Capability is still managed against roles that are evolving rapidly

Most organisations can tell you what specific roles cost. Far fewer can tell you what skills those roles actually require this year, compared to last year, let alone whether that mix still lines up with the organisation's current priorities. That evolution is something to take seriously on its own terms, without assuming which roles are hit hardest.

PwC's 2026 AI Jobs Barometer measures something called net skill change, how much the skill mix inside an occupation shifts year on year. In 2025, the most AI-exposed occupations churned their skill mix at 2.2 times the rate of the least exposed occupations. That rate of change has been moving quickly: 25% in 2023, 66% in 2024, 116% in 2025.

In AI-exposed roles, a meaningful share of what the job requires can turn over within two years. In most organisations, position descriptions are reviewed far less often than that, which leaves plenty of room for a role to drift from its written description before anyone notices.

Almost nobody gathers the data to catch it.

WTW, the global human capital consulting firm, puts a number on how rare good skills data actually is: fewer than 30% of organisations use skills effectively in even one of their core people processes, not on average across all of them, in any single one.

The concerning bit is that most people-management decisions assume that data exists somewhere. A workforce plan assumes someone knows what current capability looks like. A hiring decision assumes someone has checked whether the gap could be filled from inside first. If fewer than a third of organisations have workable skills data anywhere in their processes, most of those assumptions are running on description, not fact.

Payscale's 2026 survey of over 3,400 organisations shows the same gap from a different angle. 61% have added AI-related skills to existing role definitions, but 55% report no corresponding change to how those roles are managed or resourced.

Why the data itself is hard to get right

WTW calls this the measurability problem. Certifiable technical skills get codified because they're easy to assess. Judgement, stakeholder handling and the less countable parts of a role get left to manager discretion, because they're hard to assess. Organisations end up with precise data on the skills that are easiest to measure, not the skills that create the most value.

Buying a better system to assess skills doesn't help if the system is only capturing what's easy to certify. It will keep producing confident data about a narrow slice of the workforce and nothing usable about the rest, which is often the larger slice.

Australia's Fair Work Commission ran into a version of this from a completely different direction. Rebuilding the SCHADS Award classification structure, it found the existing structure had failed to capture what it called invisible skills, communication, organisation, judgement, because the frameworks had been built around outdated assumptions about the work. A tribunal reached this conclusion through an industrial relations process, not a skills strategy, and arrived at the same place anyway: the framework being used to describe the work had stopped matching the work.

What this means before it becomes a hiring decision

Grow, redeploy or buy decisions only work if you know what you're starting from. Growing someone into a role requires knowing the gap between what they can do and what the role now needs. Redeploying someone requires knowing what they can already do that a different role needs. Buying is the only one of the three that doesn't require knowing your own workforce first, which may be part of why it's the default.

None of the data above says the answer is more frequent job description updates. Updating the description already happens in 61% of the organisations Payscale surveyed, and it didn't change how those roles were resourced. The gap isn't in the paperwork. It's in whether anyone has current, evidenced data on organisational capability.

Where this tends to start

The organisations we see making progress here usually start narrower than a full skills overhaul. They pick a category of roles most exposed to change and work out, role by role, how the skills are evolving, what tasks are being replaced by AI, and what new skills are emerging as a result. That distinction changes what grow, redeploy or buy actually looks like in practice, because a role where the underlying skill is being replaced needs a different response to one where the skill is just being reshaped.

If you had to point to the last systematic check on what a role in your organisation actually requires now, separate from what it says on paper, where would you look?

Sources

PwC, 2026 Global AI Jobs Barometer: Two futures for jobs in an AI era, published 15 June 2026. pwc.com/gx/en/issues/artificial-intelligence/job-barometer/2026/2026-global-ai-jobs-barometer-global-findings.pdf

WTW, Tom Hellier, "Skills-based pay: the narrative, the reality and getting to practical value", 22 July 2026. wtwco.com/en-sg/insights/2026/07/skills-based-pay

Payscale, 2026 Compensation Best Practices Report (17th annual); 3,413 responses gathered October–December 2025. payscale.com/featured-content/cbpr

Fair Work Commission SCHADS Award decision, 1 June 2026; summary via McCullough Robertson, 5 June 2026. mccullough.com.au/2026/06/05/fair-work-commission-schads-award-decision-classifications-wages