---
title: "sipgate: chatbot platform and service bot | Die Lautmaler"
description: "For sipgate we build and run a chatbot platform, the service bot in the cloud PBX and the website bot, with continuous NLU and LLM optimisation."
url: https://www.lautmaler.de/en/projects/sipgate
lang: "en"
alternate_de: "https://www.lautmaler.de/projekte/sipgate"
---
[Projects](https://www.lautmaler.de/en/projects) sipgate

# sipgate

/ Customer service automation

**From chatbot to the biggest service channel.** What started as improving a chatbot is now a platform of its own that runs all of sipgate's bots. We build and run it together with sipgate, and the bot now answers more requests than the hotline.

Telecommunications Customer service LLM

![Besprechung in einem Glasbüro, davor das sipgate-Logo](https://www.lautmaler.de/uploads/xJa0zU1zrqtPQ-y5ih4Re-Teaser_Projekte_Sipgate.png)

/ Project description

## Situation

**A service chatbot that needed to do more.**

In the web app of sipgate's cloud phone system, customers set up their telephony themselves. They have plenty of questions about number porting, outages, features or invoices. The existing chatbot quickly reached its limits, and sipgate wanted more: a bot that knows the account, really helps and knows when a human should take over.

What started as a single assignment has grown into a long-term partnership over the years.

## A platform of its own

**The goal: bring the existing bot up to a solid level.**

- Our own NLU model instead of the Dialogflow standard

- Main use cases systematically captured

- Dialogues designed for the most important processes

**Result.** The bot understood much more. For the next step, however, the very simple Dialogflow solution of the time was the limit.

## Botify: The service bot in the phone system

## Quote

> XXXXXXXX

First name, Last name - Role

**New goal: a home of their own for all sipgate bots.**

Instead of patching the old bot any further, we built the Botify platform in 2021. It runs in Google Cloud and consists of many small components that can be replaced individually.

- Dialogue logic built with the open-source framework Jovo, plus language models trained specifically for telephony topics

- Connection to sipgate's login and interfaces, handover to support first via Zendesk, later via HubSpot

- Our own chat window, continuous monitoring with automatic alerts and an automated update pipeline (CI/CD)

**Result.** The bot can now do what wasn't possible before: it knows the account, detects outages on the sipgate platform and helps specifically, creates tickets on behalf of customers and hands over to humans when needed. Botify is still the basis of all sipgate bots today.

## Two bots we built for sipgate

## 1. Service bot

**In the cloud phone system.**

- The most important use case is number porting, i.e. moving your own number to sipgate. Plus help with outages, features and invoices

- Since 2024, a language model (LLM) answers questions from a knowledge base. It works together with the existing bot and now handles most of its tasks

- Proactive notices: when a chat starts, the bot points out known outages and informs about news in the account, such as a new porting status or new features

**Result.** The bot has become the largest service channel: 35.1% of all contacts go through it, more than through the hotline. And it doesn't wait for someone to ask. Customers often get the answer before they ask the question.

## 2. Website bot

**On sipgate.de.**

- Since 2022 for prospects and customers who are not logged in

- First asks whether someone is already a customer, then answers questions about products, prices and contracts

- Hands over to live chat when needed, creates a ticket or forwards to the right hotline, depending on whether it's a customer, a prospect or number porting

**Result.** Nobody ends up in a dead end, every request reaches the right team.

## Since 2021: Operations

**Operation and optimisation.**

- Continuous monitoring with automatic alerts and weekly reports with metrics per bot

- Weekly NLU cycle based on real, anonymised requests, with benchmark and review before go-live

- An LLM pre-checks conversations, a classifier labels them as resolved, redirected or escalated

- In 2026, we secured the update pipeline (CI/CD) and made the platform fail-safe

**Result.** sipgate sees week by week where the bots help and where they don't, and improvements go live in a controlled way.

## What we take away

**A bot isn't finished after go-live.**

The biggest lever wasn’t the rebuild itself, but what came after. Only with a platform of our own could we connect the bot to the account, outage data and support. And only ongoing operation with real conversations, which we evaluate and annotate, makes it better year after year.

At sipgate, we also learned that language models (LLMs) don't replace a good platform. They joined in 2024 as one component and now deliver most of the answers, because integration, handover and measurement were already in place.

[Visit sipgate](https://www.sipgate.de/)

35.1%

of all contacts go through the bot, more than through the hotline

3,800+

automated contacts per month

60+

topics covered

## Has your bot hit its limits?

Tell us about it.

[Book an intro call](https://www.lautmaler.de/en/contact)
