Voice Agent and LLM
Spezi-Haus: AI-supported order processing
Spezi-Haus specializes in shipping sweets to private customers and processes a large number of orders, especially during peak seasons. Fraiss IT was pleased to support Spezi-Haus in modernizing its order processing system , relieving the workload of its employees, and ensuring reliable processing even with high order volumes .
Task
The goal was to automate order processing, which had previously been heavily dependent on the call center, while simultaneously enabling reliable control by employees .
Orders from all input channels should be centrally recorded, automatically recognized, and compared with customer, address, and product data. Orders that are not clearly identified should be transparently displayed and can be specifically addressed.
Customer & Sector
Spezialitätenhaus
Commerce
Services
AI Consulting
Web-App
IT & DevOps
UI/UX Design
Solution
A central web application consolidates orders from telephone, email, fax, and mail. Telephone orders are processed and transcribed by an intelligent voice assistant, while written orders are automatically analyzed. Employees can see the recognition level and order details and can either manually correct errors or approve the order.
Duration
Start: 08/2026 – ongoing
Tech stack
Rust
Angular
PostgreSQL
Kubernetes
Keycloak
The problem
Orders reach Spezi-Haus via various channels – by letter, fax, email and especially by telephone.
Especially during peak seasons, taking orders by phone reaches its limits: With thousands of calls per day, orders cannot always be recorded completely or correctly. If a call lasts too long, it is sometimes terminated – and with it, potential orders and revenue are lost.
At the same time, written orders in a wide variety of formats must also be reliably recognized and processed.
The solution
Fraiss IT developed a central solution for processing orders from telephone, email, fax and letter.
Telephone orders are taken and transcribed via a voice assistant.
Emails are collected centrally, while written orders and handwritten letters are automatically processed .
The identified data is compared with customer, address, and product data. A web application displays all orders along with their recognition level and origin. Orders that are not clearly identified are flagged and either reviewed by staff or followed up with the customer.
The order will only be processed further after manual approval. An additional debugging tool assists in analyzing and correcting recognition errors.
Conclusion
Automated order processing reduces dependence on the traditional call center and creates the basis for reliably handling even very high order volumes.
Telephone orders can be taken regardless of workload, while employees only need to intervene where manual review is actually required. This reduces order abandonment, speeds up processes, and frees up valuable capacity for customer service.
Team/Credits
PM: Zaim Sarajlic
Development: Benedikt Schmatz, Silvius Lehner, Martin Unger, Stefan Klinger, Christian Rieger, Katharina Rauch
In cooperation with: Euvic AI Consulting
Wolfgang Sommer
IT, Reports and Accounting
“Our collaboration was characterized by uncomplicated processes, a deep mutual understanding, and the highest level of professionalism – always accompanied by excellent solutions and a collaborative relationship.”
Would you like to know more about this project?
