Back office ops · Production

Swisscom builds a multi-agent Network Assistant on Amazon Bedrock to reduce engineer data-retrieval time by 10%

The problem

Swisscom network engineers spent significant time manually gathering and analyzing data from multiple decoupled sources, with certain tasks consuming more than 10% of engineer availability, introducing risk of human error and preventing focus on strategic work.

First attempt

The initial RAG implementation struggled with large input files containing thousands of rows of numerical data, and the retrieval process did not return the precise number of vector embeddings required for accurate KPI formula calculations.

Workflow diagram · grounded in source
1
Engineer submits natural language query
trigger
“facilitating accurate numerical calculations retrieval using natural language from the user input prompt”
2
RAG retrieval and context augmentation
ai_action
“Retrieval – User queries are matched with relevant knowledge base content through embedding models Augmentation – The context is enriched with retrieved information Generation – The large language model (LLM) produces informed responses”
3
Supervisor agent routes to sub-agents
routing
“Supervisor agent – Orchestrates interactions between documentation management and calculator agents to provide comprehensive and accurate responses”
4
Documentation agent retrieves insights
ai_action
“Documentation management agent – Helps the network engineers access information in large volumes of data efficiently and extract insights about data sources, network parameters, configuration, or tooling”
5
Calculator agent translates and queries
ai_action
“the agent translates natural language user prompts into SQL queries. In a subsequent step, the agent runs the relevant SQL queries selected dynamically through analysis of various input parameters, providing the calculator agent an accur…”
6
Bedrock guardrails validation
validation
“a series of guardrails were defined in Amazon Bedrock. The application implements a comprehensive set of data security guardrails to protect against malicious inputs and safeguard sensitive information. These include content filters that…”
7
Insights delivered to engineer
output
“engineers can have instant access to a wide range of network parameters, data source information, and troubleshooting guidance from an individual personalized endpoint with which they can quickly interact and obtain insights through natu…”
Reported outcome

The Network Assistant is estimated to deliver a 10% reduction in routine data-retrieval time per engineer, saving nearly 200 hours annually per engineer, at operational costs less than 1% of total value generated, enabling engineers to focus on strategic tasks.

Reported metrics
Engineer time on routine data retrieval tasks10% reduction
Hours saved per engineer annuallynearly 200 hours per engineer saved annually
Operational costs as share of total value generatedless than 1% of the total value generated
Engineer availability consumed by manual tasks (baseline)more than 10% of their availability
Reported stack
Amazon BedrockAmazon Bedrock Knowledge BasesAmazon Bedrock AgentsAWS LambdaAmazon S3AWS GlueAmazon AthenaPandasSparkRAG
Source
https://aws.amazon.com/blogs/machine-learning/transforming-network-operations-with-ai-how-swisscom-built-a-network-assistant-using-amazon-bedrock?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The Network Assistant is estimated to deliver a 10% reduction in routine data-retrieval time per engineer, saving nearly 200 hours annually per engineer, at operational costs less than 1% of total value generated, ena…

What tools did this team use?

Amazon Bedrock, Amazon Bedrock Knowledge Bases, Amazon Bedrock Agents, AWS Lambda, Amazon S3, AWS Glue, Amazon Athena, Pandas, Spark, RAG.

What results were reported?

Engineer time on routine data retrieval tasks: 10% reduction; Hours saved per engineer annually: nearly 200 hours per engineer saved annually; Operational costs as share of total value generated: less than 1% of the total value generated; Engineer availability consumed by manual tasks (baseline): more than 10% of their availability (source-reported, not independently verified).

What failed first in this deployment?

The initial RAG implementation struggled with large input files containing thousands of rows of numerical data, and the retrieval process did not return the precise number of vector embeddings required for accurate KP…

How is this back office ops AI workflow structured?

Engineer submits natural language query → RAG retrieval and context augmentation → Supervisor agent routes to sub-agents → Documentation agent retrieves insights → Calculator agent translates and queries → Bedrock guardrails validation → Insights delivered to engineer.

Related back office ops cases