Starter guide

Graph-based RAG
for enterprise
with Writer Knowledge Graph

• GUIDES
 Table of contents
Introduction
1
Graph-based RAG: the key to scaling enterprise AI
2
Graph RAG vs vector RAG
4
Writer Knowledge Graph: Benefits
10
Writer Knowledge Graph: Use cases
12
Getting started with Writer Knowledge Graph
18
Introduction
This guide explores the practical
applications of graph-based RAG 

and the ease of integration into
existing systems using Writer
Knowledge Graph. By the end, you'll
understand why graph-based RAG is
the future of enterprise information
retrieval and how Writer Knowledge
Graph can drive significant value for
your organization.

1
Graph-based RAG 

is the key to scaling
enterprise AI from
pilot to production
Despite high hopes and heavy investment, enterprises have entered
the trough of disillusionment regarding generative AI. A global
survey of 500 leaders at enterprise companies found that only 17%
rate their in-house AI projects as excellent. Additionally, McKinsey
reports that only 11% of companies have adopted generative AI at
scale. In short, CIOs and CAIOs are struggling to scale initiatives
beyond the Proof of Concept (POC) stage.

One of the main issues holding companies back? Large language

models (LLMs) trained on public data lack the specialized business
knowledge necessary to support company-specific use cases,
creating a gap.

Some companies attempt to tackle this challenge by fine-tuning an

LLM with internal data. But fine-tuning can be hard to get right,
costs money, and takes time. As a result, companies often end up
using generic models without context on their business.

Retrieval-augmented generation (RAG) has emerged as a viable

approach, combining LLMs with targeted data retrieval for more
accurate and relevant AI outputs. It lets users chat with internal data
and unlock various enterprise use cases. This includes answering
customer service questions in real-time, improving sales
enablement, analyzing data for better decision-making, and
simplifying compliance processes.


2
But many teams remain stuck in "POC purgatory” as they struggle

with generating consistently accurate answers.

D ivide the
P rompt

D ata content into + LLM O utput

smaller parts

C ontext

The RAG process

This is where graph-based RAG comes in. Unlike traditional

approaches to RAG that use vector databases, which can suffer from

high hallucination rates and crude data handling, graph-based RAG

delivers the accuracy and depth needed for practical applications.

It's not just better — we believe it's the only scalable solution for

generative AI in business contexts.

W hy is the graph R AG approach better ?

W ’
riter s full-stac k
Fine-tuning Full text search Vector DB
approach

Training a large Information searching Information searching Information searching

Definition language model with based on fuzzy match based on context 
 based on data relation

selected examples and keywords and intent and similarity

Cost at scale High Low Very High Average

Accuracy Low Average High High

Adding/updating
Hard Easy Hard Easy
new information

Q&A level 1 Average Good Good Good

Q&A multiple level Average Low Low Good

Data security Low Low Low High

3
How graph RAG stacks
up to vector RAG for
accuracy, scalability,
and affordability
Both vector RAG and graph RAG are techniques used for achieving
retrieval augmented generation. The main difference is how they
handle data, especially when it's complex and connected to other
data in an enterprise.

Vector RAG: simplified but limited

Vector RAG approaches data retrieval by converting data points into
numerical vectors, which are then stored in a vector database. This
method is straightforward and powerful for simple semantic
searches, like reviewing a corpus of similarly-structured documents.

But once you start adding structured and hierarchical data to the
mix, vector RAG struggles to capture the intricate relationships and
nuances. This is because vector representations can't show complex
data structures. This means they lose context and relationship
information that's important for understanding and retrieving
hierarchical or connected data correctly.

4
 Vector retrieval: fails with concentrated data
Are there any immunizations or
vaccinations that Mary S. has
received recently or is due for?



B10 A1
V7
A5 B6 Y6
C17
C6 F7
LLM
E0 C2 C11 C9
E6
S1
C3 C10 Mary S. is up-to-date on their
routing vaccinations, including
O2
C6 ? C7 Pneumococcal vaccine (PCV13),
C8 C12 Shingles vaccine, MMR, Hepatitis B,
and Varicella. They were last
E1 C4 C1 C5 administered on 06/01/2023.

C15
E4 B1
O1
F1 A2
B3 B2
F3

1 Enterprise data can be concentrated, eg manuals frequently mention

device names.
2 Given chunking, a combination of correct (green) and incorrect (red) data
points are chosen.
3 ANN/KNN pulls “n” nearest data points. LLM generates an answer that’s
not completely accurate.

5
In large business settings, data often exhibits complexity, with
numerous interdependencies and layers of information.

For instance, electronic health records (EHRs) capture and organize

detailed patient histories, including diagnoses, treatments,
medications, and interactions with healthcare providers.

Vector databases, by focusing only on the numerical representation

of data, may oversimplify these relationships. This can lead to
retrieval results that are close in terms of numerical distance but
lack the contextual understanding needed for accurate and relevant
responses.

Graph RAG: captures complexity and

context
In comparison, graph-based retrieval offers a more sophisticated
approach to RAG. Instead of simply looking at the distance between
data points, it builds a web of relationships between data points. All
data points become nodes, and their relationships to other points
become edges.

6
 Knowledge Graph maps rich semantic
relationship between data points
Are there any immunizations or
vaccinations that Mary S. has
received recently or is due for?



Graph built by
LLM
specialized LLM
date
8/22/2024 Mary S. recently received their
due Influenza Influenza vaccine (10/15/2024) and
Shingrix
 date their Tdap booster (8/22/2024).

second

dose

vaccine
Mary S. is due for Pneumococcal
Tdap booster
vaccine (PCV13) as they are over 65
06/01/2023
and have not received it yet. 


date Mary S. vaccine

PCV13 They are due for the second dose of
Singrix Shingles vaccine (Shingrix), with
the first dose administered on
06/01/2023.

1 Cost-efficient LLM trained to map data into graph structure of nodes and
edges.
2 Rather than chunking, intelligently processes data based on context and
maps semantic relationships.
3 Reliably retrie es correct data points for LLM to generate a completely
v

accurate ans er. w

Enterprise data isn t ust about indi idual points it s about ho

’ j v — ' w

those data points interact and relate to each other.

7
 What is the most we will pay out
for a non-owned trailer?

The maximum payout for a non-owned

trailer is $500.
Thought process View contributing sources

Add source View prompts

Knowledge Graph handles complex data formats where vector retrieval struggles

With graph-based retrieval, you’re not just finding the closest match —
you’re finding the right match by understanding the deeper connections
within your data.

Vector RAG: costly to scale

There are also the issues of rigidity and cost with vector-based RAG. Every
time you need to add new data, a vector database can’t just append it to
the existing dataset. It needs to rerun all the data and assign each data
object a new value. This is because the entire dataset determines the
value given to each vector embedding. And every time you change your
embedding model, it costs money. The larger the corpus of data your
company has, the more it’ll cost.

With new data added every day, an enterprise environment demands a

more dynamic, flexible, and affordable solution.

Maintain ncrease
I Stay secure
efficient costs your speed and compliant

67% L O W E R 

COST
We’ve seen 67% lower cost than
ntegrate data sources uic ly.
I q

Our full stac platform of LLMs,

K
- k

nowledge Graph, A I
k Secure your data and comply
with privacy regulations. Our
security first architecture gives
-

leading alternative approaches guardrails, and a fle i le

x b you control and our platform
when you use Writer instead of application layer ma es it
k doesn t persist or use your data.
'

traditional RAG, with savings easy to deploy in days.

increasing as you scale.

8
Graph RAG: saves on costs as you scale
Unlike vector databases that require frequent, costly updates, graph
RAG systems retain semantic relationships, which not only improves
accuracy but also simplifies complex decision-making processes.
This makes enterprises more agile and responsive without the hefty
price tag.

Graph-based systems are designed to scale more smoothly, which is

crucial for handling enterprise-level data. This scalability means that
as the data grows, the system can handle it without a significant
increase in cost.

Knowledge Graph scales efficiently as enterprise data changes

The Writer graph-based RAG approach, Knowledge Graph, further

boosts these cost advantages by being fast, cost-effective to
implement, and allowing for easy, inexpensive updates.
b a
a a
b

b
RAG vector
a a
database explained
Learn more on the blog
b a a b
a

9
Writer Knowledge
Graph empowers teams
to quickly and easily
deploy graph-based
RAG at scale
At Writer, our unique graph-based RAG solution, Knowledge Graph,
has helped hundreds of enterprises move from POC to powerful,
scalable implementations. Knowledge Graph grounds generative AI
in your company-level context by connecting our platform to your
internal data sources.

Knowledge Graph, our innovative approach to RAG

10
Get unmatched accuracy
Writer Knowledge Graph is a one-of-a-kind RAG solution and
outperforms other approaches. In a benchmarking test, Writer
Knowledge Graph took first in accuracy when compared with seven
other popular RAG implementations using vector retrieval. It had an
accuracy score of over 86 on RobustQA.

Derisk your AI strategy

Instead of slogging through months-long build cycles, fine-tuning
LLMs, and stitching together multiple tools to build your own RAG,
Knowledge Graph takes minimal effort to set up. Plus, it’s a key
component of the Writer full-stack generative AI platform that helps
you to build and deploy tailored AI apps in just days.

Stay secure and compliant

Knowledge Graph enables you to use your most confidential and
important data with generative AI. Writer adheres to global privacy
laws and security standards, such as GDPR, HIPAA, SOC 2 Type II,
and PCI. We offer flexible deployment options for our platform.

Lower your costs

We’ve seen that Knowledge Graph can offer 67% lower costs
compared to a DIY vector retrieval solution. Plus, as you scale, these
savings become even more pronounced thanks to Knowledge
Graph’s efficient handling and updating of data.

11
Bring speed and
intelligence to every
business workflow 

with Writer Knowledge
Graph
Writer Knowledge Graph isn't just a technological innovation — it's a
versatile tool that enhances efficiency and accuracy across multiple
enterprise applications. From boosting customer support to
empowering sales teams, its capabilities are making a significant
impact.

12
Gather information from multiple sources
Handling multi-hop questions that require gathering information
from various sources and conducting multi-step reasoning is
another standout feature. This makes the Knowledge Graph a
powerful tool for research and development teams, as well as for
sales and sales enablement tasks. Imagine helping your company to
reduce time spent on RFP responses by as much as 70%.

ASk WRITER How is Home Depot thinking about the impact of 


raw material costs, growing their ESG efforts, and
Your sessions New international expansion?

Search Home Depot's gross margin has been impacted by the

increase in raw material costs. According to the sources
Today provided, in the second quarter, Home Depot's gross
Raw material costs
margin was approximately 33.1%, a decrease of
approximately 1.9% from the previous years. This decrease
was primarily driven by supply chain investments and the
need to offset significant transportation and product cost
pressures.

Home Depot has implemented several ESG initiatives to

drive cost savings and sustainability...
Show sources

Add source View prompts Knowledge Graph mode Home Depot

Knowledge Graph excels at multi-hop questions where vector retrieval struggles

13
Get explainable, transparent answers to
every query
Explainable AI is essential for building trust. Knowledge Graph chat
sessions display the thought process behind each response,
breaking down complex questions into manageable subquestions
and highlighting specific source sections.

Is our customer feedback positive? What’s an example? Contributing sources

2024 customer feedback survey analysis

Customer feedback about Acme Company is positive.

2024 customer feedback survey [Link]
Here’s a customer testimonial:
 Customer satisfaction data indicates that Acme's
"Acme Company has surpassed all my expectations! Their customer support team consistently exceeds
expectations. Customers appreciate the team's friendly
outstanding customer service, superior products, and prompt and knowledgeable approach, which instills confidence
delivery make them my top choice. I highly recommend Acme and fosters a positive experience throughout the
support process.
Company for their reliability and trustworthiness!" - Sarah Johnson Last updated: Jan 4, 2024

Thought process View contributing sources

Customer testimonials page View source
Received question
[Link]/customer_testimonials
"Acme's customer support team is simply outstanding. I
Generated subqueries
had a complex inquiry, and they patiently guided me
through the process, providing clear and concise
Subquery #1 instructions. Their dedication to resolving my issue left
me feeling valued as a customer."
Is customer feedback positive?
Across various interactions, customers consistently expressed Last updated: Mar 14, 2024

satisfaction and appreciation for the exceptional service provided by

Acme Company. They frequently used words like "excellent," Customer NPS responses
"impressive," "outstanding," and "delighted" to describe their
Customer NPS [Link]
experience.
"Acme's seamless and efficient checkout process made
my purchase hassle-free. The clear instructions and
2024 customer feedback survey [Link]
secure payment options gave me peace of mind,
Slide 23 ensuring a smooth transaction."

Subquery #2 Last updated: Feb 8, 2024

Knowledge Graph
What is an example provides
of positive feedback? transparency and explainability
Acme Company has surpassed all my expectations! Their
Sentiment analysis
outstanding customer service, superior products, and prompt
outstanding
Generated final answer customer service, superior products, and prompt
What do you need?make them my top choice. I highly recommend
delivery Acme
User research & feedback Analysis of customer [Link]
Company for their reliability and trustworthiness! A comprehensive analysis of customer feedback
highlights Acme's customer support as a key
differentiator. Customers consistently praise the team's
Customer NPS [Link]
professionalism and dedication, emphasizing their ability
Pages 6-7 to provide comprehensive solutions and exceed
expectations.
Last updated: Jan 4, 2024

14
Deploy advanced knowledge assistants for
any team
The Writer Knowledge Graph excels in advanced question-
answering tasks. It acts as a knowledge assistant, interpreting
complex queries, retrieving relevant information, and using logical
reasoning along with summarization skills to construct
comprehensive responses. This capability is invaluable for customer
support and internal information retrieval systems. Some Writer
customers have seen as much as a 30% decrease in time spent
looking up answers to customer questions.

What is the coverage for a trailer under policy

#4252432?

Policy #4252432 provides comprehensive coverage for

your trailer. It includes protection against theft, vandalism,

fire accidents, and natural disasters. The coverage is

based on the trailer’s actual cash value.

Contributing sources

Policy #4252432 [Link]

Add source View prompts Knowledge Graph mode Support agent

Support knowledge assistant

15
C
​ ompare thousands of complex documents
in seconds
Writer Knowledge Graph can analyze and identify differences and
similarities between multiple documents. This is particularly useful
for legal, academic, and corporate environments where precision is
crucial. A tech firm is using Writer Knowledge Graph to power its e-
discovery tool for lawyers, which minimizes the “human toil” of
understanding vast corpuses of legal documents.

Cecilia

Hey there, I'm Cecilia – the personal Al

assistant!

You can ask me anything about this corpus

using the input at the bottom of this section.

Why did the fraud at Enron go undetected for

so long?

Enron was able to hide losses and mislead

investors for a long time due to its financial
inscrutability and the lack of investigation from
accountants and rating agencies. Additionally,
Enron's culture of financial sorcery allowed it
to pay its employees as if they were conducting
hugely profitable business.

"Enron's financial inscrutability hid stranger stuff."

"Enron did not entirely conceal that aspect of its
A legal tech company built an AI assistant withbusiness."
Writer Knowledge Graph that allows
users to quickly and easily understand what’s in a given corpus.

16
Transform knowledge into contextually
relevant content
Whether it's creating concise executive summaries or conducting
due diligence on target list products, Knowledge Graph makes sure
that content is accurate and contextually relevant, improving
productivity across various departments. The proof is in popularity
with end-users: One F500 financial services customer reports up to
70% improved advisor enablement when using a Writer-powered
knowledge assistant.

What is the coverage for a trailer under policy

#4252432?

The investment approach of the Global Sustain Fund, as

described in the sources provided, focuses on long-term
growth and incorporates environmental, social, and
governance (ESG) criteria. The fund aims to achieve a
significantly lower greenhouse gas emissions intensity
compared to the reference universe. It primarily invests in
intrinsically carbon-light companies and maintains a
lower carbon footprint than the broader market.
Additionally, the fund has a robust carbon-related
exclusions policy and filtering process.
Contributing sources
global_sustation_fund
Mar24 Investor

market [Link] [Link]

Add source View prompts Knowledge Graph mode Support agent

17
Get started with Writer
Knowledge Graph in
five simple steps
Step 1: Identify the core use cases you want
to start with
Knowledge Graph supports analysis, search, and question-
answering at scale. Start by discovering what kinds of questions
members of each team need to answer.

Use case examples

Support Sales enablement

Access knowledge base and Find relevant customer

provide answers on plans and references, quotes, statistics
products

Sales/Business development Research

Identify proposals that have Learn insights about a 


answered similar questions or subject based on dense
tackled similar SOWs in the past to documents (medical research,
speed up the drafting of a new financial documents, scientific
proposal papers, etc.)

Step 2: Map the files and data sources you

need to include
Discover the data sources teams are using today to answer the kinds
of questions you expect them to ask your Knowledge Graph.

18
Step 3: Employ data hygiene
Is there any data that should be excluded? Garbage in, garbage out
applies — make sure you're only importing data that's relevant and
trustworthy.

Step 4: Set up your Knowledge Graph data

connectors
Connectors make it easy to keep your company's Knowledge Graph
up-to-date and relevant. By syncing Knowledge Graph with your
internal databases, you can guarantee your team is always working
from the right source documents.

M ARKETI N G
Su gge st i o ns
Kn o w le d ge gr aph
Punctuation Set up your Knowledge Graph to improve content accuracy or find
Writing style answers to questions about your company’s data. Learn more.
Clarity graph
Retry files

Delivery PMM materials for cloud connection Add a Graph

Inclusivity
Edit graph details
Compliance Description

Sales enablement materials for Acme Enterprise, inclDelete

uding case studies and battle cards
Plagiarism
Custom rules data connector

General
Google Drive connection is live
CONTENT G ENERATION

Voice Connected folders

ReWrite Writer Marketing collateral

Magic links
Next Sync previous sync
D ata
Jan 26, 2024 12:05pm Sep 24, 2023, 12:05pm
Knowledge Graph 1,232/1,232 files synced from GDrive

Sources
Recordings
Manage connector

Delete connector

Knowledge Graph files

G raph status
24 files didn’t index Reindex
900 files in Graph

Search

Name Type Status Added by Date added

Founder [Link]
.pdf
Live in Graph
Peter Hatch SJdklf Jun 7, 2021

Uoi...

documentation_.pdf
.pdf
Live in Graph
Jun 5, 2021

Peter Hatch

Team roster_.doc
.doc
Syncing
Jun 5, 2021

Data pro
connectors and easy-to-use
duct-terminology (JAN).xlsx
.xlsx
Sinterface
yncing

Peter Hatch

Jun 5, 2021

Doris Jwo

workflow_requirements.csv
.csv
Sync error
Jun 1, 2021

Doris Jwo

acme-apollo case [Link]

.docx
Indexing error
May 28, 2021

Doris Jwo

jan.23-messaging_hierarchies.pptx
.pptx
Indexing
May 28, 2021

Doris Jwo

competitive_graph_infographichtielsss.png
.png
Sync error

Doris Jwo

May 28, 2021

19
[Link]
.jpg
Added to Graph
May 28, 2021

Peter Hatch SJdklf

3.23.21_announcements.eml
.eml
Indexing
May 28, 2021

Uoi...

Blog [Link]
.pdf
Live in Graph
May 28, 2021

Writer employs OAuth apps to manage access to your data sources.

Currently, we support data connectors for Confluence, Google
Drive, SharePoint, and Notion, with additional options in
development. You can configure these connectors through Writer-
managed apps or your company’s self-managed apps.

Our API facilitates effortless management of Knowledge Graphs and

their associated files. Whether you’re developing chat applications,
recommendation systems, or other AI-powered tools, the KG API
enhances and scales your development process efficiently.

Step 5: Build and deploy AI workflows

connected to your data wherever teams
work
Writer AI Studio simplifies the complexity of our full-stack generative
AI platform. It provides an intuitive development environment that
allows you to focus on building AI applications rather than managing
a complex stack. Our flexible development options, pre-built
templates, and comprehensive tutorials enable you to start quickly
and scale efficiently.

Novaphone+ | Support assistant Overview Build Deploy Troubleshoot

Configurations App guide

Welcome message
Hi, I’m your customer support knowledge
Customize the welcome chat message that users will in new sessions
assistant. Ask anything you’d like about
Hi, I’m your customer support knowledge assistant. Ask anything you’d
like about product features, customer information, active discounts,
product features, customer information, 

active discounts, and more.
and more.

Chat options

Show voice commands z

Customi e avatar
Show feedback flag
Show send to doc

Mode configuration

Enable at least one mode for your chat app

What do you need?
Chat mode
Use Chat mode to ideate or create content

Knowledge Graph mode

Built into AI Studio for rapid deployment of apps
Use Knowledge Graph mode to get answers from your team’s data

20
Writer integrations for Slack, Chrome, Outlook, Mac, and more

connect users of all technical levels with the tools they need. These

integrations, featuring built-in graph-based RAG, help teams

improve business workflows, make informed decisions, and enhance

creativity and productivity.

Writer AI APP

Search Acme Inc Writer is the full-stack generative AI platform for

enterprises.
Writer AI APP

Writer account-novaphone 1 2
How to use Writer?

3 Pinned Resources Designs Writer is the full-stack generative AI platform for

Threads 1 Matt Brewer 11:28 AM enterprises.

Home
Writer AI APP
Morning all! I agree, Zoe. I think a customized onboarding plan 11:30 AM
that addresses their specific
3 1 How to use Writer?
1
needs and pain points would be ideal. Select a chat app to use

DMs
Starred 4
Prospect research
announcements Writer AI APP 11:30 AM
1
Zoe Maxwell 11:2 9 AM Mode
general Select a chat app to use

Activity
Has everyone looked at the preliminary feedback from Novaphone about their e pecta ons? It x ti
KnowledgeProspect
graph mode
research
helpdesk
ti
might give us some insights into their priori es.

hiring Hi there! I help

Modeyou research a prospect using
Matt Brewer 11:2 9 AM
their latest financial documents.
Assistant
j
Pro ect Gizmo Yes, they are particularly keen on ensuring there is minimal disrup tion during
Knowledge graphthe
modetransition. We

need to emphasize our support and training capabilities.

-
account novaphone Hi there! I help you research a prospect using
How has Novaphone grown financially over the
Carmen Vega 11:2 9 AM their latest financial documents. Writer AI APP

More -
gizmo leads
last few quarters?
ti
I can prepare a sec on on our technical support and the dedicated resources we will allocate during their onboarding phase.

How has Novaphone grown financially over the

Direct messages Writer is the full-stack generative
Zoe Maxwell 11:35 AM last few quarters?
Arcadio Buendia 1
Thanks Carmen and Matt! Sounds like a plan.
enterprises.

How to use Writer?

Today

Apps
Writer AI APP 11:30 AM
Writer
Select a chat app to use
Zoe Maxwell 9:28 AM
Good point - let me check with Writer. Hang tight. I’ll ask! Select

Message #account-novaphone
Message Writer

Check out 

our developer
documentation

21
Fast-track your
enterprise AI strategy
for success with Writer
Knowledge Graph
Get started with Writer Knowledge Graph and lead your company
into the AI era. Join industry leaders like Vanguard, Accenture,
Qualcomm, and Vizient, who have transformed their enterprise data
strategies with our advanced, secure, and explainable RAG
technology. Request a demo today to see the impact firsthand.

22
Writer is the full-stack generative AI platform for
enterprises. We make it easy for organizations to deploy 

AI apps and workflows that deliver impactful ROI.

Our integrated platform consists of our top-scoring
Palmyra LLMs, our highly accurate graph-based RAG, 

AI guardrails to enforce brand and compliance rules, 

easy-to-use development tools, and a library of prebuilt
apps, extensions, and desktop experiences to get started
quickly. With Writer, enterprises build highly-customized 

AI apps that accelerate growth, increase productivity, 

and ensure AI compliance.


Our enterprise-grade platform can be deployed flexibly,
keeps your data private, and adheres to global privacy 

laws and security standards. Leading enterprises choose
Writer, such as Vanguard, Intuit, L'Oreal, Accenture,
Dropbox, and Kenvue.

Visit us at [Link]