AI Agents in Customer Service

Three days after the hydraulic issue was resolved, the production manager reports back. The problem has occurred again—this time on a second machine of the same type. It is the third incident of this kind in six weeks.

The help desk agent opens the new service ticket in SAP Service Cloud. As we saw in the first article—automatic case creation, sentiment analysis, and routing—these steps have already been completed.

Here’s what happens next: The Case Action Recommendation Agent analyzes the case context and identifies the pattern. Its recommendation appears immediately upon opening the case: third recurring case involving the same machine class within 42 days, error code SC-441, SLA class: critical.

Recommended follow-up actions: Immediately contact the customer’s quality manager, provide Technical Bulletin TB-2024-441, and initiate an escalation review.

This is not a generic, predefined checklist. These are case-specific, well-reasoned recommendations that a specialist can evaluate in two minutes—instead of spending 30 minutes manually piecing together the same context.

At the same time, Joule, in conjunction with the Email Draft Recommender Agent, prepares a draft response for the customer: “We have identified the pattern and are taking immediate action.”

And in the future, the Case Management Assistant will enable a seamless workflow instead of using four separate AI functions individually—from pattern recognition to recommended actions to draft responses—all triggered by a single command to the Assistant. Agents will work in coordination with one another, not sequentially.

The Case Action Recommendation Agent changes the moment a case is initiated: Instead of an empty task, the specialist opens a well-founded recommendation. And the Case Management Assistant will also fully coordinate this process in the future.

That is the essence of AI agents in customer service when it comes to case handling.

A week later, the same production manager writes again—this time with a different request: He wants to know which preventive maintenance measures make sense for his three systems, and whether there is a maintenance package that includes regular inspections and wear parts. He asks about costs and availability.

The report is submitted via the self-service portal of the proaxia Customer Service Suite. The proaxia CSS Service Assistant receives the request. It accesses the customer’s current master and order data from SAP Service Cloud, SAP S/4HANA, and SAP FSA : three hydraulic systems, manufactured in 2021, warranty expired, last service call eight months ago, current contract class: Standard without preventive maintenance.

It correctly identifies the inquiry as a maintenance consultation request and creates a fully prepared case in SAP Service Cloud—with equipment context, contract history, and the specific customer inquiry as the starting point.

At the service desk, the inside sales representative opens the assigned case. What follows now would have been a manual process in the past: defining the scope of maintenance, researching prices, and preparing the terms of the quote.

A Quote Creation Agent can handle this step: It analyzes the equipment-product configuration, retrieves the relevant maintenance items from the service product catalog, calculates the package price based on the current customer price list, and creates a complete draft quote — featuring the “Hydraulics Preventive” maintenance package, which includes three annual inspections, seal kits, and filter replacements.

The employee reviews the quote, makes adjustments as needed, and approves it. The quote is sent back to the customer via proaxia CSS—within minutes rather than hours. And what began as a service call to address a problem has now turned into a business opportunity: a multi-year contract for preventive maintenance.

Here’s what a self-service assistant will be able to do in the future: Customers will be able to compare maintenance options or check availability directly in the portal—without having to contact the service desk. This means that the offer is generated through the interaction between the customer and the AI assistant—no longer just through an intermediate step involving a staff member.

A service request becomes a quote. A quote becomes a contract. Without any extra effort, without CRM maintenance, without a change in communication channels.
This is the essence of AI agents in customer service as the industry transitions from reactive to proactive service.

A second process is underway in parallel with the quotation process. The component manufacturer has delivered replacement parts—seal kits and pressure valves—to the plant manufacturer in two batches. The first batch has arrived at the customer’s site. The second is still in transit. Both batches have a problem. Two separate processes must be coordinated.

The production manager opens the first delivery and notices that three of the five seal kits delivered are damaged. He reports the damage via the proaxia Customer Service Suite’s self-service portal. The case is automatically created in SAP Service Cloud—classified and enriched with customer master data and order context.

Processing begins in the service back office: Which items are affected? Do the delivery terms apply? On what basis should a credit memo be posted? Without AI agents, this means manually searching for documents—such as invoices, delivery slips, and damage reports—across various modules, which is time-consuming.

The Billing Assistant—part of Autonomous Finance in SAP S/4HANA—can help with this. Once the affected line items have been identified and flagged, the Billing Adjustment Agent can then reverse the affected invoice line items and recreate the corrected documents—based on the complaint and the confirmed terms of delivery.

The service back office approves the change. The Email Draft Recommender drafts the confirmation message to the customer. There is no need for manual document search—all that remains is the decision and approval.

The second shipment is still in transit—and the customer doesn’t know it yet: A carrier issue has delayed the shipment. Without AI agents, this would have escalated much later: The customer waits, asks for updates, and receives no clear information.

The Logistics Predictive Insights and Response Agent in SAP S/4HANA has already detected the delay. It continuously monitors all logistics processes for carrier data and shipment status. At the same time, the Warehouse Outbound Optimization Agent performs an operational assessment: Which alternative warehouse has the required items? What is the fastest feasible delivery option? Based on this analysis, the Material Reservation Processing Agent triggers a special withdrawal order from the Stuttgart regional warehouse. Subsequent delivery: possible within 48 hours.

The service back office receives a structured recommendation for action, confirms the rescheduling, and forwards the information to the internal sales team. The customer is proactively notified that same day—before they even have to ask. What used to be an escalation is now a proactive service moment.

Here’s what else the Logistics Assistant will be able to do: Complaints and replacement deliveries will no longer be managed separately, but as a coordinated process. The Assistant recognizes when two delivery issues are occurring simultaneously for the same customer, prioritizes them accordingly, and ensures that no action blocks the other.

Two issues, one customer, two AI agent threads—coordinated rather than sequential. What used to be two separate escalations is now a proactive service interaction.
That’s the essence of AI agents in customer service for customer order and delivery processes.

The standard SAP agents cover the most common service patterns. For company-specific requirements—such as a warranty review agent, an escalation protocol for key customers, or a troubleshooting wizard for a specific machine class—the SAP CX AI Toolkit already offers two additional, ready-to-use modules.

In the Custom AI Tool Builder, administrators configure their own AI tools that access data directly from SAP Service Cloud, Sales Cloud, and Customer Data Platform. A prompt is linked to CRM fields—case description, equipment master data, warranty status—and generates structured output directly within the application context.

Example: An AI tool that, when a service case is opened, automatically analyzes all known cases involving the same machine class and provides the specialist with prioritized pattern recognition.

The Custom AI Agent Builder allows administrators to configure agents for company-specific processes—without requiring in-depth development knowledge. Configuration is performed using a structured wizard: agent type, trigger definition, and optional filter rules.

Triggers determine when an agent becomes active—for example, when an email is received, when a new case is created, or when a specific field is updated. Objects from the SAP Customer Experience Suite are available as data sources.

All agents run within the SAP Trust Layer: Personal data can be anonymized before LLM processing, and zero-data retention prevents data from being stored by third parties; furthermore, the data is not used for model training.

Example: An AI agent that evaluates the sentiment of each new customer interaction within a case and tracks it over time. If the tone deteriorates measurably over the course of several messages, the agent automatically triggers an escalation alert—the team leader sees the case flagged as a priority before the customer escalates it themselves.

These three scenarios demonstrate what SAP agents and assistants are capable of. At proaxia, we are also committed to developing process- and industry-specific AI agents that will fill additional gaps in the roadmap.

These include, for example, concepts for fleet-wide pattern recognition and proactive service outreach before a customer even submits a request. This is exactly what we see in the first scenario with our plant manufacturer: three instances of the same error pattern in a single machine class.

The real question, however, is: How many other customers are using the same system—and haven’t yet reported or identified the problem?

The concept of two interconnected agents is designed to fill exactly this gap:

Identifies common error patterns by equipment type and model series—not only from confirmed cases in SAP Service Cloud V2, but also from early indicators in the proaxia customer portal: unsuccessful self-diagnoses, repeated error code searches, and frequent manual lookups for a specific error pattern. Generates a structured pattern signal with confidence and severity scores. In the example of the SC-441 control module, this agent would have generated a signal after just the second instance—with an estimate of how many systems in the customer base might be affected.

Business Value: Earlier detection of fleet-wide patterns of issues before they escalate into a wave of individual cases. The service back office and team leaders receive a verified pattern rather than many unrelated cases. This enables targeted prioritization and proactive action.

It takes a sample signal verified by the Service Back Office and derives a suitable recommendation for action for each individual customer based on it. Two output paths: active outreach by the responsible account manager (subject to approval on a case-by-case basis) or a self-service notification in the proaxia customer portal (approval at the template level). For safety-related patterns—which are particularly relevant in medtech regulation—an additional signal entry is made in SAP S/4 QM for human evaluation. The decision to report remains entirely with Quality Management.

Business Value: Proactive rather than reactive customer engagement: End customers receive a notification or service offer before an outage occurs. “Human-in-the-loop” is a design principle at every decision point. Compliant with Clean Core.

We’d like to use these two agents to bridge the gap to the last article. Both agents are currently conceptual ideas within proaxia and are not existing features. However, they are intended to run entirely within a manufacturer’s system landscape, access SAP data exclusively via approved APIs, and be designed according to the “human-in-the-loop” principle: No action is executed without human approval.

This is exactly where the final article in this series begins: How agents collaborate across systems using SAP Service Cloud, S/4HANA, and SAP FSA—and what proaxia Seamless Service means in practice.