TalentJam Connections · August 2026 · 7-min read

When the role stops being the unit of account

Pay bands assume a job holds still long enough to be priced. What the trans-Tasman and global evidence actually says about skills-based pay.

An Australian regulator has just rebuilt an award because its classifications stopped describing the work. A New Zealand statute has made role comparison stricter. And almost no employer on either side has the skills data to pay for skills at all.

Editor's note

July handed us two documents that don't obviously belong together. PwC's AI Jobs Barometer reached Asia-Pacific on the 31st carrying a 62% pay premium for AI skills. WTW published a piece nine days earlier arguing that hardly anyone can measure skills well enough to pay for them. Both are right. Most remuneration teams are somewhere between the two, which is what this issue is about.

The 60-second takeaway

Run one query this week. For your technology roles, count how many position descriptions have been rewritten in the past 24 months, then how many of those roles have been re-benchmarked in the same period.

In most organisations the second number is far larger, which means your bands are priced against job descriptions nobody has updated. Fix the descriptions before you redesign the architecture.

Tech talent intelligence

What AI skills pay, and what employers can measure

Sixty-two per cent. That is the average advertised wage premium PwC found for jobs requiring AI skills, across more than a billion job ads in 27 territories, published 15 June. Last year it was 57%. Australia contributed 15.9 million postings, New Zealand 3.7 million.

Consumer markets showed 118%. Technology, media and telecommunications 84%, professional services 67%, financial services 53%. Government and the public sector: 16%. If you set pay for a department, a Crown entity or a state agency, that last figure is the one to put in front of your executive team, roughly a quarter of the all-sector average, on our arithmetic.

Why bands go out of date

PwC measures net skill change (how much the mix of skills demanded inside an occupation shifts over time). In 2025 the most AI-exposed occupations churned theirs at 2.2 times the rate of the least exposed, and that gap is widening: 25% in 2023, 66% in 2024, 116% in 2025.

A pay band assumes the role holds still long enough to be priced. When the contents turn over that fast, the position description anchoring it describes a job that has partly stopped existing, and the first symptom is usually a hiring manager asking for an exception.

Payscale's 2026 survey of 3,413 organisations, fielded from October to December last year, puts a number on the lag: 61% have added AI-related skills to existing role definitions, while 55% have made no compensation change for those skills. Respondents were 80% North American, so read it as direction rather than a trans-Tasman benchmark.

Australia's regulator got there first

On 1 June the Fair Work Commission finalised its rewrite of the SCHADS Award, closing a gender-based undervaluation review begun in June 2024 across five modern awards. The Full Bench found the classification structure no longer fit for purpose: complex, inconsistent between inherited streams, and a reliable source of disputes about who sits where.

The Commission concluded the structures had failed to capture what it called “invisible” skills (communication, organisation, empathy, patience) because they rested on outdated assumptions about the work. Four schedules become one integrated structure, with an interim 15% increase proposed for disability home care workers from 1 October 2026 and the broader structure following on 1 October 2027. Progression moves onto a performance basis.

Strip out the industrial relations context and a tribunal reached the same conclusion as the skills-based pay literature, by a completely different route: the role had stopped describing the work.

New Zealand moved the other way

The Equal Pay Amendment Act 2025, passed under urgency on 6 May last year, lifted the threshold for predominantly female work to 70% sustained over ten years, ranked comparators strictly by closeness to the same employer, and widened the grounds for discontinuing a claim where comparator work is not the same or substantially similar. All 33 live claims were discontinued. A select committee inquiry reported in February 2026, claimants took a complaint to the UN in May, and a three-day High Court hearing concluded in Wellington this month with judgment reserved. Five unions want the Act declared inconsistent with the New Zealand Bill of Rights Act 1990, and the Human Rights Commission has joined as an intervener. A declaration would not change the law, but it would force a select committee process and a parliamentary debate.

Australia's regulator decided role classifications described the work badly and rebuilt them. New Zealand's Parliament decided comparison between roles should be harder. Both answer the same question, how much weight can a role bear, and they point opposite ways. My reading, not either government's.

Almost nobody is actually doing it

Fewer than 30% of organisations use skills effectively across any single core people process, according to WTW, and fewer still have the taxonomy and assessment machinery that paying for skills demands. Most, on WTW's read, are not replacing job-based structures at all. They are recalibrating where skills inform pay decisions, which is a much narrower change than the conference-stage version.

Mercer found the same from the budget side: of just over a thousand organisations surveyed in October 2025, 83% expected to spread salary increase budgets evenly rather than direct money toward scarce skills or market gaps. Mercer's foundations data shows why that is the sensible choice: 55% map skills directly to jobs, up from 47% in 2023, and 38% run a single enterprise-wide skills library, up from 30%. Real movement, but not yet the coverage a pay architecture would need.

AI-skill pay premiums run from 16% to 118% by sector, while 83% of employers still spread pay rises evenly A bar chart of advertised wage premiums for jobs requiring AI skills by sector in 2025: consumer markets 118%, technology, media and telecommunications 84%, professional services 67%, financial services 53%, and government and public sector 16%, against an all-sector average of 62%. Beside it, three figures: 83% of employers plan to spread pay rises evenly rather than target scarce skills; fewer than 30% of organisations use skills effectively in any single core people process; and the skill mix in the most AI-exposed jobs changes 2.2 times faster than in the least exposed. AI-skill pay premiums run from 16% to 118% by sector. 83% of employers still pay evenly. Advertised wage premium for jobs requiring AI skills, by sector, 2025 data What the market pays All-sector average 62% Consumer markets 118% Technology, media & telecoms 84% Professional services 67% Financial services 53% Government & public sector 16% What employers actually do 83% plan to spread pay rises evenly rather than target scarce skills or market gaps <30% use skills effectively across any single core people process 2.2× faster skill-mix change in the most AI-exposed jobs than the least Sources: Sector wage premiums and skill-change rate: PwC 2026 Global AI Jobs Barometer, published 15 Jun 2026 (2025 data; >1bn job ads, 27 territories). Pay-rise distribution: Mercer QuickPulse US Compensation Planning Survey, fielded Oct 2025 (n=1,013). Skills maturity: WTW, 22 Jul 2026. TalentJam
Advertised premiums show what employers offer, not what sitting staff earn. PwC notes the comparison does not control for education, experience or location.

This month's playbook

Four steps before you redesign the architecture

Nothing in the evidence supports a wholesale switch away from job-based pay, and I would be sceptical of anyone selling you that particular bridge this year. What it does support is pulling apart the several different jobs your band does at once. TalentJam is an accredited SFIA partner, so read the framework preference below as a house view, the logic works with any levelling scheme that has explicit responsibility attributes.

Step one

Separate job size from skill scarcity

Most bands express two things at once: how large a job is, and what the market charges for people who can do it. Those used to move together and no longer do. SFIA's seven levels of responsibility give you the first axis cleanly, because the generic attributes (autonomy, influence, complexity, the business skills, and knowledge) describe the shape of the accountability, not which tools are scarce this quarter. Set the level first. Argue about money second.

Step two

Put the premium outside the band

WTW identifies three mechanisms with real traction: a premium for genuinely scarce skills, pay or progression tied to a credential, and an incentive for deploying a skill rather than merely holding it. All three share two useful properties: they are both additive and removable. Bake a scarcity premium into a base band and you have made a permanent decision about a temporary market condition. Attach it separately, with a review date, and you can let it lapse when the scarcity does.

Step three

Settle the evidence question before the money question

Certifiable technical skills get codified and paid; judgement and stakeholder handling get left to manager discretion. WTW calls this the measurability problem: you end up paying precisely for what is easy to measure and vaguely for what creates the value.

Decide in advance what counts as evidence for each skill you mean to price. It could be a certification, an assessed deployment, a peer panel, and write down who signs it off. If you cannot answer that, do not price it yet. The Fair Work Commission's move to performance-based progression under the new SCHADS structure makes the same demand, with a dispute process attached.

Step four

Check who had access in the first place

Where access to the skills you are about to pay a premium for has been uneven historically, a premium does not correct that unevenness. It prices it in and hands it a governance-approved justification.

Australia's March release put the private-sector average total remuneration gender pay gap at 21.1%. Whatever premium scheme you design will land on top of that distribution, not instead of it. Model the effect before you launch, because someone else could very well model it afterwards.

Tools & signals

Three things worth a diary note

  • The C13 phase-out, Australia. The Commission's 2 June annual wage review lifted modern award minimums 4.75% from the first full pay period on or after 1 July, and began removing the C13 classification. C13 rates rose by a further third of the gap to C12, affecting roughly 100,000 of the lowest-paid workers. The Commission has said the phase-out will require the classification descriptions themselves to be reviewed.
  • The gap between two wage measures, New Zealand. Stats NZ released June-quarter figures on 5 August. Wage rates rose 2.0% over the year on the Labour Cost Index, which holds the quality and quantity of labour constant, and 3.1% on the unadjusted index, which does not. That 1.1-point gap is pay moving because people changed roles or progressed, not because the price of the same job rose. Annual CPI was 4.1%, so adjusted wage rates went backwards in real terms. Public sector 1.7%, private sector 2.0%.
  • APS job families, recalibrated. The APSC's workforce planning centre has finished a strategic review of the Job Family Framework, aligning it to the ABS occupation standard released in December 2024. As at 30 June 2025, 12.2% of APS employees (24,135 people)had no job family recorded at all!

A question

If you had to defend one pay decision from the last quarter to a room containing the people it affected, which would you pick, and would the defence rest on the role or on the person? Hit reply.

Next month: internal talent marketplaces across the Tasman: who is running one, and what the deployment data says about whether they work.

Sources