Stop Automating the Old Process: Why Agentic AI Demands a New Way of Thinking

Most GBS organizations are about to make the same mistake with Agentic AI that they made with RPA: pointing it at the process they already have instead of asking what the process should be.

That was the core warning from a recent SSON, Hypatos and EY webinar on why traditional process mapping falls short in the age of agentic AI. The speakers: Moritz Treutwein, Head of Partnerships, Hypatos; Jimmy Marquis, Managing Director, Technology Consulting; and Nitee Gupta, Executive Director, EY, shared a clear message: patching up old workflows with a smarter layer of automation will produce gains, but nowhere near the value that agentic AI is actually capable of.

The Difference Between RPA and Agentic AI

RPA thrived on predictability. Every step had to be templated, sequenced, and locked down. If you change one column header in a report, an entire bot fleet could grind to a halt, which made things very difficult if a supplier changed their invoice format. Agentic AI works the opposite way. It thrives on ambiguity and the freedom to reason toward an outcome, which means the moment you wrap it in RPA-style guardrails and rigid step-by-step logic, you cap its value before it has a chance to deliver.

The governance model has to shift, too. RPA governance was about control at the step level. Agentic AI governance is about verifying outcomes, tracing decision logic, and building trust in a system that's allowed to skip steps or take a nonlinear path when the data supports it.

Where is Agentic AI Most Effective?

Consider invoice processing, a favorite AI use case. The conventional approach starts with a metric, cost per invoice, then goes hunting for the bottleneck, typically an OCR exception queue where manual data entry is slowing things down. Deploy an agent there, watch the cost per invoice drop, and declare success.

It works. But it leaves value on the table.

The alternative starts with a bolder outcome: what if invoices could be approved and paid within the 10-day window vendors offer for early-payment discounts — terms most finance teams have long since written off as unrealistic? Chasing that outcome doesn't stop at the exception queue. It pulls in cash positioning, new approval structures, and a different combination of AI, agents, and humans working together. The redesign goes deeper, but so does the payoff.

The distinction matters: don't just automate where work is visibly stuck. Step back and ask what greater outcome your team could be driving, then decide which mix of AI, agents, and people gets you there.  Although there can be a tendency to test the water with small use-cases that aren’t mission-critical, the biggest ROI comes from processes with the largest transactional volume, e.g., those taking place between the supplier and buyer. So Hypatos’ suggestion is always to go for large volume processes such as procure-to-pay.

How to Approach Reimagining Processes, Not Optimizing

This is the philosophy behind EY's value blueprint methodology, built specifically to help GBS functions rethink how they operate with AI agents rather than retrofit AI onto legacy workflows. It works through a business function layer by layer, going from customer interactions, workforce collaboration, process design, trust and governance, and the underlying AI and technology foundation, to reimagine end-to-end value chains like hire-to-retire, procure-to-pay, and order-to-cash.

The principle underneath it is simple to state and hard to execute: design from zero. Stop optimizing existing processes and start without outcomes, then build systems that keep learning and improving as transactions flow through them. That's a fundamentally different exercise than automating today's tasks and steps.

Why the Timing Matters

There's real urgency behind this shift. The Hackett Group reports that SG&A costs have hit their highest levels in five years in both Europe and the US.  Cost pressure is real, but the bigger opportunity is what many GBS leaders are still underestimating: agentic AI's ability to create genuinely new value, not just trim expenses.

Part of that value comes from a capability humans simply don't have. Agents can share granular knowledge about every transaction they process, in real time, with every other agent in the system, building a form of collective intelligence no team of people could replicate through monthly review meetings and shared notes. That changes what a GBS operating model can look like, not just how efficiently it runs.

The Real Shift

The throughline across the entire discussion was this: agentic AI isn't RPA with better branding, and treating it that way is the single biggest way organizations shortchange themselves. RPA asked, "How do we do this faster?" Agentic AI asks "what should we actually be trying to achieve, and what's the smartest combination of people, agents, and AI to get there?"

That's a harder question to answer. It's also the one that determines whether an organization captures 20% of the value on the table, or closer to 80%.

For GBS leaders starting this journey, the practical takeaway is to resist the instinct to map your current process and drop AI on top of it. Start with the outcome you actually want — faster payment cycles, better cash visibility, higher-value work for your team — and let that outcome shape the process, not the other way around.

Access the whole webinar on-demand here.