Redesigning the GBS workforce for an agentic future.

The workforce equation is changing

The conversation about AI in enterprise operations has moved beyond efficiency gains. We are now confronting a structural question: as AI agents take on end-to-end process execution and escalate only exceptions to humans, what does this mean for how we design our organisations, where we source talent, and what capabilities we need to build and sustain?

This is not a theoretical exercise. PwC's Global Business Services (GBS) Study 2025 finds that 96% of organisations cite increased efficiency as the primary driver for AI and RPA adoption, while 85% target cost reduction and 84% prioritise quality improvement. The implication is clear: AI is being deployed first as an efficiency and quality lever—creating a follow-on operating model challenge as work shifts from routine execution to exception handling, oversight, and continuous improvement.

Key data points from PwC’s GBS Study 2025

• AI expectations: 96% cite increased efficiency; 85% cite cost reduction; 84% cite quality improvement.

• Technology maturity: AI (Generative and Agentic) is still early-stage for most (only 14% report widespread use; 45% have not considered it yet or are only in ideation).

• Operating model direction: 52% of organisations run GBS with multiple global sites performing end-to-end processes and expert functions (up from 39% in 2023). 

• Centres of excellence (COEs) are mainstream: approximately 75% of respondents have invested in or plan to invest in centres of excellence.

For GBS leaders in Australia, where talent scarcity, high labour costs, and regulatory complexity converge, the imperative is clear. The question is no longer whether AI will reshape your workforce. It is whether you are designing deliberately for that future or allowing it to happen to you.

What workforce redesign through AI actually means 

When AI absorbs transactional processing layers, the impact extends far beyond headcount. It rewires the entire organisational architecture:

• Organisation design: As automation and AI scale, operating models tend to move along a continuum—from a largely human “execution engine”, to “digital augmentation”, and (longer term) towards an “expansive digital workforce” where agentic automation can deliver more of the transactional processing. This typically widens spans of control, compresses layers, and shifts management attention from throughput to overall system performance, including the outcomes delivered by AI-driven processes.

• Role architecture: As transactional work is automated, human roles shift towards supervision, exception handling, process ownership, and improvement—supported by stronger AI literacy and clearer escalation pathways. Practically, that means redesigning job families around (a) exception mastery, (b) human-AI interaction and prompt/workflow design, and (c) governance disciplines such as controls, auditability, and data stewardship.

• Talent profiles: Skill profiles are shifting towards a combination of advanced technical capabilities and applied AI literacy across the broader workforce. In the study, 76% of organisations prioritise building data analytics skills, 66% process automation/RPA, 61% project and change management, and 54% (Gen)AI expertise. Notably, only 29% cite data security and risk management as a key skill to build—an important gap given increasing centralisation of data and end-to-end processes. 

• Organisational wiring: Reporting lines, governance forums, and escalation pathways must be reconfigured around human-AI collaboration rather than human-only hierarchies. Performance systems need realignment - measuring outcomes, model stewardship, and improvement contribution rather than task throughput.

These shifts also require a reset of how performance is measured and communicated. The study shows dashboards are the dominant reporting method (93%), complemented by process audits/reviews (66%), internal customer surveys (64%), and benchmarking (53%). Yet measuring value remains difficult—driven by poor data availability/quality (45%), unstandardised KPIs (43%), and challenges linking KPIs to corporate goals (42%).

Two distinct capability sets: Build AI and run AI 

GBS leaders must invest across the build-and-run lifecycle of AI-enabled services. The study’s digital enablement chapter reinforces that value typically comes from combining capabilities (e.g., workflow, RPA, data visualisation, advanced analytics, and increasingly GenAI and agentic AI) and then scaling them beyond pilots. In that context, it helps to distinguish between capabilities needed to build AI solutions and those needed to run them safely and effectively in production.

Build AI — the talent required to design, develop, and deploy AI solutions:

• AI engineers, data scientists, and solution architects

• Prompt engineers and workflow designers who translate business problems into agentic configurations

• Integration specialists who connect AI agents to ERP, CRM, and operational systems

• Governance designers who establish guardrails, explainability standards, and bias controls

Run AI — the talent required to operate, monitor, and continuously improve AI in production:

• AI operations analysts who monitor agent performance, confidence scores, and drift

• Exception specialists who handle the cases that do not flow straight through and turn them into learning signals

• Model stewards responsible for data quality, retraining triggers, and regulatory compliance

• Process product owners who manage the AI-enabled service as a product, iterating based on performance data

Across the report, a consistent theme is that scaling AI requires more than experimentation: it requires governance, auditability, and clear human oversight—especially as organisations move from automation and GenAI pilots into more autonomous (agentic) patterns to ensure AI-enabled processes operate reliably at scale. Practically, that means defining action limits and approval thresholds, maintaining audit logs, and building a workforce that can interpret outputs, manage exceptions, and continuously improve services based on performance data.

The orchestration model: Humans in and on the loop 

As organisations explore more autonomous patterns, the report distinguishes two complementary operating approaches for agents:

• Human-in-the-loop: Humans approve or intervene at defined decision points before the agent proceeds. Appropriate for high-risk, high-judgement activities where regulatory, ethical, or financial consequences require human accountability.

• Human-on-the-loop: Agents execute autonomously while humans monitor performance, review outputs on a sampling or exception basis, and intervene when thresholds are breached. Appropriate for scaled, lower-risk processes where speed and volume demand autonomy.

The design task is to choose the right level of autonomy per process and then operationalise the guardrails. The report highlights mechanisms such as staged autonomy (observe → propose → approve → limited execute), clear escalation paths, action limits and approval thresholds, and audit logs. For higher-risk contexts, it also points to the need for simulation environments and “kill switches” to safely test and control agent behaviour.

The global capability centre connection: Where capability gets built at scale

This workforce transformation does not happen in isolation. For Australian organisations, the Global Capability Centre (GCC) model provides the structural answer to the question of where you build these new capabilities.

GCCs are presented as an evolution of GBS: outcomes-centric entities that combine traditional shared services scope with COEs and 'frontier capabilities' (e.g., digital, analytics, engineering and R&D), acting as an ‘equal strategic partner’ of the operating model alongside corporate functions and business units. With around three-quarters of organisations investing in or planning COEs, GCCs provide a repeatable way to concentrate specialised talent, methods, and reusable assets (data, prompts, workflows, controls) and then scale them enterprise-wide.

For Australian organisations, the choice is less about “offshoring for cost” and more about designing a location portfolio that balances capability, scale, resilience, and time zone coverage. The study’s location strategy chapter emphasises blending mature hubs with emerging spokes and being explicit about each location’s role (scale delivery, specialised expertise, time zone coverage, or lower-value activities). In practice, this supports a model where you centralise build/run capability (e.g., digital, analytics, automation operations) where talent ecosystems are deepest—while retaining business ownership, governance, and stakeholder leadership close to the enterprise.

The leadership imperative

The leaders needed for AI-enabled operations are different from the past: they must combine product thinking with iterative change, domain expertise with data literacy, and the ability to translate between technology teams, risk/controls, and the business. They also need to modernise management routines—moving from volume-based reporting to outcomes and value tracking (for example, through a cross-functional “value office” that baselines performance and validates benefits as automation scales).

Overall, the study reinforces a direction of travel: GBS is shifting from transactional delivery towards a broader transformation role—powered by process excellence, COEs, data/analytics, and (increasingly) AI-enabled services. It also challenges common assumptions: fully virtual delivery has not become the dominant model; instead, hybrid working is now the standard (53% in 2025, expected to rise to 56% by 2027).

For GBS leaders, the challenge extends beyond automation. AI is not simply changing how work is performed—it is changing what work exists, which skills matter, and how organisations create value. As transactional activity becomes increasingly automated, competitive advantage will depend less on execution capacity and more on an organisation's ability to orchestrate human expertise, digital capabilities and AI-driven services as an integrated workforce. The most successful organisations will not be those that deploy the most AI, but those that most effectively redesign their operating models, workforce and governance around it. Those that get this right—across roles, measures, governance, location portfolio, and talent—will be best positioned to capture both efficiency gains and long-term strategic value.

The work is clear. The time to start is now.