Democratised by Default

Why contact centres are the test case for PwC's two-track thesis, and why the track is a decision, not a destiny

PwC's 2026 Global AI Jobs Barometer, drawn from over a billion job advertisements, makes an argument that deserves to be taken seriously. AI is not eliminating jobs at the rate the forecasts promised. It is splitting them.

Some roles are being “Professionalised”. AI absorbs the routine tasks and leaves the human with the harder, more expert work. Others are being Democratised. AI takes on the complex tasks and leaves the human with what remains. In Australia, PwC puts half of advertised jobs in the Democratised category, roughly a quarter in the Professionalised category, and the remaining quarter as barely exposed.

The two tracks have visibly diverged. Professionalised roles are adding new skills at twice the rate of Democratised ones and have seen forty two per cent faster wage growth since 2021. The gap has widened every year since AI adoption accelerated.

Buried in the occupational lists is a line that should stop anyone running a large service operation. Contact centre information clerks sit on the Democratised side.

I want to take that finding seriously, and then argue that it is being read the wrong way round.

What PwC has measured, and what it cannot see

The methodology is sound for what it sets out to do. Classify occupations by the tasks they contain, assess which of those tasks AI can perform, and observe what happens to skill demand, posting volumes and advertised wages over time. The result is the clearest empirical picture we have of AI's labour market effect.

But the unit of analysis is the occupation. That choice carries an assumption with it: that the track is a property of the job.

It isn't. It is a property of the deployment.

Two superannuation funds can run a structurally identical member services role and land on opposite tracks within eighteen months of the same technology decision.

The first deploys conversational AI at the front of the queue with a containment target attached to it. The system is measured on how many contacts it resolves without human involvement. It performs well on the simple, high volume, well formed enquiries, because those are the ones it can complete. What arrives at the human is what the system could not finish: the ambiguous, the emotionally charged, the multi system, the ones where the member has already tried twice and is now angry. The consultant inherits a queue with the pattern stripped out of it and the difficulty left in, and no additional time, authority or support to handle it. Handle time rises. Attrition rises. The role has been “Democratised”, not because AI took the expert work, but because the architecture stripped the recoverable work out and left the residue.

The second fund draws the boundary differently. AI contains a defined tier of routine transactions completely, end to end, with no partial handoffs. Everything outside that tier is routed to a human with full context, pre assembled, along with the authority to resolve it. Supervisor capacity is rebuilt around coaching judgement rather than auditing adherence. The consultant's day is now composed of the cases that require discretion, and the organisation has made the discretion possible to exercise.

Same occupation. Same technology. Opposite tracks.

PwC's data would eventually register the difference, but only after the fact, in the job advertisements, once the divergence had already happened to real people. The classification describes the outcome. It does not describe the mechanism, and the mechanism is where the choice sits.

Why the default runs Democratised

If the track is a choice, the obvious question is why so many organisations are choosing badly. They aren't, mostly.

They are not choosing at all.

The Democratised outcome is what you get when nobody designs the boundary. It is the residue of a business case, and the business case is almost always written in the language of deflection. Contained contacts. Reduced volume to human. Cost per interaction. Every one of those metrics rewards the system for finishing what it can and passing on what it cannot, which is precisely the mechanism that hollows the human role out.

I have written elsewhere about what I call ‘The 85% Trap’: the point at which an automation programme delivers most of its promised containment, declares success and quietly leaves the residual fifteen per cent, the hardest, most consequential, most reputationally exposed fraction of the work, sitting on a human workforce that has been resourced as though the job got easier. The trap is not a technology failure. Every component performs to specification. It is an architecture failure, and it is the most common way an organisation ends up on the Democratised track without ever deciding to.

Contact centres are unusually exposed to this for a reason PwC gestures at without naming. The report argues that the traditional workforce pyramid, high volume at the base, attrition as the management tool, is finished, because AI raises the skill bar on entry level work. Their data supports it: the most AI exposed junior roles are seven times more likely to demand capabilities previously expected of senior staff and entry level roles that have been reshaped this way grew thirty five per cent between 2019 and 2025, while those that were not declined ten per cent.

The contact centre is the purest surviving form of that pyramid. It is an operating model built on high volume junior intake, compressed training, scripted execution and planned turnover. If PwC is right that the model is finished, then contact centres are the most structurally exposed sector in the Australian labour market and the exposure is not primarily about which tasks AI can perform. It is about an operating model that was designed for a distribution of work that no longer exists.

What professionalising the role actually requires

The report's recommendation to leaders is to redesign workflows rather than automate tasks. That is correct, and it is where the analysis stops. It is also where the operational question begins.

Redesigning for the Professionalised track means, concretely:

• Draw the automation boundary at a tier, not a percentage. Define what AI owns completely and resolve it end to end. Partial handoffs are the primary mechanism of role hollowing. A contact that AI half completes is worse for the human than one it never touched.

• Design the exception, not just the happy path. The work that reaches a human is now, by construction, the hardest work in the operation. It needs to be treated as the primary design object rather than as overflow. Context assembly, decision rights, escalation authority and time allowance all have to be set for the residual distribution, not the original one.

• Rebuild supervisor capacity before you need it. Judgement is developed through coaching, and coaching capacity is the first thing consumed when automation compresses headcount. Organisations routinely remove the supervisory layer that the new work profile depends on, then wonder why quality falls.

• Instrument the divergence. If the two tracks are real, an organisation should be able to state which one each of its roles is on, and the direction it is moving. Almost none can. PwC's own advice to leaders is to map the split, and it is the right first move, because you cannot manage a trajectory you cannot see.

• Change what the business case measures. As long as the return is expressed as contacts deflected, the architecture will optimise for Democratisation regardless of what the strategy document says.

Two futures, one decision

PwC frames it as two futures and asks which one you are building toward. The framing is right, but the choice is more immediate than the phrasing suggests.

For most Australian service operations, the decision is not upcoming. It is being made now, implicitly, in containment targets, routing rules and business case metrics written by people who do not think of themselves as making workforce policy.

AI performs to the level of the architecture it is deployed into. So does the workforce standing next to it. The Democratised track is not what AI does to contact centres.

It is what happens when nobody decides where the boundary goes.

Half of Australian jobs are on that track by default.

Default is the operative word.