says.

“Instead of using AI to drive productivity, organisations must evolve to focus on business growth. That's where Appian comes in," he adds. 

AI is not yet built into how work gets done, limiting its ability to drive higher-level business outcomes, the Harvard study notes.

Appian adds that organisations must focus on "embedding AI into core processes/workflows," an action that the report shows 62 per cent of the respondents are currently doing, and 71 per cent of these respondents note that this was already yielding notable value.

Legacy systems continue to limit AI’s impact

One of the most striking findings is the correlation between infrastructure modernisation and AI success.

While only 45 per cent of organisations have prioritised modernising legacy infrastructure, this action yields the highest percentage at 76 per cent.

This indicates that many businesses are attempting to build advanced AI capabilities on top of outdated foundations, with nearly seven in 10 respondents agreeing that legacy systems have limited their ability to scale.

This reinforces the need for modernisation and better integration across systems and data.

Siloed or low-quality data (34 per cent), a lack of integration across systems (31 per cent), and a lack of AI talent/skills (30 per cent) are also among the most cited barriers to embedding AI into workflows.

AI agent adoption lags in core operations

Another important finding is that AI agent adoption lags in core operations.

Organisations are actively deploying AI agents in software development (35 per cent), IT operations (31 per cent), marketing and sales (26 per cent), and customer service (25 per cent).

In contrast, agent adoption is more limited in core operational areas such as procurement (nine per cent), manufacturing (10 per cent), and supply chain (11 per cent).

These were areas where processes tend to be more complex and require greater control and consistency.

As organisations look to expand AI into these environments, governance becomes critical.

In relation to governance, an important finding was that most organisations lack the guardrails needed to scale AI agents safely.

Need for rules-based guardrails

Ninety-two per cent of respondents agree that AI agents need rules-based guardrails to operate safely and effectively, but fewer than half (48 per cent) have defined such rules.

As organisations explore agentic AI systems (currently used by 25 per cent of organisations and considered by 62 per cent), the need for clearly defined processes and guardrails will become even more critical.

Without clear guardrails, agents can act unpredictably across systems, increasing the risk of unintended outcomes. 

Realising the full value of AI and achieving sustainable return on investment (ROI) requires rethinking how work is structured and governed.

Organisations are increasingly focused on better defining rules/guardrails that AI must follow (50 per cent), standardising processes/workflows across functions (49 per cent), and increasing cross-functional coordination (47 per cent) to improve the success of AI implementations. 

Long-term, enterprise-wide impact requires a holistic approach that combines technical modernisation with organisational change.

By embedding AI into the flow of work and establishing the necessary guardrails, businesses can move from incremental productivity gains to the transformative potential of agentic AI and beyond.

Those that do not make these foundational investments risk falling short of their ambitions, while those that succeed will be positioned to turn early AI progress into lasting, measurable results.