Most technology upgrades simply give us faster tools to do the same old tasks. But every few decades, a breakthrough arrives that completely changes who or what executes the labor. Think of the automated assembly line: it didn't just help mechanics build faster; it handed the physical labor to machines so humans could focus on design and quality. Today, customer experience is having its own assembly line moment. Gartner predicts that by 2029, agentic AI will autonomously resolve 80% of common customer service issues without human intervention, and cut operational costs by 30%. Read that line again. That is not a chatbot answering FAQs. That is software that decides, and then acts.

So what does that actually mean for you if you run SAP — and how would we approach it at VASS? That is what this series is about. We'll take the art of the possible and break it down into practical scenarios that show how AI agents can work across your SAP landscape. Every scenario you’ll read here is buildable today with Custom SAP Joule Agents, built using SAP Business Application Studio and SAP AI Core, working across SAP Sales & Service V2, SAP S/4HANA, and SAP Commerce Cloud.

Part 1 . Explores what autonomous CX really means—and, just as importantly, where it delivers the most value.

Part 2. Walks through a retail product recall and shows how an AI agent can transform a complex, reactive claims process into an autonomous, end-to-end service experience.

PART 1

The Autonomy Ladder: What “Autonomous CX” Actually Means

Autonomy isn’t a switch you flip. It’s a ladder you climb — one rung at a time.

Let me start with a story you’ll recognize. A retailer launches a new robot vacuum, and it sells like hot cakes - more than 100,000 units in a matter of weeks. Customers register their purchase so the one-year warranty kicks in. Then a handful of complaints surface a serious defect, and the company makes the right call: a voluntary recall and free replacement for all affected customers.

There is one problem. Those robot vacuums were sold through dozens of retailers, so the manufacturer doesn’t know who most of the end customers are. It can’t just email them. Instead, it announces a recall campaign through retailers, social media, and the news, and launches a claims portal where customers can enter their product information, upload proof of purchase, and request a replacement — provided the purchase falls inside that one-year window.

Now imagine tens of thousands of claims arriving within days. In a traditional service organization, each claim is handled by a person. Someone reviews the receipt, verifies the product and serial number, checks the warranty dates, validates the claim, initiates the return, and triggers the replacement process. The process works—but only at a cost. It is slow, expensive, difficult to scale and often inconsistent. For a customer who paid for a faulty product that is unusable, every additional delay erodes trust.

We’ll revisit this exact story in Part 2 and rebuild it using autonomous AI Agents. For now, it illustrates the underlying idea behind this series: Autonomous CX isn't an all-or-nothing destination. It's a journey. Organizations move toward autonomy one capability at a time, climbing an autonomy ladder where each rung delivers more intelligence, more independence, and more business value than the last.

The three rungs of the autonomy ladder

Rung 1 — Manual. A person makes every decision and performs every action. The software is simply the system of record.

In our recall example, a service agent reviews the receipt, verifies the warranty, approves or rejects the claim, creates the return, initiates the replacement, and updates the system. Every claim consumes skilled human time, and consistency depends on who happens to handle the case.

Rung 2 — Assisted. Software helps people make better and faster decisions, but a person still owns every transaction.

Generative AI might summarize the case, draft the customer response, identify an expired warranty, determine the incident category, or flag a suspicious receipt. The human reviews the recommendations, makes the final decision, and completes the process.

This is where most organizations are today, and it's already delivering meaningful value. McKinsey estimates that applying generative AI to customer-care functions can lift productivity by 30 to 45 percent of the function’s cost<, with real cases showing a 14% jump in issues resolved per hour and a 9% cut in handle time. The important point, however, is that every case still requires human involvement.

Rung 3 — Autonomous. For the recall, the AI agent validates the receipt, confirms warranty eligibility, approves or rejects the claim, creates the return order, initiates the replacement, notifies the customer, and issues a shipping label. Human employees no longer process every claim—they supervise the system, monitor performance, and step in only when an exception or policy violation requires judgment. This is the rung Gartner is pointing at with that 80%-by-2029 forecast.

The question is no longer “should we use AI in CX?” It’s “which moments deserve to climb to Rung 3 — and which should deliberately stay on Rung 2?”

Isn’t this just workflow automation with a new name?

It's the first pushback we hear from customers: Isn't this just the workflow automation we already build in SAP Build Process Automation? It's a fair question. The difference isn't that agents replace workflows,  it's that they extend what workflows can do.

Think of a deterministic workflow as a self-checkout kiosk. Scan a barcode it recognizes, and everything moves quickly. Present something unexpected like an item without a barcode, a coupon that won't scan, and the process stops, waiting for a human to step in.

An agent is more like the experienced cashier. Instead of failing immediately, it examines the situation, applies the relevant policies, and either resolves the issue or escalates when it's genuinely uncertain.

In our product recall example, the workflow rejects a receipt because it's a screenshot instead of a perfectly scanned document. The agent reads the screenshot, extracts the product, serial number, and purchase date, evaluates the recall rules, and continues the claim without unnecessary human intervention.

In summary, deterministic, compliance-heavy, high-volume activities still belong in a workflow, and a well-designed agent will happily invoke one when consistency and governance matter most. What changes is who owns the judgment. In our autonomy ladder, traditional workflows excel at Rung 2 - automating predictable processes, while agents unlock Rung 3 by reasoning, adapting, and orchestrating work toward a business outcome.

So where does a fully agentic scenario actually make sense?

Not everywhere — and this is where good judgement matters. Chasing full autonomy for every interaction is a recipe for disappointment: Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, because the costs escalated, business value wasn't clearly demonstrated, and governance couldn't keep pace with the technology. We've found that the best candidates for autonomous execution share a few common characteristics. A customer interaction is typically a good fit for Rung 3 if it meets most of these criteria:

• High volume and repeatable. The same decision is made thousands of times with little variation – such as is the product under warranty? or is the proof of purchase available. Volume is what creates the business case for autonomy.
• Verifiable inputs and rules. The decision is based on facts the system can validate - such as product, serial number, warranty status, or purchase date.
• A clear system of action. Once a decision is made, the next steps are well defined and executable – such as creating a return order in S/4HANA, issuing a replacement order, notifying the customer or generating a shipping label.
• Safe exception process. When confidence is low or the situation falls outside policy, the agent can hand the case to a human without disrupting the customer experience. That’s what keeps autonomy safe.

This isn’t just our perspective. A Harvard Business Review Analytic Services report, sponsored by SAP, reaches a similar conclusion advising organizations to focus AI on experiences that are “frequent, easy, and predictable” — the tasks that, as Bain’s Darci Darnell puts it, “show almost no variation.” Where those conditions are weak - emotionally charged complaints, regulatory grey zones, or high-stakes negotiations - the right answer isn't more autonomy. It's defining the right level of autonomy. Knowing which rung fits which process is what separates successful AI programs from expensive experiments.

Why Autonomous CX Is Finally Practical for SAP Customers

It's fair to ask: Why now?

For years, autonomous customer experiences were more vision than reality. Three things had to happen before they became practical: technology had to become capable, organizations had to prove it worked at scale, and enterprise platforms had to make it safe to deploy.

Today, all three are true.

First, AI is capable. Foundation models can now understand unstructured business content—receipts, invoices, emails, and images—and reason over it with enough accuracy to automate real business processes.

Second, the market has validated it. Organizations have moved beyond experimentation. The momentum is real. McKinsey’s 2025 research finds 23% of organizations are already scaling an agentic AI system, with another 39% experimenting.

Third, SAP provides the enterprise foundation. An autonomous agent needs access to more than just a LLM - it needs business context, the ability to execute business processes, and the governance to operate safely in production. For SAP customers, those capabilities already exist within the SAP ecosystem. AI agents can securely access business data, execute transactions across applications such as SAP Sales & Service Cloud V2, SAP S/4HANA, and SAP Commerce Cloud, and operate within existing identity, authorization, and audit controls.