Exclusive Insights from

Dorota Zablocka, Head of GBS EMEA, Randstad Enterprise

_{EXCLUSIVE EXPERT INSIGHTS}

Why Agentic AI Success Starts with People, Process and Governance – Not the Platform

Dorota Zablocka, Head of GBS EMEA, Randstad Enterprise explains why the biggest breakthroughs have come from empowering frontline teams, strengthening governance and redesigning workflows, not simply deploying new AI tools.

When you first explored agentic AI, what outcomes were you expecting to achieve, and how do those expectations compare with reality?

When we first dipped our toes into agentic AI over two years ago, our expectations were very ambitious.

We anticipated a plug-and-play solution that could autonomously orchestrate complex workflows across our talent solutions from day one. The reality, however, highlighted that successful adoption requires much more than technology alone. Like many organisations, we initially approached agentic AI as a standalone productivity tool before recognising the importance of embedding it into broader operational workflows.

It quickly became clear that realising the full value of agentic AI requires strong foundations in data, processes, and organisational readiness. Fragmented data and slower-than-expected adoption initially limited our progress. This experience closely mirrors what Randstad Digital’s report describes as the “Productivity Paradox” – organisations often deploy technology faster than they develop the human capabilities needed to maximise its value. Increasing tool velocity without upskilling simply scales technical debt.

The turning point was shifting our strategy away from the technology itself. We developed a rigorous evaluation scorecard and decision matrix to assess scalability against strict requirements. We realised that before unleashing agents, you must first master the fundamental work architecture: identifying the exact problem, mapping the precise workflow, and clearly defining the human-to-agent boundaries.

Where have you seen agentic AI create the greatest business impact so far, and what surprised you most about the results?

The greatest business impact hasn’t come from a top-down, boardroom executive mandate, but from an empowered, bottom-up movement. The real magic happened when we properly trained our frontline GBS teams, and they shifted from fearing displacement to actively submitting their own "wishlists" for agent deployment.

In our world of RPO, MSP and SOW management, routine tasks used to burden our specialists with repetitive administrative activities... By leveraging agentic AI to offload this manual labour, our people were instantly released to focus on high-value work, which radically boosted both our operational quality and client experience.

What surprised me most was how quickly this transformed our internal talent dynamics. According to Randstad’s data, AI fluency paired with uniquely human capabilities, like emotional intelligence and critical judgment, has become the new career currency, accelerating promotions across industries.

Our employees didn’t just become more efficient; they evolved into orchestrators of AI, treating agents as advanced collaborators rather than simply tools. This shift from manual execution to strategic oversight has redefined what human potential looks like in shared services.

Can you share an example where poor data quality, fragmented process ownership, or inconsistent governance limited the effectiveness of an AI initiative?

A good example occurred when we attempted to deploy an agentic AI

to automate and measure time-to-hire metrics across several high-volume RPO accounts. The ambition was to have an autonomous agent track bottlenecks in our interview loops and dynamically reschedule talent pipelines to optimise delivery.

As we scaled the initiative, we quickly realised that differences in regional processes, data definitions, and governance models limited the effectiveness of the solution. Different talent hubs were using varying recruiting definitions within their Applicant Tracking Systems (ATS), making standardisation essential before the agent could operate effectively.

Rather than abandoning the initiative, we proactively addressed the challenges by establishing centralised governance... We halted the rollout, mapped the global data architecture, and established strict accountability by assigning dedicated process owners to manage data parameters at the regional level. This active intervention successfully bridged what Randstad's workforce insights identify as the global AI integration gap.

By resolving the underlying workflow fragmentation rather than just blaming the technology, we turned the initiative around. Today, the standardised agent safely and accurately optimises our pipelines, proving that robust human management is what prevents a sophisticated agentic ecosystem from defaulting to chaos

What processes or functions are proving most suitable for agentic AI today, and what characteristics make a process ready for autonomous execution?

Today, the functions proving most suitable for agentic AI within our GBS ecosystem are those characterised by high-volume, rule-based complexity

rather than purely creative ambiguity. In our operations, workflows like multi-jurisdictional candidate credential validation, first CV screening, first outreach to candidates, and job reposting are particularly well suited for automation.

To determine readiness, we utilise Randstad’s "4 A's" framework (Automate, Augment, Archive, Amplify) to systematically assess and deconstruct tasks.

A process is ready for autonomous execution when it exhibits three key characteristics:

Structured, Digitised Inputs: The data must flow through clean APIs or standardised forms rather than unstructured, unpredictable channels.
Immutable Logic Parameters: The underlying business rules must be clearly defined, with minimal reliance on subjective, ad-hoc human policy interpretations.
Explicit Hand-off Thresholds: There must be an unambiguous, pre-programmed boundary where the agent knows exactly when to execute and when to trigger a human escalation loop.

When a workflow meets these criteria, agentic AI shifts from a simple scripted assistant to a true autonomous engine capable of executing end-to-end operational cycles safely.

As agentic AI takes on more decision-making and execution responsibilities, how are you approaching governance, accountability, and human oversight?

As agentic AI transitions from a passive assistant to an autonomous decision-maker, our approach to governance has shifted from passive monitoring to robust, human-led oversight. In the talent management sector, saying "the AI made the decision" is a fast track

to regulatory, ethical and legal risks. Accountability must always remain entirely with humans.

To manage this, we have institutionalised a strict "Human-in-the-Lead" operational framework. We don’t just monitor outputs; we require our systems to maintain transparent, auditable decision paths so our teams can immediately understand why an agent took a specific action and verify its alignment with strict compliance standards.

This shift has sparked a massive structural change in our hiring profiles. Reflecting Randstad Digital's data, we are seeing a massive surge in demand for specialised oversight roles, such as AI Trainers (which have grown 281% globally) and Process Automation Specialists.

These professionals don't just build the technology; they serve as critical human gatekeepers who continuously audit model reliability, mitigate bias, and ensure that autonomous execution never outpaces ethical judgment.

Where are agentic AI initiatives being held back today, and how much of that is a data, process, or ownership challenge rather than a technology limitation?

From our experience, agentic AI initiatives today are rarely limited by the underlying technology itself. The large language models and multi-agent frameworks available are exceptionally powerful. Instead, the real bottleneck is often fragmented data, non-standardised processes, and unclear ownership models.

While legacy process ownership is usually well-defined within traditional business lines, ownership of the AI agent itself remains a massive corporate grey area. Who is responsible when an agent

drifts? Who audits its cost-to-compute or handles its optimisation?

Furthermore, Randstad’s "AI Capability Gap" report underscores a critical enterprise bottleneck: a severe shortage of operational leadership and integration talent, with roles like AI Solutions Leads taking up to 54 days to fill.

In fact, over half of tech professionals are forced to seek training independently because internal corporate upskilling programs simply cannot keep pace with tech evolution. We are experiencing a capability crisis; organisations are aggressively buying advanced tools but failing to build the necessary talent infrastructure required to scale and govern them effectively.

What foundational changes have had the biggest impact in enabling agentic AI to deliver measurable value?

The single most impactful foundational change we made was rewriting how we define and measure the ROI of automation. Historically, shared services leaders have primarily measured success through “time saved” or headcount reduction. However, we realised that agentic AI often produces work outcomes or deep analytical audits that humans would simply never have the bandwidth to perform in the first place.

Therefore, our primary value metric has shifted from traditional time savings to “time avoidance”, preventing costly compliance errors, avoiding bad talent placements, and removing the friction of manual rework. Introducing this level of flawless quality into reputable, high-volume tasks has completely changed our operational baseline.

Another foundational pillar has been fostering an "AI-orchestrator mindset" among our management tiers, a concept heavily championed in

Randstad's Advisory services. Instead of looking at AI as a fancy typing assistant, our managers now treat agents as cognitive co-thinkers and operational leverage. This cultural pivot from individual contributor to AI orchestrator has finally allowed us to unlock measurable, scalable value.

If you could give one piece of advice to a GBS leader who is currently evaluating agentic AI, what would it be, and what is the one mistake you would urge them to avoid?

If I could offer one key piece of advice to a GBS leader currently evaluating agentic AI, it would be this: Treat your workforce capability as a foundational layer of your technology stack, not as an operational overhead.

The biggest mistake leaders can make, and one I urge you to avoid at all costs, is investing in technology before clearly identifying the business problem and mapping the underlying workflows. Do not attempt to force an autonomous agent into a broken, fragmented process; you will only succeed in generating technical debt and operational errors at an unprecedented scale.

As Randstad’s research continually reminds us, technology alone cannot drive enterprise growth. The real bottleneck is the human element. Before writing a single check to a software vendor, invest heavily in building a high-velocity AI fluency toolkit within your teams.

Redesign your workflows for a human-centric future first, establish your data governance, and only then select the specific agentic tool that matches your long-term requirements for scaling.