How a Swiss Pension Fund Automated 1,300 Customer Enquiries per Month
Many organisations face the same challenge.
The customer service team is competent, friendly and well-organised. But staff still spend a significant portion of their time answering the same questions over and over.
For pension funds, these questions are particularly sensitive. They concern financial decisions, retirement planning and personal life situations.
For example:
- How high will my pension be?
- Can I take a lump sum instead of a pension?
- When will I receive my payout?
- How does partial retirement work?
These questions must be answered correctly. At the same time, they generate a high volume of recurring enquiries every day.
This is exactly the challenge that a project at the Aargau Pension Fund (APK) set out to solve.
The Starting Point: Too Many Standard Enquiries
The APK had long handled member enquiries in the traditional way: phone, email and in-person counter contact.
The customer service team was highly experienced and could answer questions competently. But a large proportion of enquiries covered standard topics – pension amounts, conversion rates, lump-sum withdrawals, voluntary top-up contributions, and administrative processes. These questions are important, but they repeat constantly.
The result: a significant amount of time was spent on routine enquiries, leaving less bandwidth for complex advisory conversations. And outside office hours, members simply had to wait.
The Idea: An AI Chatbot as a Knowledge Interface
As part of a website relaunch, the APK decided to try a new approach. Rather than simply offering a traditional website, an AI would help answer questions directly.
A chatbot was implemented based on the pension fund’s own content. The knowledge base includes:
- Website content
- Pension and organisational regulations
- Fact sheets
- FAQ documents
- Internal process information
The AI draws exclusively on these sources. This means answers are not invented – they are based on the organisation’s actual documents.
How the Chatbot Works
Technically, the chatbot is built on an approach known as Retrieval Augmented Generation (RAG).
The AI first searches the existing documents for relevant information, then formulates an answer. Crucially, the answer can reference concrete sources directly. Users don’t just receive a statement – they can also see which document it is based on.
If a question falls outside the available knowledge base, the user is automatically referred to the customer service team. Control always remains with the organisation.
Results in Production
After going live, it quickly became clear that many members were using the new service.
Today, the chatbot processes around 1,300 enquiries per month fully automatically.
Key figures from operations:
- 96.7% success rate in answering questions
- Average of 2.5 messages per chat
- 97% of answers are based directly on official documents
Most members receive their answer within seconds.
Most Common Question Types
Enquiries fall into three main categories.
Pensions and Benefits: Questions about future pension amounts, conversion rates, partial retirement options and lump-sum withdrawals.
Contributions and Changes: Questions about voluntary top-up contributions, advance withdrawals for home ownership, and the impact of salary changes.
Processes and Deadlines: Administrative questions about forms, deadlines and procedures. This information often already exists but is hard to find on traditional websites. The chatbot makes it immediately accessible.
Benefits for Members
The most important benefit is speed. Instead of searching for information, members simply ask a question and receive an answer within seconds.
Availability is another key advantage: the chatbot is accessible around the clock, regardless of office hours.
Benefits for Customer Service
Routine questions are now handled automatically, freeing up time for conversations that genuinely require personal advice. Especially for complex pension topics, this personal exchange is particularly valuable.
Conclusion
A large proportion of customer enquiries are recurring and based on information that already exists. When this information is made structurally accessible, AI can take over a significant portion of the workload.
Customer service is not replaced. It is supported. And that is the real value.