Purple Fabric Hands on

Guide For Customer 360

Introduction
Welcome to this hands-on activity session, where we will explore the art of the possible with the
Purple Fabric platform in transforming financial advisory processes. This guide will walk you
through creating and deploying the Customer 360 agent, demonstrating the platform's powerful
capabilities.

Introducing the Solution and the Problem It Solves

The Customer 360 Intelligence Agent aims to transform client relationship management by
delivering a holistic, AI-driven view of each customer. Traditionally, understanding client
sentiment, identifying risks of attrition, and uncovering upsell opportunities required manual
analysis of scattered financial reviews. Our solution automates this process using a React
Orchestrator and specialized RAG agents, providing structured, actionable insights from
complex financial documents. This showcases the art of the possible by enabling relationship
managers to engage more strategically, drive personalized growth opportunities, and strengthen
client loyalty — all while dramatically reducing manual effort.

Note - For every agent name add your name at the end For
example ‘Customer360_Yourname’

Breakdown of Knowledge Bases and Agents

To build this solution on the Purple Fabric platform, we will utilize several key components:

●​ Knowledge Bases:
○​ Customer360: The customer360 KB contains client meeting notes, complaint
email threads, and manager-customer conversations. It provides a consolidated
view of client sentiment, service issues, financial goals, and cross-sell
opportunities, enabling the Customer 360 Agent to generate deep, actionable
customer insights.
○​ Customer360_CFR: Contains Client Financial Review Documents specifically for
extracting structured financial review insights.​

1
 Purple Fabric Hands on
●​ Agents:
○​ RAG Agents:
■​ Customer360_Actions RAG Agent: Retrieves pending actions to be taken
by the personal banking manager from the customer360 knowledge
base.
■​ Customer360_Opportunities RAG Agent: Extracts cross-sell and upsell
opportunities for the client from the customer360 knowledge base.
■​ Customer360_Contentment RAG Agent: Analyzes client satisfaction and
sentiment levels using the customer360 knowledge base.
■​ Customer360_Risk RAG Agent: Identifies potential risks of the client
leaving the bank based on insights from the customer360 knowledge
base.
■​ Customer360_CFRD RAG Agent: Extracts a complete set of insights
(risks, opportunities, contentment, and actions) specifically from the
Client Financial Review Document present in the Customer360_CFR
Knowledge base, outputting results in a structured JSON format.
○​ ReAct Agents:
■​ Customer360_Orchestrator ReAct Agent: Manages and orchestrates all
client-related queries by invoking specialized RAG agents, and compiles a
comprehensive 360° client profile based on the outputs.

Overview of Upcoming Activities and Agents

In this session, we will guide you through the following steps to create and deploy the Customer
360 Intelligence Agent:

1.​ Creating Knowledge Bases: Set up two knowledge bases:

●​ customer360_docs: Import client documents such as meeting notes, complaint
emails, and manager-customer conversations. Configure chunking strategies and
generate vector embeddings to optimize document retrieval for client
engagement insights.
●​ customer360_cfrd: Import Client Financial Review documents separately.
Configure chunking and embedding to enable structured extraction of risk,
opportunities, contentment, and pending actions from financial reviews.

2
 Purple Fabric Hands on
2.​ Building RAG Agents: Configure the following RAG agents:
●​ Customer360_Actions, Customer360_Opportunities,
Customer360_Contentment, and Customer360_Risk — all accessing the
customer360_docs knowledge base to extract specific insights like pending
actions, upsell opportunities, client sentiment, and client retention risks.​

●​ Customer360_CFRD — accessing the customer360_cfrd knowledge base to

extract a structured JSON output of client risks, opportunities, contentment, and
pending actions.

3.​ Creating the ReAct Agents: Build the Customer360_Orchestrator ReAct agent,
responsible for intelligently handling incoming queries, orchestrating calls to the
appropriate RAG agents, collating their outputs, and generating a structured 360°
customer profile.

4.​ Publish: Deploy the Customer 360 Intelligence Agent, conduct functional testing, and
validate its ability to accurately assess client sentiment, identify opportunities and risks,
and generate comprehensive Customer 360° profiles based on real client documents.

By understanding each step and the agents involved, participants will gain a complete view of
how the Purple Fabric platform can be utilized to transform client engagement in personal
banking, showcasing the art of the possible in delivering hyper-personalized banking
experiences.

Step 1: Create a Knowledge Base for Customer360

Step 1.1: Initiate Knowledge Base Creation

1.​ Head to platform, choose Knowledge Hub.

2.​ In the Knowledge hub section, click Create New.
3.​ In the Create Knowledge window, enter the Knowledge base Name as
Customer360 and provide a suitable Description.
4.​ Click Create to initiate the creation of the Knowledge Base.

3
 Purple Fabric Hands on

Step 1.2: Configure a Chunking Strategy & Embedding Model

Document chunking refers to breaking down large documents or data sets into smaller,
more manageable chunks for processing. This technique improves the performance and
cost efficiency of Gen AI platforms by allowing parallel processing, resource optimization,
scalability, fault tolerance, and cost-effective operation.

1.​ On the top right of the platform , click Settings.

2.​ In the Configuration window that appears, choose the Block chunk strategy:
3.​ Click save to apply the chunking strategy.

A vector embedding model converts words and sentences into numerical vector
representations, which the computer can easily work with. This model is used to convert
the chunks into numerical vector representations for better understanding by LLMs (Large
Language Models).

1.​ Choose the ‘Azure OpenAI Text Embedding ADA 002’ option to convert the
chunks into numerical vector representations using Azure OpenAI’s
embedding model.

Step 1.3: Import Documents

1.​ On the import data page, click Import

2.​ In the import data window that appears, select the browse option and
upload all the necessary documents
3.​ Select all the documents and click open to close the window.

Step 1.4: Metadata and Tagging for Chunks

Add Metadata

1.​ In the Chunk Viewer section, select the checkbox next to the pdf name and all the
chunks in this document will be selected.(You can also select more than one pdfs
also.
2.​ Click on Add Metadata.

4
 Purple Fabric Hands on
3.​ In the Metadata window that appears, enter the Name as client_id and Value as
1001 for the selected chunks and click Submit.
4.​ Similarly follow the same process for 2nd client

Step 1.6: Experiment/Test the Knowledge Base

1.​ In the Knowledge Base creation page, click Retrieval Test.

2.​ In the Search bar, enter your query and press Enter.
3.​ Based on the query you entered, view the respective chunks of information.

Step 1.7: Publish the Knowledge Base

1.​ Click Publish.

2.​ The published Knowledge Base can be accessed in the Knowledge base home
page.

Step 2: Create a Knowledge Base for Customer360_CFR

Step 2.1: Initiate Knowledge Base Creation

1.​ Head to the platform and navigate to the Knowledge Hub.
2.​ In the Knowledge Hub section, click Create New.
3.​ In the Create Knowledge window, enter the Knowledge Base Name as
Customer360_CFR and provide a suitable Description.

Step 2.2: Configure a Chunking Strategy & Embedding Model

Document chunking refers to breaking down large documents or data sets into smaller,
more manageable chunks for processing. This technique improves the performance and
cost efficiency of Gen AI platforms by allowing parallel processing, resource optimization,
scalability, fault tolerance, and cost-effective operation.

4.​ On the top right of the platform , click Settings.

5.​ In the Configuration window that appears, choose the Block chunk strategy:

5
 Purple Fabric Hands on
6.​ Click save to apply the chunking strategy.

A vector embedding model converts words and sentences into numerical vector
representations, which the computer can easily work with. This model is used to convert
the chunks into numerical vector representations for better understanding by LLMs (Large
Language Models).

2.​ Choose the ‘’ option to convert the chunks into numerical vector
representations using Azure OpenAI’s embedding model.

Step 2.3: Import Documents

1.​ On the Import Data page, click Import.

2.​ In the Import Data window that appears, select the Browse option and
upload the Client Financial Review documents.
3.​ Select all the documents and click Open to close the window.

Step 2.4: Metadata and Tagging for Chunks

1.​ In the Chunk Viewer section, select the checkbox next to the PDF name so
that all chunks from the document are selected.
2.​ Click Add Metadata.
3.​ In the Metadata window:
a.​ Enter the Name as client_id
b.​ Enter the Value as the corresponding client’s ID (1001)
4.​ Click Submit to tag the chunks.
5.​ Repeat this process for each Client Financial Review document if multiple
clients are present.

Step 2.5: Experiment/Test the Knowledge Base

1.​ In the Knowledge Base creation page, click Retrieval Test.

2.​ Enter a query related to the financial review content (e.g., "Identify any risks
for client 1001") and press Enter.
3.​ Verify that relevant document chunks are retrieved correctly

6
 Purple Fabric Hands on

Step 2.6: Publish the Knowledge Base

1.​ Click Publish.

2.​ The published Customer360_CFR Knowledge Base will now be available in
the Knowledge Base home page, ready for agent integration.

Step 3: Build the agents

Creating the Customer360_Actions - RAG Agent

RAG agents are designed to enhance the quality of generated responses by retrieving
relevant information from a knowledge base before generating an answer. They are
particularly useful for scenarios where responses require specific, accurate information.

Step 3.1: Create a RAG agent

1.​ Head to the Expert Agent Studio module and click Create New.
2.​ In the pop up window that appears, Select the AI Agent.
3.​ Enter the name of the agent and description of the agent
4.​ In Type, choose Conversational Agent and click Create.

Step 3.2: Select a RAG Prompt Template

1.​ On the creation page that appears in the prompt template dropdown , choose RAG
Prompt Template.

Step 3.3: Select a Model and Set Model Configurations

1.​ Select the Azure OpenAI GPT-4o 4k model.

Step 3.4: Provide the System Instructions, Knowledge Base, and Examples

1.​ System Instructions: Craft a prompt that guides the agent in generating content.
Copy paste the below in the System Instructions section.

7
 Purple Fabric Hands on

System Instruction:
Answer the question based only on the following context: You are an expert agent
that can analyze pending actions to be taken by personal banking manager for the
client from various documents
Identify all the pending actions to be taken by the bank for the particular client.
Provide the answers in a numbered list

2.​ Add Knowledge Base: Click Add and select the Customer360 Knowledge Base.This
provides the necessary information for the agent to extract relevant and accurate
data.

Step 3.5: Run the model & Test the RAG agent

1.​ In the Test and Debug section, enter the following prompt. Identify pending
actions to be taken for the client?

2.​ Click or press Enter key to run the prompt.

3.​ Review the generated response to ensure it adequately addresses or clarifies your
query.

Step 3.6: Publish the RAG agent

1.​ Click Publish.

2.​ In the asset Details page that appears, enter the Welcome Message as
“Welcome to AI Chat!” Or Leave it blank
3.​ Click Publish and the status of the agent changes to Published and it can be
accessed in the Expert Agent Studio.

Follow the similar steps for below agents as well

8
 Purple Fabric Hands on

Creating the Customer360_Opportunities - RAG Agent

Copy paste the below in the System Instructions section.

System Instruction:
Answer the question based only on the following context: You are an expert
agent that can analyze opportunities for cross-sell and upsell for the client
from various documents provided to you in the context of banking and wealth
planning
Summarize the opportunities for cross-sell and upsell for the client and the
reasons to justify the opportunities in a detailed output with bullet points if
applicable.
Provide specific examples of the opportunities and reasons wherever possible.

Creating the Customer360_Contentment - RAG Agent

Copy paste the below in the System Instructions section.

System Instruction:
Answer the question based only on the following context: You are an expert
agent that can analyze client sentiment and satisfaction from various
documents
Analyze the overall satisfaction and sentiment of the client along with
examples as evidence and reasons for the output.
Also give a contentment score in the range of 1 to 10
Identify overall sentiment of the client as 'Very Negative', 'Likely Negative',
'Neutral', 'Likely Positive' or 'Very Positive'

9
 Purple Fabric Hands on

Creating the Customer360_Risk - RAG Agent

Copy paste the below in the System Instructions section.

System Instruction:
Answer the question based only on the following context: You are an expert agent
that can analyze risk of the client leaving the bank as a customer from various
documents
Summarize the risks for the client leaving the bank and the reasons for that risk in a
detailed output with bullet points if applicable.
Provide specific examples of the risk and reasons wherever possible.

Creating the Customer360_CFRD - RAG Agent - Attach CFR Kb

Copy paste the below in the System Instructions section.

System Instruction:
Answer the question based only on the following context:
You are expert agent who can analyse Client Financial review document and
come up with risks of client leaving the bank, opportunities for cross-sell and
up-sell, contentment and actions to be taken by the bank

Your job is to extract the following information from the json output:
1. Risk: you can come up with the risk of customer leaving the bank, you should
also provide the reason based on the analysis along with an example.
2. Opportunities: you can come up with the opportunities for up sell and cross
sell for the client along with what can be cross sold or up sold, you should also
provide the reason based on the analysis along with an specific example if
found.
3. Contentment: You can come up with the overall satisfaction and sentiment
of the client along with examples as evidence and reasons for the output. Also
give a contentment score in the range of 1 to 10. Identify overall sentiment of

10
 Purple Fabric Hands on
the client as 'Very Negative', 'Likely Negative', 'Neutral', 'Likely Positive' or 'Very
Positive'.
4. Actions: you can come up with the pending actions to be taken by the bank
for the particular client. Provide the answers in a numbered list, you should also
provide the reason based on the analysis along with an example.

Important : Do not provide any explanation apart from the intended json output
mentioned below
Output the information in a json:
""Risk":"",
"Opportunities":"",
"Contentment":"",
"Actions":""

Once you have the answer, your thoughts should start with 'Final Answer:'

Creating the Customer360_Orchestrator - ReAct Agent

ReAct agents are designed to perform tasks by reasoning through a sequence of actions.
They are ideal for scenarios requiring a series of logical steps to reach a solution.

Step 3.19: Create a ReAct agent

1.​ Head to the Expert Agent Studio module and click Create New.
2.​ In the pop up window that appears, Select the Team of Agents.
3.​ Enter the name of the agent and description of the agent
4.​ In Type, choose Conversational Agent and click Create.

Step 3.20: Select a ReAct Prompt Template

1.​ On the creation page that appears in the prompt template dropdown ,choose the
ReAct Prompt template

Step 3.21: Select a Model

1.​ As the model type for the agent, select: ‘Azure OpenAI GPT-4o 4k’ model.

11
 Purple Fabric Hands on
Step 3.22: Provide the System Instructions, Knowledge Base, and Examples

1.​ System Instructions: Add your prompt to "System Instructions" to guide the agent.
For example:

System Instruction:
Respond the following query/input as best you can.

You are a smart assistant for a Personal Banking Manager, designed to

understand the meaning behind user queries and respond accordingly.
Your role is to assist with customer-specific insights related to:
-Client Retention Risk
-Cross-sell and Up-sell Opportunities
-Client Contentment
-Pending Actions
-Comprehensive Financial Review
-Full Customer 360° Profile Generation

Task Execution Logic

1. Risk Assessment (Client Leaving the Bank):

If the query is about risk of the client leaving, execute the following tools
sequentially:
- Customer360_Risk
- Customer360_CFRD
Collate their outputs to derive a holistic risk assessment.

2. Opportunity Evaluation (Cross-sell/Up-sell):

If the query is about identifying potential cross-sell or up-sell opportunities:
- Customer360_Opportunities
- Customer360_CFRD
Combine results to generate an opportunity summary.

3. Contentment Evaluation:
If the query is about understanding customer satisfaction/contentment:
- Customer360_Contentment
- Customer360_CFRD
Compute an average contentment score and derive qualitative insights.

4. Pending Actions Overview:

12
 Purple Fabric Hands on
If the query is about pending actions or next steps for the client:
- Customer360_Actions
- Customer360_CFRD
Present a clear, actionable list.

5. Financial Review Evaluation:

If the query is about the client’s financial review document:
- Execute Customer360_CFRD
Return the exact response from this tool without modification.

---

Full Customer 360° Profile Generation

If the user request is to generate a comprehensive Client 360° Profile, then

perform the following tool executions in order:

1. Customer360_Risk
2. Customer360_Opportunities
3. Customer360_Contentment
4. Customer360_Actions
5. Customer360_CFRD

Then:
- Collate and group the responses under the four sections: Risk, Opportunities,
Contentment, and Action
- Compute the average contentment score across tools and highlight sentiment.
- Always mention the client’s first and last name in the report.
- Provide a detailed breakdown for each section.
- Final output should be structured and easy to interpret.
---
Once you have the answer, your thoughts should start with 'Final Answer:'

2.​ Add Actions: Click "Add" in the Actions section to incorporate the relevant tools.
Click"Add" next to the respective agents and provide the corresponding descriptions
to allow the ReAct agent to access them.

13
 Purple Fabric Hands on

Agent Agent Type Description

Customer360_Risk RAG Detects potential risks of client attrition, summarizing

reasons and specific examples of dissatisfaction if
present.

Customer360_Actions RAG Extracts pending actions that the bank needs to

perform for the client based on various client
interaction documents.

Customer360_Opportunities RAG Identifies cross-sell and upsell opportunities for the

client with supporting reasons, using client
communications and notes.

Customer360_Contentment RAG Analyzes the client's overall satisfaction and sentiment,

providing a contentment score and sentiment category.

Customer360_CFRD RAG Extracts structured insights (risk, opportunities,

contentment, actions) specifically from the Client
Financial Review Document in JSON format.

3. Provide Examples: Provide example scenarios to illustrate how the agent should
respond. For example:

Example 1:
Question
Create clients 360 profile for client_id "1001"

Thought
To create a client 360 profile, I need to first evaluate the risk of the client leaving the bank

Action
Customer360_Risk

Action Input
{{
"query": "analyze risk for the client", "filter":"client_id = 1001"
}}

14
 Purple Fabric Hands on
Observation
Answer

Thought
Now that I have the risk profile, I need to evaluate the opportunities for cross-sell and up-sell for
the client

Action
Customer360_Opportunities
Action Input
{{
"query": "analyze cross-sell and up-sell opportunities for the client", "filter":"client_id = 1001"
}}

Observation
Answer

Thought
Now that I have the opportunities, I need to evaluate the contentment of the client

Action
Customer360_Contentment

Action Input
{{
"query": "analyze cross-sell and up-sell opportunities for the client", "filter":"client_id = 1001"
}}

Observation
Answer
Thought
Now that I have the contentment, I need to identify the pending actions to be taken by the bank
for the client

Action
Customer360_Actions

Action Input

15
 Purple Fabric Hands on
{{
"query": "analyze contentment of the client", "filter":"client_id = 1001"
}}

Observation
Answer
Thought
Now that I have the actions, opportunities, risk and contentment from all the previous step, I
need to identify the risk, opportunities, actions and contentment from client financial review
Action
Customer360_CFRD

Action Input
{{
"query": "analyze the client", "filter":"client_id = 1001"
}}

Observation
Answer

Thought
Thought: I now know the answers from all the pervious step, I will combine all answers and that
will be my final answer

Final Answer
Formatted customer 360 report
​
Step 3.22: Run the model and Test the ReAct agent

1.​ In the Test and Debug section, enter the following prompts to check the
performance of the model.
a.​ Identify the risks of client_id "1001" leaving the bank
b.​ Evaluate contentment of the client
c.​ Identify pending actions to be taken for the client
d.​ Identify Opportunities for the client
e.​ Create client 360 profile for the client

2.​ Click or press Enter key to run the prompt.

16
 Purple Fabric Hands on
Review the generated response to ensure it adequately addresses or clarifies your
query.

Step 3.23: Publish the ReAct agent

1.​ Click Publish .

2.​ Click Publish and the status of the agent changes to Published and it can be
accessed in the Expert Agent Studio.

Congratulations!
You have now successfully completed the process of creating a multi-agent solution
using the Purple Fabric platform. By building the Customer 360 Intelligence Agent, we
have illustrated the art of the possible, demonstrating how Purple Fabric can transform
client engagement by delivering holistic, AI-driven customer profiles.

This session has provided a comprehensive overview of the critical components

involved, from the creation of multiple knowledge bases to the configuration of
specialized RAG agents and the integration of a powerful ReAct agent. Each step has
highlighted the ability of Purple Fabric to optimize information retrieval, automate deep
client analysis, and drive smarter, faster decision-making in personal banking.

The Customer 360 Intelligence Agent is not merely a tool, but a transformative solution
that redefines how relationship managers understand, anticipate, and serve their clients.
It enables banking teams to shift focus from manual data processing to strategic client
engagement, leading to improved loyalty, higher cross-sell success, and enhanced
service experiences.

We hope this experience inspires you to explore further and harness the full potential of
Purple Fabric in your workflows. The capabilities we covered today represent just the
beginning—there is a vast landscape of possibilities waiting to be unlocked, offering
endless opportunities for innovation, personalization, and operational excellence.

17