Data+AI 2025 | www.factor-insights.com
Graphs & GraphRAG
Essential Ingredients
for Enterprise GenAI
Thursday | March 27
Factor Data+AI Day
Events Centre, Collins Square Melbourne
Emil Pastor
Head of Solutions
Engineering A/NZ
Neo4j
Data+AI 2025 | www.factor-insights.com
"Science Finds, Industry Applies, Man Conforms”
“Sociotechnical” Principles
Motto of the 1933 Chicago World’s Fair
Data+AI 2025 | www.factor-insights.com
AI requires a Holistic Approach1
IT Infrastructure & Management
Data Engineering & Quality Management
Machine Learning & Analytics
Project Management
Organisational Change
Industry Transformation
Regulation
Principles & Philosophy (e.g. Ethics)
1 – T. Evgeniou, INSEAD
Your business
is agile and
dynamic
But your data
is locked in
rows and
columns
Valuable
patterns are
hidden in
your data
Model your
data like
your
business
Data, meet graph
Employees
Network & Security
Suppliers
Product
Customers
Transactions
Process
The power of
the graph model
Graph is flexible
Employees
Network & Security
Services
Citizens
Transactions
Process
Suppliers
Graph is insightful
Employees
Network & Security
What’s important?
Process
Suppliers
What’s unusual?
Services
Transactions
Citizens
What’s next?
13
Data+AI 2025 | www.factor-insights.com
Generative AI races
toward $1.3 trillion in
revenue by 2032
Generative AI Revenue
Generative AI as a % of Total Technology Spend However, 71% of
organizations are
stuck piloting
GenAI projects
2024 IBM CEO Survey
Bloomberg Intelligence
Data+AI 2025 | www.factor-insights.com
85% of IT leaders
cite data quality as
their top AI
challenge
KPMG AI Quarterly Pulse Survey 2025
100 senior executives from $1B+ revenue
companies
It’s a data management issue!
Lack of Standardisation
Scale of Query Diversity
Data Organization
AI needs data organized in
ways it can effectively access
AI requires high-quality
harmonized data to be reliable
AI demands rapidly accelerated
question-to-query execution
Data+AI 2025 | www.factor-insights.com
Flexible Standardised
Enables AI to grasp meaning
and access data across
sources
Supports diverse query types
and evolving data needs
Maintains consistent
understanding at scale
What makes data AI-ready?
Data exposes
entities and
relationship
understanding
Adapts to new
patterns, structures
and access logic
Consistent,
governed
data foundation
Contextual
Data+AI 2025 | www.factor-insights.com
Connected Data
Node
Property Graph Data Model
Person
Name:Terry
Person
Name:Sam
Car
Brand:Volvo
Since:10/Jan/11
DRIVES
OWNS
SISTER_OF
BROTHER_OF
Since:10/Feb/11
Relationshi
p
Property
= AI-ready data
Knowledge graphs
Data+AI 2025 | www.factor-insights.com
Flexible Standardised
Entities and their connections
are built into the data model
Evolve your model while
queries keep running
Consistent meaning across
all your data sources
Relationships are
treated as primary
data structures
Schema grows &
adapts to meet
evolving business
needs
Expressive data
model provides
a common
language
Contextual
= AI-ready data
Knowledge graphs
Data+AI 2025 | www.factor-insights.com
GenAI Applications
LLM Orchestration inc. retrievers, tools & agents
AI & KG Architecture
LLM Platform
Models & Reasoning Engine
Enterprise Data Platform & External Databases
mapping unstructured + structured data
Knowledge Graph
Data+AI 2025 | www.factor-insights.com
Ordinary RAG
Data+AI 2025 | www.factor-insights.com
GraphRAG: RAG with graph
It’s actually quite broad…
GraphRAG is Retrieval
Augmented Generation
(RAG) using Knowledge
Graphs
Improves GenAI by taking
advantage of rich graph
data structures.
Community summaries &
graph enrichments
Graph enhanced vector
search
Dynamic graph query
generation
+ more: graph vectors,
parent-child retrievers, etc.
Retrieval Patterns
Data+AI 2025 | www.factor-insights.com
AI Evolution & Going to Production
Graphs are the next
step – a central data
component for the
AI stack
Generative models
RAG
Agents
Knowledge Graph
GraphRAG
Trust | Performance | Extensibility
Data+AI 2025 | www.factor-insights.com
Language
Statistics
Creativity
Data+AI 2025 | www.factor-insights.com
Knowledge
Facts
Context
Language
Statistics
Creativity
Data+AI 2025 | www.factor-insights.com
Data+AI 2025 | www.factor-insights.com
Benefits of
25
Data+AI 2025 | www.factor-insights.com
GraphRAG
Performance
RAG
Performance
Lettria Analysis1
81.67% 57.50%
Writer Knowledge Graph2
(RobustQA Benchmark)
86.31% 32.74%–75.89%
RAG vs. GraphRAG: Multi-hop
Question Answering3
77% 66%
GenUI Experiments
(MultiHop-RAG Dataset)4
Successfully answered
complex, multi-step queries
Struggled integrating data
from multiple sources
GraphRAG delivers up to
3x higher accuracy than
traditional RAG, with
better multi-hop
reasoning for context-
rich AI applications.
1) https://writer.com/engineering/rag-benchmark/ 2) https://www.lettria.com/blogpost/vectorrag-vs-graphrag-a-convincing-comparison
3) https://arxiv.org/abs/2502.11371 4) https://www.genui.com/resources/graphrag-vs.-traditional-rag-solving-multi-hop-reasoning-in-llms
1.) 3x Higher Accuracy
1
Data+AI 2025 | www.factor-insights.com
2. Ease of Development
Natural language description: “Apples & oranges are both fruits”
2
Data+AI 2025 | www.factor-insights.com
X
Customer
Service
Doctor
Social Security
Number
Social Security
Number
Patient
Bob
Phone
Number
Health
Diagnosis
2. Explainability
3
Data+AI 2025 | www.factor-insights.com
Benefits of GraphRAG
Higher
Accuracy
Easier
Development
Improved
Explainability
Data+AI 2025 | www.factor-insights.com
Data+AI 2025 | www.factor-insights.com
What does this look
like in
30
Data+AI 2025 | www.factor-insights.com
Internal
Documentation
Wikis
Enterprise
Systems
Klarna transforms knowledge access
with GraphRAG
HR Systems
Data+AI 2025 | www.factor-insights.com
Internal
Documentation
Wikis
Enterprise
Systems
Klarna transforms knowledge access
with GraphRAG
HR Systems
Daily queries processed
250K Employee questions
answered in first year
2,000
85% Employee adoption
Data+AI 2025 | www.factor-insights.com
Klarna deploys Neo4j graph
to fuel AI initiatives
What we’ve started using extensively in-house is Neo4j and
graphs. And we’ve also looked at how people collaborate on
building great information.
And then on top of that we have Kiki, who then explores
that information and brings it to life. So we can go and ask
Kiki about anything!
We’re seeing that it’s having a tremendous impact on productivity
internally. So Kiki is basically our own internal chatbot based on that
growing internal knowledge graph.
- Sebastian Siemiatkowski, Co-founder & CEO
Data+AI 2025 | www.factor-insights.com
34
Internal
Forecasts
Analyst Reports
Gaming Data Sales
Subscriptions Marketing
Competitor
Launches
Developmen
t Decisions
Revenue
Promotions
Player
Engagement
Pricing
Strategy
Market
Condition
s
10x faster 92% 150+
Time-to-insight
reduction
Reduction in
analyst time on
routine
requests
Business users
daily
When millions in revenue hang in the balance - knowledge graph powered agents prove invaluable
Gaming Giant Transforms Analytics with
GraphRAG
● Maps structured & unstructured
data from multiple sources
● Semantic context centralized &
controlled in the data layer -
Unmappable with traditional databases
● Speed, scale, traceability & most
importantly trust & confidence in
enterprise AI responses
AI Agents
Data+AI 2025 | www.factor-insights.com
Architectur
e
Documents
Design
Diagrams
Technical
Specifications
VM02 turns docs into design intelligence
Enterprise
Metadata
Data+AI 2025 | www.factor-insights.com
Architectur
e
Documents
Design
Diagrams
Technical
Specifications
VM02 turns docs into design intelligence
Enterprise
Metadata
Time reduction
20%
↑ Driving up productivity
↑ Enhancing quality of deliverables
↑ Enabling faster delivery
Data+AI 2025 | www.factor-insights.com
37
Intelligence
Data
Financial
Data
Unlocking defense and energy insights with GraphRAG
Operational
Records
Data+AI 2025 | www.factor-insights.com
38
Operational
Records
Intelligence
Data
Financial
Data
Market for AI
in energy sector
Analyst workload
reduction
50% $13B
Unlocking defense and energy insights with GraphRAG
FROM DATA TO DECISIONS
Data+AI 2025 | www.factor-insights.com
The Necessity of Graph Databases
Agile Development Query Performance Lower Total Cost of
Ownership
Fast implementation and
iteration with a flexible
schema
High-speed retrieval
across connected data
Better results with
smaller models through
contextual understanding
Real-Time
Transactional & ACID with
access controls, high
availability, durability, etc.
()-[*]->()
Data+AI 2025 | www.factor-insights.com
Graph DB
Agility, Speed, Low TCO & Real-Time
AI Evolution & Going to Production
Graphs are the next step - a central data component for the AI stack
Trust | Performance | Extensibility
Knowledge Graph
Context, Flexibility,
Standardisation
GraphRAG
Accuracy, Explainability,
Governance
Data+AI 2025 | www.factor-insights.com
Data+AI 2025 | www.factor-insights.com
The Future of
AI Adoption is Connected.
How will you harness
Graph + AI to drive
smarter applications?
Data+AI 2025 | www.factor-insights.com
Data+AI 2025 | www.factor-insights.com
Thank you!
Emil Pastor | emil.pastor@neo4j.com
Head of Solutions Engineering, A/NZ
Catch us on the booth area

Graphs & GraphRAG - Essential Ingredients for GenAI

  • 1.
    Data+AI 2025 |www.factor-insights.com Graphs & GraphRAG Essential Ingredients for Enterprise GenAI Thursday | March 27 Factor Data+AI Day Events Centre, Collins Square Melbourne Emil Pastor Head of Solutions Engineering A/NZ Neo4j
  • 2.
    Data+AI 2025 |www.factor-insights.com "Science Finds, Industry Applies, Man Conforms” “Sociotechnical” Principles Motto of the 1933 Chicago World’s Fair
  • 3.
    Data+AI 2025 |www.factor-insights.com AI requires a Holistic Approach1 IT Infrastructure & Management Data Engineering & Quality Management Machine Learning & Analytics Project Management Organisational Change Industry Transformation Regulation Principles & Philosophy (e.g. Ethics) 1 – T. Evgeniou, INSEAD
  • 4.
  • 5.
    But your data islocked in rows and columns
  • 6.
  • 7.
  • 8.
    Data, meet graph Employees Network& Security Suppliers Product Customers Transactions Process
  • 9.
    The power of thegraph model
  • 10.
  • 12.
    Employees Network & Security Services Citizens Transactions Process Suppliers Graphis insightful Employees Network & Security What’s important? Process Suppliers What’s unusual? Services Transactions Citizens What’s next?
  • 13.
  • 14.
    Data+AI 2025 |www.factor-insights.com Generative AI races toward $1.3 trillion in revenue by 2032 Generative AI Revenue Generative AI as a % of Total Technology Spend However, 71% of organizations are stuck piloting GenAI projects 2024 IBM CEO Survey Bloomberg Intelligence
  • 15.
    Data+AI 2025 |www.factor-insights.com 85% of IT leaders cite data quality as their top AI challenge KPMG AI Quarterly Pulse Survey 2025 100 senior executives from $1B+ revenue companies It’s a data management issue! Lack of Standardisation Scale of Query Diversity Data Organization AI needs data organized in ways it can effectively access AI requires high-quality harmonized data to be reliable AI demands rapidly accelerated question-to-query execution
  • 16.
    Data+AI 2025 |www.factor-insights.com Flexible Standardised Enables AI to grasp meaning and access data across sources Supports diverse query types and evolving data needs Maintains consistent understanding at scale What makes data AI-ready? Data exposes entities and relationship understanding Adapts to new patterns, structures and access logic Consistent, governed data foundation Contextual
  • 17.
    Data+AI 2025 |www.factor-insights.com Connected Data Node Property Graph Data Model Person Name:Terry Person Name:Sam Car Brand:Volvo Since:10/Jan/11 DRIVES OWNS SISTER_OF BROTHER_OF Since:10/Feb/11 Relationshi p Property = AI-ready data Knowledge graphs
  • 18.
    Data+AI 2025 |www.factor-insights.com Flexible Standardised Entities and their connections are built into the data model Evolve your model while queries keep running Consistent meaning across all your data sources Relationships are treated as primary data structures Schema grows & adapts to meet evolving business needs Expressive data model provides a common language Contextual = AI-ready data Knowledge graphs
  • 19.
    Data+AI 2025 |www.factor-insights.com GenAI Applications LLM Orchestration inc. retrievers, tools & agents AI & KG Architecture LLM Platform Models & Reasoning Engine Enterprise Data Platform & External Databases mapping unstructured + structured data Knowledge Graph
  • 20.
    Data+AI 2025 |www.factor-insights.com Ordinary RAG
  • 21.
    Data+AI 2025 |www.factor-insights.com GraphRAG: RAG with graph It’s actually quite broad… GraphRAG is Retrieval Augmented Generation (RAG) using Knowledge Graphs Improves GenAI by taking advantage of rich graph data structures. Community summaries & graph enrichments Graph enhanced vector search Dynamic graph query generation + more: graph vectors, parent-child retrievers, etc. Retrieval Patterns
  • 22.
    Data+AI 2025 |www.factor-insights.com AI Evolution & Going to Production Graphs are the next step – a central data component for the AI stack Generative models RAG Agents Knowledge Graph GraphRAG Trust | Performance | Extensibility
  • 23.
    Data+AI 2025 |www.factor-insights.com Language Statistics Creativity
  • 24.
    Data+AI 2025 |www.factor-insights.com Knowledge Facts Context Language Statistics Creativity
  • 25.
    Data+AI 2025 |www.factor-insights.com Data+AI 2025 | www.factor-insights.com Benefits of 25
  • 26.
    Data+AI 2025 |www.factor-insights.com GraphRAG Performance RAG Performance Lettria Analysis1 81.67% 57.50% Writer Knowledge Graph2 (RobustQA Benchmark) 86.31% 32.74%–75.89% RAG vs. GraphRAG: Multi-hop Question Answering3 77% 66% GenUI Experiments (MultiHop-RAG Dataset)4 Successfully answered complex, multi-step queries Struggled integrating data from multiple sources GraphRAG delivers up to 3x higher accuracy than traditional RAG, with better multi-hop reasoning for context- rich AI applications. 1) https://writer.com/engineering/rag-benchmark/ 2) https://www.lettria.com/blogpost/vectorrag-vs-graphrag-a-convincing-comparison 3) https://arxiv.org/abs/2502.11371 4) https://www.genui.com/resources/graphrag-vs.-traditional-rag-solving-multi-hop-reasoning-in-llms 1.) 3x Higher Accuracy 1
  • 27.
    Data+AI 2025 |www.factor-insights.com 2. Ease of Development Natural language description: “Apples & oranges are both fruits” 2
  • 28.
    Data+AI 2025 |www.factor-insights.com X Customer Service Doctor Social Security Number Social Security Number Patient Bob Phone Number Health Diagnosis 2. Explainability 3
  • 29.
    Data+AI 2025 |www.factor-insights.com Benefits of GraphRAG Higher Accuracy Easier Development Improved Explainability
  • 30.
    Data+AI 2025 |www.factor-insights.com Data+AI 2025 | www.factor-insights.com What does this look like in 30
  • 31.
    Data+AI 2025 |www.factor-insights.com Internal Documentation Wikis Enterprise Systems Klarna transforms knowledge access with GraphRAG HR Systems
  • 32.
    Data+AI 2025 |www.factor-insights.com Internal Documentation Wikis Enterprise Systems Klarna transforms knowledge access with GraphRAG HR Systems Daily queries processed 250K Employee questions answered in first year 2,000 85% Employee adoption
  • 33.
    Data+AI 2025 |www.factor-insights.com Klarna deploys Neo4j graph to fuel AI initiatives What we’ve started using extensively in-house is Neo4j and graphs. And we’ve also looked at how people collaborate on building great information. And then on top of that we have Kiki, who then explores that information and brings it to life. So we can go and ask Kiki about anything! We’re seeing that it’s having a tremendous impact on productivity internally. So Kiki is basically our own internal chatbot based on that growing internal knowledge graph. - Sebastian Siemiatkowski, Co-founder & CEO
  • 34.
    Data+AI 2025 |www.factor-insights.com 34 Internal Forecasts Analyst Reports Gaming Data Sales Subscriptions Marketing Competitor Launches Developmen t Decisions Revenue Promotions Player Engagement Pricing Strategy Market Condition s 10x faster 92% 150+ Time-to-insight reduction Reduction in analyst time on routine requests Business users daily When millions in revenue hang in the balance - knowledge graph powered agents prove invaluable Gaming Giant Transforms Analytics with GraphRAG ● Maps structured & unstructured data from multiple sources ● Semantic context centralized & controlled in the data layer - Unmappable with traditional databases ● Speed, scale, traceability & most importantly trust & confidence in enterprise AI responses AI Agents
  • 35.
    Data+AI 2025 |www.factor-insights.com Architectur e Documents Design Diagrams Technical Specifications VM02 turns docs into design intelligence Enterprise Metadata
  • 36.
    Data+AI 2025 |www.factor-insights.com Architectur e Documents Design Diagrams Technical Specifications VM02 turns docs into design intelligence Enterprise Metadata Time reduction 20% ↑ Driving up productivity ↑ Enhancing quality of deliverables ↑ Enabling faster delivery
  • 37.
    Data+AI 2025 |www.factor-insights.com 37 Intelligence Data Financial Data Unlocking defense and energy insights with GraphRAG Operational Records
  • 38.
    Data+AI 2025 |www.factor-insights.com 38 Operational Records Intelligence Data Financial Data Market for AI in energy sector Analyst workload reduction 50% $13B Unlocking defense and energy insights with GraphRAG FROM DATA TO DECISIONS
  • 39.
    Data+AI 2025 |www.factor-insights.com The Necessity of Graph Databases Agile Development Query Performance Lower Total Cost of Ownership Fast implementation and iteration with a flexible schema High-speed retrieval across connected data Better results with smaller models through contextual understanding Real-Time Transactional & ACID with access controls, high availability, durability, etc. ()-[*]->()
  • 40.
    Data+AI 2025 |www.factor-insights.com Graph DB Agility, Speed, Low TCO & Real-Time AI Evolution & Going to Production Graphs are the next step - a central data component for the AI stack Trust | Performance | Extensibility Knowledge Graph Context, Flexibility, Standardisation GraphRAG Accuracy, Explainability, Governance
  • 41.
    Data+AI 2025 |www.factor-insights.com Data+AI 2025 | www.factor-insights.com The Future of AI Adoption is Connected. How will you harness Graph + AI to drive smarter applications?
  • 42.
    Data+AI 2025 |www.factor-insights.com Data+AI 2025 | www.factor-insights.com Thank you! Emil Pastor | emil.pastor@neo4j.com Head of Solutions Engineering, A/NZ Catch us on the booth area

Editor's Notes

  • #9 That's it Nodes Labels Rels Properties No 1200 page book
  • #10 You can connect any nodes with any number, type, and direction of relationships You don’t have to sacrifice fidelity for your data model The tyranny of up front schema and schema migrations is over. Graphs allow organic growth With light-touch constraints where you need strong governance E.g. Bank accounts must be connected to beneficial owner
  • #11 Index free adjacency vs join bomb Query latency proportional to amount of graph explored Not to data set size Not to number of joins And actually there are no joins (at least in Neo4j, it’s just pointers which are the only things computers are really good at) So what can you do with graphs? Loads! <click>
  • #12 Nearly three hundred years of graph theory have given us a myriad ways of gaining insight from structure James Fowler story San Fran 2013 - public health graph - depth 2 Some of the Web hyperscalers (google, facebook, linked in, twitter) even made their business from this.
  • #19 From an architectural standpoint, the way we are seeing this shake out with customers is as follows….
  • #23 LLMs are not a system. They excel in certain probabilistic things. Language, stats, creativity. But you need their complement.
  • #24 Need to ground the LLM in something deterministic, factual To add context and increase accuracy. Better data gives better context that leads to better accuracy. That is the role of a Knowledge Graph
  • #33 This is the CEO of Klarna Ponder that for a minute. Knowledge Graphs are such an important part of their business - especially their GenAI projects - that the CEO is involved. I don’t know if this is a trend, but your CEO already knows about GenAI, will it be long before they start asking you about KGs? I hope not too long 
  • #35 Virgin Media O2's architects were facing a common enterprise challenge - their technical knowledge was scattered across documents, diagrams, and specifications. Traditional AI solutions couldn't make sense of it all. They needed something that could understand complex relationships between their systems and architecture. How different components connect, how changes impact the broader landscape, and how past designs inform new solutions.
  • #36 They built a knowledge graph that mapped their entire technical domain. This gave their AI agents the context to understand how everything fits together. When architects need a new solution design, the system can now understand the full picture, from system dependencies to technical constraints. Design work that used to take weeks can now start with AI-generated drafts. Each design can be traced back to source documentation through the knowledge graph, giving architects confidence in the output. The impact is clear - 20% faster design time and architects rate the output as "great starts." The result is higher quality solution designs that the business can trust and build upon.