Human workers are not built for the kind of work GBS performs at scale. Leaders should not view AI as advanced automation but apply it to the redesign of enterprise services as a growth platform.
Financial transactions, data reconciliation, compliance monitoring, case processing, exception handling—these require absolute consistency across millions of instances, the ability to hold enormous quantities of data in mind, and accurate decisions at speed, without error, day after day. Other activities are more routine but equally unforgiving: repetitive tasks that cannot fail in accuracy, relevance, or timing, sustained for years without variation. Humans are not designed to sustain either indefinitely. GBS, as built, has depended on them doing exactly that.
This matters beyond the operational. The GBS value proposition was never simply cost reduction: it was consolidating transactional work to free capacity to identify and pursue new sources of enterprise value. But classic GBS delivered that promise only partially. Without meaningful automation, it replaced dispersed human transactional effort with centralised human transactional effort. Those intended to pursue higher-value work found themselves managing the operation instead, and in time seeking work elsewhere. Governance stiffened to compensate, consuming the bandwidth that was supposed to drive transformation. The result was a model that promised to liberate human capacity and instead trapped it.
The same logic that reveals the constraint reveals the opportunity. The transactional, data-intensive work that has held GBS back is precisely what agentic AI is designed to do: at scale, without fatigue, with continuous improvement. When machines absorb the work for which humans were never suited, human capacity is not displaced: it is released for the judgment-intensive, relationship-driven work that generates new value. As set out in our other articles (From services to spine, The intelligence mandate for GBS and The Intelligent Enterprise Spine), this is the transition from GBS as managed cost centre to GBS as Intelligent Enterprise Spine: the platform through which AI capability and human ingenuity combine to drive growth.
This article maps how that transition happens, what the agentic-led GBS model looks like in practice, and why most organisations are not yet ready to make it.
When machines absorb the work that humans were never suited for, human capacity is not displaced: it is released.
Classic GBS: a climb that most organisations have not made
GBS development has been well-mapped for decades. The journey runs from fragmented, business-unit-based operations to a fully integrated enterprise services platform across five levels, each requiring greater structural commitment, technology investment, and organisational discipline than the last.
| Level | What it represents | Where organisations get stuck |
|---|---|---|
| (1) Fragmented / decentralised | Functions run independently across business units. No shared processes, no economies of scale. | High cost of complexity; duplicated effort; no baseline from which to improve. |
| (2) Centralised shared services | Common processes pooled into a shared centre. Labour arbitrage the primary value driver. | Cost savings plateau; service quality inconsistent; change resistance from business units. |
| (3) Global Business Services | Multi-function, multi-geography model with genuine process ownership and enterprise scope. | Governance complexity; difficult to maintain strategic alignment as scope expands. |
| (4) Enterprise services platform | Technology-enabled, analytically capable; strategic partner to business leadership. | Requires sustained executive sponsorship and significant investment. Rarely achieved. |
| (5) Intelligent enterprise services ecosystem | Autonomous, AI-native operations. Continuous self-improvement. Intelligent Enterprise Spine. | Has been theoretical for most organisations. Until now. |
Classic GBS development curve. Most organisations operate at levels 1–2, few reach level 3, and level 4 remains rare.
Most organisations that set out on this journey never complete it. Many spent the 2000s and 2010s building basic shared services and are still there. Some are still starting. The reasons are consistent across industries and geographies, and they begin with the business case.
The initial consolidation case is made on labour arbitrage: reduce headcount, move work to lower-cost locations and harvest the savings. This is the metric that leadership approves, and the one against which GBS leaders are measured. It rewards cost-per-transaction, not capability development. A low-cost, mediocre-quality operation can appear successful for years. Moving up the curve requires investment, disruption, and a longer time horizon—none of which the original business case supports.
The standard playbook compounds the problem. Organisations consolidate first and plan to standardise later. In practice, later rarely arrives. Moving a fragmented, exception-ridden process into a shared services centre does not discipline it: the complexity travels. Standardisation requires the authority to say no to business unit variation, and GBS organisations are rarely granted that authority. Exceptions accumulate, and the fragmented model rebuilds itself under one roof.
Technology reinforces the ceiling. GBS inherits legacy systems from which the front office has already moved on. The platform investment that could enable genuine standardisation is always two years away. Without it, data quality, process consistency and automation that advanced GBS requires cannot be built.
The result is a model that plateaus—not because of poor management, but because incentives, governance and technology never align with the ambition. This is the landscape within which agentic AI now emerges.
Agentic AI as the curve compressor
Agentic AI is not a faster version of robotic process automation. It represents a different operating capability. RPA accelerated the execution of pre-defined rules within a fixed process. Agentic AI introduces something categorically different: systems that can reason across tasks, take sequential actions in response to context, handle ambiguity, coordinate with other agents, and escalate intelligently when they reach the limits of their authority. The practical consequence for a GBS organisation is that the activities that previously required either expensive labour or years of process standardisation can be initiated, orchestrated, and completed autonomously.
Analyst forecasts reinforce the scale of the shift underway. Gartner predicts that by 2026, more than 40 per cent of enterprise applications will include task-specific AI agents, up from less than five per cent in 2025.1 As these capabilities mature, agent-based orchestration of operational work is expected to become a standard feature of enterprise platforms.
This does two things to the GBS maturity curve. First, it compresses the time required to traverse the early stages. An organisation that would previously have spent five years standardising processes before it could contemplate automation can now deploy targeted agents into messy, partially standardised processes and use the deployment itself as the standardisation mechanism. The process improves through use. Second, agentic AI extends the value potential: a genuinely autonomous GBS platform and, beyond it, a cognitive enterprise ecosystem, become achievable endpoints rather than theoretical ones.
The agentic-led GBS development model that emerges from this shift looks different from the classic curve in important ways. The early stages focus on deploying agents into specific high-volume workflows, not on building the infrastructure first. Value is demonstrated quickly, in weeks rather than years, sustaining organisational confidence and funding subsequent expansion. Each stage builds on the deployed capability of the previous one, rather than on a blueprint designed in advance and then executed.
| Level | What it represents | Where organisations get stuck |
|---|---|---|
| (1) AI-enabled baseline | Targeted agents handle high-volume, rule-adjacent tasks: invoice matching, query routing, payroll exceptions, case triage. Manual effort reduces 30–50%. Accuracy and throughput improve immediately. | Identifying the right initial workflows; data quality in source systems; building the organisational confidence to expand beyond the first deployment. |
| (2) AI-orchestrated platform | Connected agents work across functions and systems. Multi-step workflows orchestrated end-to-end. Handoffs between teams reduced; cycle times cut by 2–5x. Governance framework established. | Integration complexity across legacy systems; defining accountability when agents span functions; resistance from teams whose workflows are being redesigned. |
| (3) Autonomous GBS | Services become largely self-managing. Agents handle exceptions, escalate outliers, and re-route work without human initiation. Human oversight focuses on policy and performance, not transaction processing. | Building organisational trust in autonomous decision-making; redefining human roles before resistance builds; satisfying audit and compliance requirements for AI-driven processes. |
| (4) Intelligent enterprise ecosystem | GBS operates as the enterprise AI spine: governing data standards, orchestrating cross-functional agent networks, and managing AI risk enterprise-wide. | Securing the cross-functional authority that enterprise data governance requires; building AI risk and oversight frameworks at scale; sustaining executive sponsorship beyond the initial deployment case. |
| (5) Cognitive enterprise | The enterprise adapts and improves in real time through operational intelligence. Services are outcome-priced and continuously optimised. The services-as-software model in practice. | Reached by fewer than one in ten organisations globally. The frontier is still being defined. |
Agentic-led GBS development model. The critical insight: agentic AI compresses the classic curve and extends beyond its ceiling.
The roadmap emerges through well-managed AI deployment: each agent deployed reveals where the next opportunity lies.
The deployment evidence
The case for agentic GBS is not theoretical. Across financial services, logistics, professional services and operations, organisations that have deployed agent-based automation at scale are reporting results that exceed those delivered by traditional automation programmes. The benchmarks are significant enough to change the strategic conversation.
- •JPMorgan Chase's Legal Agentic Workflows (LAW) system, a network of AI agents managing legal document review and compliance tasks, reports 92.9 per cent task accuracy and saves an estimated 360,000 working hours annually across the legal function alone.2
- •DHL deployed HappyRobot AI agents across its freight operations in late 2025, autonomously handling hundreds of thousands of emails and millions of voice minutes, improving carrier communication speed and reducing the manual burden on operations teams.3
- •Verizon reports that generative AI can predict the reason for a customer call roughly 80 per cent of the time, allowing calls to be routed to the most appropriate agent before the interaction begins.4
- •Goldman Sachs deployed AI systems that have saved analysts an estimated 30 to 50 per cent of time previously spent on research synthesis and document preparation.5 Invensis, the business process outsourcing firm, reported a 38 per cent productivity improvement and approximately a 40 per cent cost reduction following agentic implementation in core service delivery workflows.6
Early deployments of agentic systems across enterprise service operations have reported significant improvements in productivity, cycle time, and accuracy. In several cases, these outcomes are observed in production environments rather than controlled pilots. While results vary by process and organisational context, the evidence is building: when AI agents orchestrate high-volume operational work, substantial gains in speed, consistency and transparency follow.
At the same time, enterprise adoption of agentic AI is accelerating sharply at the infrastructure level. Gartner forecasts that 40 per cent of enterprise applications will feature task-specific AI agents by 2026, up from fewer than five per cent in 2025.1 The Hackett Group's 2025 GBS AI Adoption Survey found that 42 per cent of GBS organisations had trialled generative AI in 2024, and 63 per cent of those recorded measurable productivity gains.7 The direction of travel is clear: the question is not whether agentic AI will reshape GBS but how quickly and on whose terms.
The real blockers: debt that technology cannot pay
Despite the evidence of deployment, meaningful enterprise-wide agentic AI remains the exception. Only 7 per cent of Global 2000 enterprises have achieved enterprise-wide agent deployment at scale, according to HFS Research.8 Deloitte's 2026 technology trends analysis found that just 11 per cent of organisations are actively deploying agentic AI in production; 42 per cent are still developing a strategy.9 IDC data indicates that 88 per cent of AI pilots fail to reach production: for every 33 proofs of concept initiated, four reach production deployment.10
The blockers are not the AI systems. They are what HFS Research describes as the four enterprise debts: accumulated deficits in process, data, talent, and technology that compound over time and prevent organisations from deploying AI at scale, regardless of how good the technology itself is. HFS estimates the global cost of these debts at approximately $10 trillion, with process debt the largest single category:
- •Process debt exists where workflows are fragmented, undocumented, exception-ridden or inconsistently executed. Agents require legible processes to orchestrate. An organisation that has never properly documented how a given process works, or where dozens of undocumented local variations exist, will find that agents cannot be deployed safely into that environment. The standardisation that GBS was supposed to deliver, but often did not complete, becomes a prerequisite for agentic deployment.
- •Data debt is the consequence of years of fragmented system landscapes, inconsistent data governance, and poor master data management. Agents depend on clean, accessible, trustworthy data. Where data sits in siloed systems, is of poor quality, or lacks the access controls required for AI to use it safely, deployment stalls at the first stage.
- •Talent debt accumulates where workforce skills have not kept pace with technological change. Deploying agents into a GBS operation where the workforce lacks the AI literacy to supervise, correct and improve them creates both operational risk and cultural resistance. HFS research shows that 60 per cent of workforces approach agentic AI with uncertainty or concern rather than enthusiasm.2 Managing that transition is as important as the deployment itself.
- •Technology debt reflects the accumulated legacy of systems that were not designed for the integration demands of an agentic environment. Agents need to connect to systems, extract information, trigger actions and pass outputs to other agents. Where the technology landscape is fragmented, proprietary and poorly integrated, the plumbing required for agentic orchestration becomes expensive and slow.
HFS Research's enterprise survey data is illuminating on the relative weight of these debts. When asked to identify the single most important constraint on achieving organisational goals, 35 per cent of senior leaders cited process inefficiencies, 19 per cent data limitations, 17 per cent people challenges, and only 16 per cent technology constraints. The biggest blocker to AI deployment is not the AI. It is everything that needs to be in place before the AI can work.
The question is not whether agentic AI will reshape GBS. It is how quickly, and on whose terms.
For GBS leaders, this reframes the strategic priority. Getting to agentic at scale is not primarily an AI investment decision. It is a debt-reduction programme. The organisations that can deploy AI at scale have already built the process foundations, data governance, workforce capability, and integration infrastructure that agents require. This is precisely the territory where GBS already operates, and where a well-run GBS function can generate a competitive advantage that is difficult to replicate externally.
The renegotiation signal
While internal GBS organisations are working through the debt problem, the external market is sending a clear signal. HFS Research data from 2025 shows that 75 per cent of Global 2000 enterprises sought to renegotiate their business process outsourcing contracts, and 72 per cent their IT services arrangements. The stated reasons are consistent: services are seen as overpriced, slow, and built for a labour model that no longer represents value for money. FTE-based pricing, billable hours, and multi-year rigid contracts written for a pre-AI world are losing credibility with enterprise buyers.
This is simultaneously a threat and an opportunity for internal GBS. The threat is obvious: if the external market is moving toward AI-led, outcome-priced service delivery, an internal GBS model that remains primarily labour-based will face the same pressure as external providers. Business units that can access AI-led services more cheaply and quickly from the market will ask why they are paying an internal overhead to maintain an operation that is not keeping up.
The opportunity is less obvious but more significant. Internal GBS has structural advantages that external providers cannot easily replicate: deep knowledge of the business, trusted access to enterprise data, existing relationships with business unit leadership, and the absence of the commercial friction that characterises outsourcing arrangements. Where an external provider has to negotiate data access and integration rights, an internal GBS function already has them. In an agentic environment, where data access and system integration are the primary prerequisites for deployment, that advantage is real.
HFS Research frames the broader shift as the emergence of services-as-software: a new category of AI-driven, autonomously delivered, outcome-priced service that is projected to become a $1.5 trillion total addressable market by 2035, absorbing revenue from both traditional IT services and conventional SaaS.8 The GBS organisation that gets ahead of this shift can position itself as the delivery vehicle for enterprise services-as-software. The one that does not risks being disintermediated by external providers who do.
The biggest blocker to AI deployment is not the AI. It is everything that needs to be in place before the AI can work.
The iterative path
One of the most important insights from organisations that have successfully deployed agentic AI at scale is that the roadmap did not exist before the deployment began. It emerged through it. Each implementation revealed the next opportunity. Each deployed agent surfaced process gaps that became the next standardisation priority. Each successful implementation built the organisational confidence and technical capability to attempt something more ambitious. The roadmap is not designed in advance and then executed. It is discovered through iterative deployment.
This has practical implications for how GBS organisations should approach the transition. The temptation is to build a comprehensive AI strategy before deploying anything: extensive assessments, target operating model design, platform selection and governance workshops before a single agent goes live. This approach is not wrong, but it is slow and tends to produce impressive documentation rather than demonstrated capability. Gartner projects that over 40 per cent of agentic AI projects initiated through 2025 will be cancelled by end of 2027, the majority likely victims of over-engineering and under-delivery.11
The more effective approach is to start with the process and data foundations closest to production-ready, deploy into them, and build outward from evidence. Three principles govern this:
- •Start where the debt is lowest. Identify the processes with the most documented workflows, the cleanest data, and the highest volume. These are the natural first deployment targets. Success here generates organisational credibility and funds the next phase.
- •Build the trust infrastructure in parallel. Governance, human oversight protocols, exception management and AI risk frameworks cannot be retrofitted after deployment at scale. They must be designed for the agentic model from the outset. The organisations that are succeeding at scale have clear answers to who is accountable when an agent makes an error, how outputs are monitored, and how agents are retrained when performance drifts.
- •Treat the workforce transition as a design constraint, not an afterthought. Every agentic deployment changes what people do, not just how much of it they do. The GBS leaders who manage this well define the new human roles (agent supervisors, exception handlers, process designers, AI governance leads) before deployment, not after resistance begins.
The improving-the-present-as-stepping-stone principle is important here. Every agent deployed into today's operation simultaneously delivers near-term efficiency and builds the foundation for tomorrow's platform. These are not separate phases. The near-term deployment is the long-term foundation, accumulated one agent at a time.
How human roles in GBS change
The shift toward agentic-led services does not eliminate human roles in GBS. It changes them. Traditionally, GBS organisations rely on large teams of human operators to execute high-volume transactional work. In an agentic environment, machines perform much of that operational coordination while humans supervise, guide and intervene where judgement is required.
Several role transitions are already becoming visible:
Roles that shrink or disappear
- High-volume transaction processing
- Manual reconciliation and validation
- Rule-based document handling
- Routine operational coordination
Roles that evolve
- Process owners become service product owners responsible for AI-enabled workflows
- Operations managers become performance supervisors overseeing human–AI systems
- Risk and compliance teams focus on AI governance and oversight
Roles that emerge
- Agent supervisors, responsible for monitoring and guiding AI agent behaviour
- Agent product owners, defining service capabilities and performance metrics
- Process-to-agent designers, translating operational workflows into AI-enabled systems
- AI risk and assurance leaders, responsible for safety, auditability and governance
In this model, human capability shifts from operational execution toward design, judgement and supervision. GBS becomes less a workforce model and more an operating platform combining machine-scale execution with human oversight. This emerging model of human–AI collaboration is explored in more detail in our article Centaurs at scale: building high-performance human–AI teams.
What GBS leaders must do now
The strategic choice facing GBS leaders is not whether to pursue agentic capability. The market, the technology, and the deployment evidence have made that question redundant. The choice is whether to lead the transition or to manage the consequences of having allowed others to lead it. The organisations that move now will establish the process foundations, the data governance, and the deployment confidence that become the barriers to replication. The organisations that wait will find that those barriers now sit between them and the position they were trying to reach.
HFS Research frames 2026 as the year of "How to AI." The why is settled: labour economics, service buyer expectations, and the emergence of services-as-software have resolved the strategic question. The what is becoming clearer: agentic orchestration, AI governance, and outcome-based service models. What remains missing for most organisations, and where GBS leadership is now decided, is the how. How to build the AI business case that goes beyond cost reduction. How to systematically solve process, data, talent, and technology debt. How to structure the operating model for an AI-first delivery environment. These are precisely the questions that a structurally well-designed GBS organisation is positioned to answer.
The structural readiness required—operating model, capability map, governance and decision authorities, the performance system, and the change architecture—is within the broader transformation discipline. Getting the structure right is the prerequisite for getting the AI right. Neither succeeds without the other.
For GBS leaders, the practical agenda is threefold: assess and address the enterprise debt that limits deployment readiness, build the agentic capability iteratively from where the foundations are strongest, and position the GBS function as the governance and orchestration layer for AI across the enterprise—not just its own AI but the organisation's. That is what an enterprise spine does. That is what an agentic-led GBS makes possible.
References
- Gartner (2025). Gartner Predicts 40% of Enterprise Applications Will Feature Task-Specific AI Agents by 2026. www.gartner.com/en/newsroom/press-releases/2025-08-26-gartner-predicts-40-percent-of-enterprise-apps-will-feature-task-specific-ai-agents-by-2026-up-from-less-than-5-percent-in-2025
- JPMorgan Chase AI Research (2024). LAW: Legal Agentic Workflows. arXiv. arxiv.org/abs/2412.11063
- DHL (2025). DHL Boosts Operational Efficiency and Customer Communications with HappyRobot's AI Agents. group.dhl.com/en/media-relations/press-releases/2025/dhl-boosts-operational-efficiency-and-customer-communications-with-happyrobots-ai-agents.html
- Verizon (2024). Verizon Uses AI to Predict 80 Percent of Customer Service Call Reasons Before Agents Answer. www.verizon.com/about/news/verizon-ai-predict-customer-service-call-reasons
- Goldman Sachs (2024). Goldman Sachs Annual Report 2024: Artificial Intelligence and Technology Investment. www.goldmansachs.com/investor-relations/financials/annual-reports/
- Invensis (2024). AI in Business Process Outsourcing: Productivity and Cost Benchmarks. www.invensis.net/
- The Hackett Group (2025). GBS AI Adoption Report: 63% of Organisations See Early Gains. www.thehackettgroup.com/the-hackett-group-report-reveals-gbs-ai-adoption-accelerating-with-63-seeing-early-gains/
- HFS Research (2025). Looking back to look forward: What 2025 taught us and where 2026 is taking us. hfsresearch.com
- Deloitte (2025). Tech Trends 2026: Agentic AI Strategy. www.deloitte.com/us/en/insights/topics/technology-management/tech-trends/2026/agentic-ai-strategy.html
- CIO / IDC (2025). 88% of AI Pilots Fail to Reach Production. www.cio.com/article/3850763/88-of-ai-pilots-fail-to-reach-production-but-thats-not-all-on-it.html
- Gartner (2025). Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027. www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027