The new rules of execution: redesigning work to unlock AI value

AI has rapidly moved from experimentation to expectation, with a study from the US National Bureau of Economic Research showing that over two thirds (69%) of businesses are using AI in their operations.

Yet while organisations have been confidently investing in AI, far fewer have worked out how to translate deployment into meaningful performance improvement. Success is often seen as a rudimentary numbers game: pilots launched, licenses deployed, or hours saved. Yet these metrics reveal little about whether true value is being created, or whether ways of working, behaviours and decision-making are changing to deliver better outcomes.

Scratch beneath the surface of ‘hours saved’ and it quickly becomes clear why it’s misleading. Research from PA Consulting shows that 37% of employees spend as long reviewing AI output as they would completing the work manually.

This is symptomatic of a wider challenge. One-third of workers in PA’s survey say AI’s impact still lags the attention it receives. “The assumption is that using more AI always equals more value, therefore we only need to measure usage,” says Dean Collins, people and technology adoption expert at PA. “But doing so doesn’t show you where it’s delivering impact.”

From solutions to systems

The organisations creating the greatest value from AI are taking a different approach. Rather than treating AI as a technology deployment exercise, they are redesigning how work gets done, how decisions are made and how performance is measured across the enterprise.

For example, each year, tens of millions passengers pass through Amsterdam’s Schiphol Airport, one of the world’s busiest passenger hubs and Europe’s third-largest cargo airport.

Working with the airport, PA supported the development of Deep Turnaround, which gives operators earlier warning of delays during aircraft turnaround. Teams can then intervene before departure, helping to prevent missed connections and disruptive gate changes while making better use of gates, stands and ground-service resources. Earlier intervention also helps absorb late arrivals before they ripple through the network.

Rahul Harlalka, global digital lead – transport at PA, says Deep Turnaround creates value because its predictions are embedded in operational decisions, rather than treated as the output of a standalone tool. Organisations seeking similar gains need to shift their focus away from hype around a single tool or model, which simply moves problems elsewhere, and towards better capability and outcomes.

“You start from where you want to be,” says Harlalka. “Then work back and put the interventions in place from the foundations onwards. This way, you can ensure that any actions happen in aid of an end goal.”

Too often, organisations focus on the processes that are easiest to automate rather than those that create the most value. As Warwick Goodall, global roads and mobility lead at PA, puts it: “Yes, you can automate a summary of yesterday’s emails, or isolated processes, but is that really delivering value? 

This philosophy — where you start with the end goal and work backwards to the right tool — shaped PA’s work with National Highways. As the organisation invests in digital roads, intelligent infrastructure and data-driven operations, its objective was not simply to deploy new technologies, but to ensure the strategic road network remained secure, resilient and trusted.

Throughout, it was crucial to maintain the confidence of both road users and the teams responsible for operating the network. If controllers don’t trust the outputs they receive enough to act, or road users don’t trust the network, the benefits of digital innovation quickly unravel. 

PA helped National Highways establish its architecture, governance and operating frameworks before introducing AI-enabled controls. Embedding security and resilience into day-to-day operations strengthened its ability to anticipate emerging risks and support safer, more reliable journeys.

“What we’re seeing from the intelligent enterprises leading on AI,” says Harlalka, “is a focus on how the entire system helps create value: orchestration, guardrails, monitoring, measurement, and so on. And building things in a way that creates repeatable, scalable delivery rather than ad-hoc, single use cases.”

The same shift is under way in clinical trials. AI can help researchers identify drug targets and automate tasks such as protocol development, patient recruitment and data review. Yet isolated automation does not prevent delays when operating models cannot cope with the volume of data.

Building on these advances, PA is exploring the concept of self-optimising clinical trials: studies that use AI to dynamically improve protocol execution, patient engagement, site performance and operational decision-making in near real time.

Jenna Phillips, a life sciences expert at PA, says, “Before launch, digital simulations identify likely points of friction. During execution, AI continuously evaluates recruitment, site performance, data quality, protocol adherence and patient engagement, helping teams intervene before problems become delays.

“This isn’t a future-state concept,” she continues. “Pilots are already demonstrating what’s possible, and the opportunity now is to scale isolated successes across studies and organisations.”

Building judgement into the system

If AI success comes from systems and capability, leaders must own the outcome, set a clear vision and build the organisation’s ability to deliver it.

“There’s a surprising number of leaders who throw tools at their people without a clear idea of the outcomes they wish to achieve, without clear lines of accountability, and without clear guardrails and structures,” says Harlalka. “What people need is a compelling leadership vision about how the organisation is going to change as a result.”

At the UK Department for Transport, senior leaders presented AI as a business and workforce opportunity that empowered civil servants, rather than an IT programme imposed from elsewhere. That gave employees clearer signals about where AI should be trusted, challenged and owned.

This kind of leadership is necessary given PA’s research found that one fifth (20%) of workers already feel overly dependent on AI, while over one quarter of frequent users (28%) trust AI outputs more than their own judgement. 

Collins adds: “The more AI is embedded across organisations, the more valuable human judgement becomes. Evidence has shown that teams with high psychological safety — where team members feel safe to challenge each other — achieve better results. Now people also need to feel confident assessing and challenging AI-generated outputs, whether they’ve created them themselves or received them from elsewhere in the organisation.”

For critical national infrastructure, that confidence is vital. Harlalka says: “Decision quality is the whole game. Split-second calls, dynamic timetable management, incident response and recovery — these are environments where a better decision, made faster, is the entire point of the investment.”

Time saved is just the start

Organisations have spent years measuring output and efficiency. AI’s next wave of value will depend on what people do with the capacity it creates.

That distinction between productivity and performance is summed up neatly by Goodall: “Saving time is not the same as creating value. Organisations need to make sure that any time saved is reinvested into high-value outcomes. That could be upskilling, or time used for higher-value work.”

Harlalka adds, “If you start with the hours saved and ask what to do with them afterwards, you’ve already lost the thread back to why AI was used at all.” 

At a major European bank, a large-scale rollout of Microsoft 365 Copilot had to deliver more than adoption. As one digital workplace leader explained: “AI only matters if it delivers value. We were not going to force-fit technology; M365 Copilot had to prove quantifiable gains for our employees and our business.”

PA created a Value Measurement Framework spanning individual productivity and business results. One result was a streamlined credit-application process with the potential to benefit around 150 lenders.

Goodall says internal surveys must also test whether AI has changed the work. “Too many surveys ask people if they have the right tools,” he says. “But AI adoption isn’t equivalent to having the right office chair. We need to find out if ways of working have fundamentally changed.”

“Here, success is not a single metric but a story across four: adoption, behaviour, outcomes and reinvested capacity. They need to be read together, because organisations can optimise for individual measures while nothing that matters actually moves.” 

Read together, those signals show whether AI is changing execution and where leaders need to intervene through coaching, workflow redesign or clearer governance.

Where time is saved, leaders need to direct it towards the next desired outcome: time saved as fuel for what comes next, not as a bonus that nobody’s tracking. Harlalka adds: “If AI saves an hour, that hour only counts once someone’s decided what it buys — more resilience, faster recovery, better customer outcomes — and can prove it did that.”

Asking the right question

Today’s intelligent enterprises are no longer asking whether people are using AI. They’re asking whether behaviour, decisions and outcomes have fundamentally changed. 

In turn, they are redesigning how work flows across people, technology and operations, then reinvesting the capacity AI creates in better decisions, stronger resilience and measurable performance. That is how value is created.

AI has been described as the great leveller of our time. Yet new research from PA Consulting suggests that organisations cannot assume those benefits will emerge automatically.

With the average worker using AI four times a day, providing access to the technology is proving much easier than ensuring that value is realised consistently across an organisation. 

PA’s study of 2,000 private and public sector workers across multiple industries reveals a critical challenge: many organisations are introducing AI without fully redesigning how employees use it, how work is assessed or where decisions sit. Often, employees are left to determine these outcomes themselves.

Too often, AI is deployed without a clear strategy for what it is intended to achieve or how it should improve work. This lack of direction can lead to uneven benefit distribution and increase the risk of over-reliance on the technology.

To realise value consistently across the workforce, leaders need to set clear outcomes for AI and rethink how expertise is developed, how judgement is exercised and how responsibility is assigned.

Leaders must set the direction and remove the obstacles

PA’s research reveals a clear difference between the AI experiences of those in entry-level, junior, and non-managerial roles (individual contributors) with those in senior leadership roles. 

Even allowing for the fact that senior leaders report higher usage, they report substantially greater gains in productivity, capability and quality than individual contributors: 70% of senior leaders say AI enables them to do things they previously could not, compared with 38% of individual contributors. Similarly, 72% of senior leaders report quality improvements from AI, compared with 43% of individual contributors. 

Rather than acting as an equalising force, AI appears to benefit employees who already have the experience, confidence and professional judgement to recognise what a good output looks like and challenge it when necessary. 

Part of the reason for this may be that AI is changing the very activities through which expertise is traditionally acquired. Dean Collins, people and technology adoption expert at PA, notes that early in many careers, people learn by being close to work that AI threatens to replace. He says: “Junior employees are missing out on opportunities to build expertise, judgement and confidence through observation, trial, feedback and repetition. 

“With AI doing more of the heavy lifting, traditional on-the-job learning is compressed, if not entirely bypassed,” he continues. “As a result, this cohort of workers may see slower skill growth and reduced long-term adaptability. Organisations need to think carefully about the learning opportunities they will provide in their place.”

For example, if AI removes apprenticeship tasks, organisations must deliberately design new ways for employees to develop expertise through mentoring, structured observation, challenge and practice. Otherwise, short-term efficiency may come at the expense of long-term capability.

Capability or dependence?

As AI usage increases, so does trust in it. PA’s research found a net trust of 92% among people who have used AI for four years or more, compared with 24% among newer users.

Yet the data also casts a shadow. One-fifth of employees feel overly dependent on AI, rising to 28% among frequent AI users. Heavy users are also more than 2.5 times as likely as light users to favour an AI answer over their own judgement, even when the two disagree.

Employees need to understand where AI systems can fall short and how to distinguish an adequate output from an excellent one. As AI handles more routine work, leaders must make their own judgement more visible.

“The best leaders make time to explain the difference between okay, good and brilliant,” says Collins. “If people have fewer opportunities to build judgement by doing the work themselves, you have to create other opportunities for them to develop it.”

One way to drive change, Collins adds, is to “set AI to be less flattering and more challenging. Leaders in particular should be doing that for their systems, because they are already the least likely to be challenged.”

While it’s widely accepted that AI systems require a human-in-the-loop, Collins proposes a practical method for defining the extent. He suggests two criteria: value and risk. Low-value, low-risk work may be more fully automated. Human involvement should increase with the consequences of a decision.

Building on this, Warwick Goodall, global roads and mobility lead at PA, recommends mapping role-specific workflows, identifying important decisions and defining how people and AI should work together. He adds: “The risk of dependency means that human expertise always remains essential – either in the workflow mapping, or in the live review cycle.”

Redesigning responsibility

As systems become more autonomous, AI also complicates accountability. PA’s survey found that two-thirds of workers believe accountability remains with the output owner whatever AI’s involvement. But 34% said they feel less responsible for quality when AI is involved.

This fog of responsibility has consequences: just over a quarter of workers (26%) feel uncomfortable signing off work that has been shaped by AI.

For Collins, leaders need to look for signs of interrogation and explanation, not just increased outputs. Employees should be able to explain AI-assisted decisions. 

He adds: “A team using AI selectively but making higher-quality decisions may outperform one with heavier usage and weaker judgement. The goal isn’t ubiquitous use. It’s capable and accountable use.”

As organisations embed AI more deeply into critical processes, accountability increasingly extends beyond individual outputs to questions of governance, oversight and operational management. Leaders need clarity not only on who owns decisions, but also on the level of dependence placed on external AI systems and providers.

Ensuring that work works

The organisations that realise the greatest value from AI will be those that treat it as workforce transformation, not simply technology deployment. They’ll be clear about the outcome they want AI to achieve and create the conditions for success through leadership, learning, capability, accountability and trust.

The firms that get it right may not have the most AI users. But they will have those who have learned to work differently and better.