vertical cloud solution GmbH

Tessa

AI support that answers on its own and knows when humans need to take over.

With Tessa, we built an AI-based support system for gastromatic that handles incoming inquiries based on existing knowledge sources. The goal of the project is the fast, safe, and scalable processing of support inquiries. Tessa answers standard inquiries directly, assists with more complex cases by suggesting responses, and escalates sensitive or uncertain inquiries to the support team.

The distinctive approach lies in its clear guardrails: automation only where it is professionally and operationally responsible. The result is a support system that enables fast answers, relieves teams, and supports growth.

What convinced me: Tessa does not simply answer everything. It recognizes when a case is a fit for it, and when it is not. That is exactly what gives us the confidence to really put it to work."
Antonia Kühngastromatic | vertical cloud solution GmbH

How it works

Starting point and objective

From the status quo to a clearly defined objective

Starting point
  • gastromatic is growing continuously without letting customer support grow proportionally.
  • Support should be able to respond to inquiries faster and more efficiently.
  • Standard cases tie up a disproportionate amount of capacity, which is where the greatest potential for relief lies.
  • Clustered inquiries about current issues add to the workload.
Objective
  • Build an AI-based support system on top of existing knowledge sources in the Intercom system.
  • Tessa answers standard inquiries independently; for more complex cases it creates suggestions for support staff.
  • Critical or uncertain inquiries are handed over to people in a controlled way, so that human oversight is preserved in sensitive cases.
  • During acute issues, Tessa intercepts similar inquiries at high volume and integrates suitable notices into the answers on short notice.

Process chain

From inquiry to answer, safe and scalable

  1. 1.A user asks a question in the chat of the gastromatic system or via email.
  2. 2.The inquiry is captured as a ticket.
  3. 3.Tessa assesses whether the inquiry can be answered without risk. If so, it answers directly. If not, a draft response and a conversation summary are created and the inquiry is handed over to the customer support team.
  4. 4.Conversation histories can be retrieved later via the history.
  5. 5.Sources and answers are continuously reviewed and improved. This includes nightly source updates, the daily analysis of poor answers, and weekly ticket reviews.

Architecture highlights

Core building blocks

  • Intercom as the central interface: All incoming support inquiries run through Intercom, so Tessa integrates seamlessly into the existing workflow.
  • Diverse knowledge sources: The foundation is the existing knowledge base in Intercom as well as transcribed video content that is specifically prepared and integrated.
  • Human-in-the-loop approach: Tessa only answers autonomously where this is classified as safe. In critical or uncertain cases, human support staff take over, either directly or based on a suggestion from Tessa.
  • Guard rails for risk assessment: Incoming inquiries are systematically assessed before Tessa answers on its own. These guardrails ensure that sensitive or complex cases are reliably detected and escalated.
  • Answers with further sources: Users receive not only a direct answer but also a concrete reference to further content, for example an article in the manual or a matching video.