Back office ops · Production

Schneider Electric leverages Retrieval Augmented LLMs on SageMaker to ensure real-time updates in their CRM systems

The problem

Schneider Electric's account teams had to manually sort through new customers daily and link them to the correct parent entity in their CRM. Using LLMs alone was insufficient because their knowledge is limited by their training cutoff date, causing them to miss recent acquisitions, market news, and corporate restructurings.

First attempt

The initial blanket prompt approach achieved only 55% accuracy and did not generalize well to education or healthcare domains, where the notion of 'parent company' is not meaningful.

Workflow diagram · grounded in source
1
New account triggers linking need
trigger
“new customers are added daily, and their account teams have to manually sort through these new customers and link them to the proper parent entity”
2
Domain classification via RAG
ai_action
“we also had to identify the domain a given account belongs to. For this, we also used a RAG where a multiple choice question "What is the domain of {account}?" as a first step”
3
Web search retrieval
ai_action
“The given company name is combined with a question like "Who is the parent company of X", where X is the given company) and passed to a google query using the Serper AI”
4
LLM synthesizes parent company answer
ai_action
“The extracted information is combined with the prompt and original question and passed to the LLM for an answer”
Reported outcome

Domain-specific prompt engineering boosted overall linking accuracy from 55% to 71%, reduced manual effort in linking workflows, and delivered faster data access to downstream analytics teams.

Reported metrics
Account linking accuracy after domain-specific prompts71%
Account linking accuracy before domain-specific prompts55%
Manual linking effortreduce the manual effort
Downstream analytics data access speedfaster data access
Reported stack
Amazon SageMaker JumpStartFlan-T5-XXLLangChainSerperAmazon SageMaker Studio
Source
https://aws.amazon.com/blogs/machine-learning/schneider-electric-leverages-retrieval-augmented-llms-on-sagemaker-to-ensure-real-time-updates-in-their-crm-systems?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Domain-specific prompt engineering boosted overall linking accuracy from 55% to 71%, reduced manual effort in linking workflows, and delivered faster data access to downstream analytics teams.

What tools did this team use?

Amazon SageMaker JumpStart, Flan-T5-XXL, LangChain, Serper, Amazon SageMaker Studio.

What results were reported?

Account linking accuracy after domain-specific prompts: 71%; Account linking accuracy before domain-specific prompts: 55%; Manual linking effort: reduce the manual effort; Downstream analytics data access speed: faster data access (source-reported, not independently verified).

What failed first in this deployment?

The initial blanket prompt approach achieved only 55% accuracy and did not generalize well to education or healthcare domains, where the notion of 'parent company' is not meaningful.

How is this back office ops AI workflow structured?

New account triggers linking need → Domain classification via RAG → Web search retrieval → LLM synthesizes parent company answer.