AI is now commonplace in corporate life. Yet while organisations have invested confidently in the technology, evidence it is improving performance remains harder to find.
New research from PA Consulting captures the gap. The average worker uses AI four times a day, yet one third of employees say its impact still lags the attention it receives. And more activity does not necessarily mean more value: 37% spend as long reviewing AI outputs as they would completing the work manually.
“The assumption is that using more AI always equals more value, therefore we only need to measure usage,” says Charlotte Townsend, people and change expert at PA. “But doing so doesn’t show you where it’s delivering impact.”
The organisations getting more from AI begin with a desired business outcome, not a deployment target. They work backwards from that result, redesigning the enterprise around it: the decisions people make, the roles and capabilities they need, where accountability sits and how capacity released by AI is reinvested.
Start with the destination
Each year, tens of millions of passengers pass through Amsterdam Airport Schiphol, one of the world’s busiest passenger hubs. A delay during aircraft turnaround can quickly spread beyond a single flight, disrupting gates, ground services and passenger connections.
Working with the airport, PA supported the development of Deep Turnaround, which uses AI technology to give operators earlier visibility of delays and emerging disruption. Instead of reacting shortly before departure, teams can intervene sooner, creating smoother passenger experiences. For the airport, it means better use of gates, stands and ground-service resources, while safeguarding operational resilience.
An earlier warning is useful only if someone can act on it. At Schiphol, the prediction changes when operators intervene and what they can prevent.
“You start from where you want to be,” says Rahul Harlalka, global digital lead — transport at PA. “At Schiphol, that meant understanding the impact of the delay before deciding where to intervene. You then work backwards and put the right interventions in place, so that each one contributes to the end goal.”
This means looking beyond the processes that are easiest to automate. Automating a simple task may improve a dashboard without improving the business. Leaders must identify the decisions that constrain performance, then determine what needs to change around them. Doing so means redesigning roles, decision rights, incentives, governance and performance measures, not just introducing a faster tool.
Redesign capability with the work
Redesigning the flow of work also changes how people acquire the knowledge needed to exercise judgement within it. PA’s study of AI use across 2,000 private and public sector employees shows that senior leaders report substantially greater gains in productivity, capability and quality than those in junior, entry-level and non-managerial roles. Senior leaders are nearly twice as likely to say AI enables them to do things they previously could not, and around 1.7 times more likely to report quality improvements.
Experienced leaders can assess an AI output against knowledge built over years. Junior colleagues are still developing that frame of reference, and have historically done this through the routine research, drafting and analysis that AI is beginning to absorb.
If some of these routine tasks junior employees traditionally learn from are automated, organisations must deliberately create new pathways for employees to develop expertise through mentoring, observation, challenge and practice, rather than assuming development will happen on its own.
Senior leaders must also make their own standards more visible. “The best leaders make time to explain the difference between okay, good, and brilliant,” says Townsend. “If people have fewer opportunities to build judgement by doing the work themselves, you have to create other opportunities for them to develop it.”
Dependence may be the other side of familiarity. One fifth of employees feel overly dependent on AI, rising to 28% among frequent users, while heavy AI users are also more than 2.5 times as likely as light users to favour an AI answer over their own judgement.
This makes ongoing training as important as initial adoption. AI capabilities and limitations continue to evolve, and extensive use can change how employees perceive both the technology and their own judgment. Without regular refreshers, employees risk losing the ability to distinguish a genuinely useful output from one that merely sounds convincing.
The more capable the technology becomes, the more important it is that employees can assess its output, identify its limitations and explain why a particular decision was made.
Make responsibility visible
AI produces answer, not responsibility for how that answer is used. Although two-thirds of workers believe responsibility remains with the output owner, 34% feel less responsible for quality when AI is involved — and more than a quarter feel uncomfortable signing off work shaped by AI.
Leaders therefore need to define where human judgement is required, who owns the decision and what evidence is needed before work is approved. A faster decision is not necessarily a defensible one. Employees should be able to interrogate AI-assisted work and explain how the result was validated.
As Townsend puts it: “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.”
Decide what saved time should buy
The same shift is needed in measurement. Pilots launched, licences activated and hours saved are easy to count but say little about whether work has improved.
“Saving time is not the same as creating value. Organisations must make sure that any time saved is reinvested into high-value outcomes, including upskilling,”
Warwick Goodall, Global Roads and Mobility Lead, PA Consulting
Reinvestment should be designed into change from the outset. If leaders wait until time has been saved to decide what to do with it, efficiency may never become enterprise value. This principle has been applied in practice during large-scale Microsoft 365 Copilot deployments. Working with PA, a leading European bank established a value measurement framework to support the expansion of users. Built around individual productivity and business results, the framework was designed to measure not only whether employees worked faster, but what value the organisation generated from the capacity that was freed up.
Adoption is only the first signal. Leaders also need evidence that behaviour has changed, outcomes have improved and the capacity released by AI has been put to better use. Otherwise, a successful rollout can coexist with stagnant performance.
Together, these signals show where progress is stalling. If adoption rises but outcomes do not, the obstacle may lie in the operating process, workforce capability, decision rights or the use of saved time.
Enterprise AI will not be won by the organisation with the most licences or the busiest users. The advantage will belong to those that redesign the enterprise around the outcomes they want: turning intelligence into better decisions, saved time into additional value, and widespread adoption into stronger performance.
Explore the Intelligent Enterprise.
AI is now commonplace in corporate life. Yet while organisations have invested confidently in the technology, evidence it is improving performance remains harder to find.
New research from PA Consulting captures the gap. The average worker uses AI four times a day, yet one third of employees say its impact still lags the attention it receives. And more activity does not necessarily mean more value: 37% spend as long reviewing AI outputs as they would completing the work manually.
“The assumption is that using more AI always equals more value, therefore we only need to measure usage,” says Charlotte Townsend, people and change expert at PA. “But doing so doesn't show you where it’s delivering impact.”
