Energy suppliers receive thousands of customer requests every month – by e-mail and on the hotline. Many of them are standard cases: payment plans, payment extensions, invoice copies, moves. Large language models (LLMs) can take over a large part of this work, but only if they are connected to SAP securely and reliably. These are the lessons from a productive project.
Start with clear, high-volume business cases
“Use AI in customer service” is not a use case. The following are – and together they cover a large share of the daily workload:
- Payment plans: recognise the request, check the conditions and set up the installment plan in SAP S/4HANA Utilities.
- Payment extensions: grant a deferral of the due date within defined rules.
- Invoice copies: identify the customer and the invoice and send the copy automatically.
- Moves: capture move-in and move-out data and trigger the move process in SAP.
AI e-mail triage
Every incoming e-mail is classified by the LLM: which business case is it, which contract account, which data is missing? Clear cases are processed automatically, all others are routed to the right team in SAP Customer Engagement – with a pre-filled summary and a suggested answer.
Connecting the hotline: Cognigy and SAP
The same business cases are also available on the phone. A voice bot based on Cognigy answers calls, identifies the customer and executes the request directly in SAP – for example sending an invoice copy or registering a payment extension – without waiting in the queue.
Architecture: Azure, Databricks and SAP BTP
- Azure OpenAI provides the language model, hosted in a region that meets data protection requirements.
- Databricks is used for data preparation, analytics and the AI pipelines around the model.
- SAP BTP Integration Suite is the integration layer: it validates the model output, enriches it with master data, calls the right SAP APIs and handles errors.
- SAP S/4HANA Utilities and SAP Customer Engagement remain the leading systems for all business transactions.
Important: the LLM never talks to the SAP back end directly. The integration layer keeps SAP consistent even when the model is wrong.
Keep a human in the loop
Uncertain cases go to an agent, with the AI’s suggestion attached. The share of fully automated cases grows step by step – based on evidence, not hope.
Data protection and authorizations
- Define where the model runs and which data it may see.
- Use technical users with minimal authorizations for every system call.
- Log decisions so that they can be traced and audited.
Plan operations from the start
An AI solution is never “done”. Prompts, models and SAP releases change. Monitoring, a release process via ChaRM or Cloud ALM and a clear owner are essential. Also keep an eye on SAP’s own offerings such as SAP AI Core, Joule and the Model Context Protocol (MCP).
Conclusion
AI in SAP customer service works – when it is treated as an integration and governance project, not only as a model question. Read more about our AI in SAP services or get in touch.