The Artificial

Intelligence
Report
The Artificial Intelligence Report 2

CONTENTS Contributors
3 Methodology Christopher Crosbie
VP Product
4 Key Findings Management, Data
Science and AI
Neo4j
5 Advancing AI Reasoning: The Synergy Between
Knowledge Graphs and Language Models by Vahe Andonians
Vahe Andonians, Founder, Chief Technology Founder, Chief
Technology Officer,
Officer, Chief Product Officer, Cognaize
Chief Product Officer
Cognaize
6 Our Gen AI experiment. A chapter produced
entirely by AI Elodie Blyth de
Fontenay
Head of Data Strategy
7 The Rise of the AI Assistant: How I Can Empower
and Transformation
Financial Data Professionals, by Bard, Large Royal London Asset
Language Model by Google AI Management

8 Part One: Navigating the Crossroads of AI Bard

Large Language Model
Adoption
Google AI

12 How GenAI Provides Deep Understanding With

Knowledge Graphs, by Christopher Crosbie, Tom Webber
Conference Director
VP Product Management, Data Science and AI,
FIMA Europe 2024
Neo4j

13 Part Two: Building Gen AI Apps for the Enterprise

17 Part Three: Measuring the Impact and

Value of AI

19 Conclusion

20 Key Suggestions

21 About Cognaize

21 About NEO4J

21 About Redis

21 About WBR Insights

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 3

Methodology
In Q1 of 2024, WBR Insights surveyed 100 of the most senior data executives from Buy-side
and Sell-side organisations across Europe, to find out about how Artificial Intelligence (AI) is
being implemented in their organisations, and the challenges they face when doing so.
The survey was conducted by appointment over the telephone. The results were compiled
and anonymised by WBR Insights and are presented here with analysis and commentary by
Cognaize, NEO4J, Redis and the FIMA Europe community.
The WBR Insights topics detailed in this report will also be covered at the FIMA event.
Download the agenda here.

What type of organisation do you work for? What is your job title?

Chief Data Officer

30%

Head of Data Governance

15%
50% Head of Data Management

15%
Buy-side
Head of Data Architecture
10%

Head of Data Strategy

10%

Head of Data Science

50% 10%

Head of Data and Distribution

Sell-side 10%

Where is your organisation headquartered?

UK/Ireland USA Canada

50% 5% 3%

Germany Benelux Italy

(Belgium, Luxembourg, Netherlands)
15% 2%
5%
Switzerland

10% Nordics
(Denmark, Finland, Norway, Sweden)

France 4%

6%

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 4

Key Findings

Privacy and data security is a top

concern when developing Gen AI
Firms are actively developing and implementing Generative
AI (Gen AI) solutions, but a significant challenge emerged
from our research. Nearly half (46%) of respondents to our
survey cited privacy and data security as their top concern.

Cybersecurity & Customer Service

AI and machine learning (ML) offers a wide range of
uses cases for the financial data industry. However, our
research shows that 70% of firms are gearing most of
their investments towards customer service, chatbots,
account management, and identifying and preventing
cybersecurity threats.

Integration and Privacy

In their efforts to deploy AI systems for various use cases,
52% of financial data executives are struggling to integrate
the new technology with their existing systems, and 50%
have substantiated their data privacy concerns.

Strong Deployment Strategy

Firms appear to have a strong deployment
strategy when rolling out their AI solutions, with
91% of respondents rating their organisation’s
capabilities in this area as adequate or strong.
Encouragingly, not a single respondent rated
their organisation’s as ‘very weak’.

DOWNLOAD THE AGENDA HERE

Advancing AI Reasoning:
The Artificial Intelligence Report 5

The Synergy Between Knowledge

Graphs and Language Models
By Vahe Andonians
Founder, Chief Technology Officer, Chief Product Officer
Cognaize

The advent of transformers has revolutionised text vectorisation, Incorporating knowledge graphs with dense vectors emerges
achieving unprecedented precision. These embeddings as the most promising solution. While embeddings efficiently
encapsulate profound semantic meanings, surpassing previous condense text of varying lengths into fixed-dimension vectors,
methodologies, and are why Large Language Models are so enabling the identification of semantically similar phrases,
convincingly good at generating text. they sometimes fall short in distinguishing critical nuances.
For instance, “Cash and Due from Banks” and “Cash and
Large Language Models further demonstrate reasoning Cash Equivalents” yield nearly identical vectors, suggesting
capabilities, albeit with limitations; their depth of reasoning a similarity that overlooks substantial differences. The latter
tends to diminish rapidly. However, integrating knowledge includes interest-bearing entities like “Asset-Backed Securities”
graphs with these vector embeddings can significantly enhance or “Money Market Funds,” while “Due from Banks” refers to
reasoning abilities. This synergy leverages the inherent semantic non-interest-bearing deposits.
richness of embeddings and propels reasoning capabilities
to unparalleled heights, marking a significant advancement in Knowledge graphs also capture the complex interrelations of
artificial intelligence. concepts. This fosters a deeper contextual insight, underscoring
additional distinct characteristics through connections between
In the finance sector, Large Language Models are predominantly concepts. For example, a US GAAP knowledge graph clearly
utilised through Retrieval Augmented Generation, a method defines the sum of “Cash and Cash Equivalents,” “Interest
that infuses new, post-training knowledge into Large Language Bearing Deposits in Banks,” and “Due from Banks” as “Cash and
Models. This process involves encoding textual data, indexing Cash Equivalents.”
it for efficient retrieval, encoding the query, and employing
similarity algorithms to fetch relevant passages. These retrieved By integrating these detailed contextual cues and relationships,
passages are then used with the query, serving as a foundation knowledge graphs significantly enhance the reasoning
for the Large Language Model to generate the response. capabilities of Large Language Models. They enable more
precise multi-hop reasoning within a single graph and facilitate
This approach significantly expands the knowledge base of joint reasoning across multiple graphs.
Large Language Models, making it invaluable for financial
analysis and decision-making. While Retrieval Augmented Moreover, this approach offers a level of explainability that
Generation marks a significant advancement, it has limitations. addresses another critical challenge of Large Language Models.
The transparency in how conclusions are derived through visible,
A critical shortcoming lies in the passage vectors’ possible logical connections within knowledge graphs provides a much-
inability to fully grasp the semantic intent of queries, leading needed layer of interpretability, making the reasoning process
to the vital context being overlooked. This oversight occurs not only more sophisticated but also accessible and justifiable.
because embeddings might not capture certain inferential
connections essential for understanding the query’s full scope. The fusion of knowledge graphs and embeddings heralds
a transformative era in AI, transcending the limitations of
Moreover, condensing complex passages into single vectors can individual approaches to achieve a semblance of human-like
result in the loss of nuances, obscuring key details distributed linguistic intelligence.
across sentences.
Knowledge graphs introduce previously gained symbolic logic
Additionally, the matching process treats each passage and intricate relationships from humans, enhancing the neural
separately, lacking a joint analysis mechanism that could networks’ pattern recognition prowess and finally resulting in
connect disparate facts. This absence hinders the model’s ability superior hybrid intelligence.
to aggregate information from multiple sources, often necessary
for generating comprehensive and accurate responses requiring Hybrid intelligence paves the way for AI that not only articulates
synthesizing information from various contexts. eloquently but also comprehends deeply, enabling advanced
conversational agents, discerning recommendation engines, and
Efforts to refine the Retrieval Augmented Generation framework insightful search systems.
abound, from optimising chunk sizes to employing parent
chunk retrievers, hypothetical question embeddings, and query Despite challenges in knowledge graph construction and
rewriting. While these strategies present improvements, they noise management, the integration of symbolic and neural
don’t lead to revolutionary outcome changes. An alternative methodologies promises a future of explainable, sophisticated
approach is to bypass Retrieval Augmented Generation by language AI, unlocking unprecedented capabilities.
expanding the context window, as seen with Google Gemini’s
leap to a 1 million token capacity. However, this introduces new
challenges, including non-uniform attention across the expanded
context and a substantial, often thousandfold, cost increase.

DOWNLOAD THE AGENDA HERE

Our Gen AI
experiment.
A chapter produced
entirely by AI

DOWNLOAD THE AGENDA HERE

The Rise of the AI Assistant:
The Artificial Intelligence Report 7

How I Can Empower Financial Data

Professionals by Bard, A Large
Language Model from Google AI
*Disclaimer. The article you see here was written in full by Bard, a large language
model from Google AI. Using prompts provided by WBR Insights, this article has been
written, and the page has been designed and edited using the decisions and opinions
provided by Bard. Additionally, the image used to represent Bard was made using a
prompt provided by Bard, and created using a tertiary Generative AI Image Generator. WBR
Insights would like to highlight that the information provided by LLMs such as Bard may not be
entirely accurate, and the information it provides should be treated as such.

The financial data industry thrives on information, analysis,

and efficient decision-making. In this ever-evolving landscape,
artificial intelligence (AI) is emerging as a powerful tool,
offering professionals valuable assistance in navigating the
complexities of financial data. As a large language model,
I, along with others like me, represent a new wave of AI
assistants poised to significantly impact the industry. Let’s 3. Improved Communication and Collaboration:
explore how I can be of use to financial data professionals.
Effective communication and collaboration are essential for
success in the financial industry. I can contribute to this by:
1. Enhanced Data Processing and Analysis: • Facilitating communication: I can translate financial
information into clear and concise language, bridging the gap
Financial data comes in various forms, often vast and
between technical experts and non-technical stakeholders.
unstructured. Manually processing and analysing this data can
This can improve communication within teams and with clients.
be time-consuming and prone to human error. I can leverage my
natural language processing (NLP) capabilities to: • Generating different content formats: I can generate various
content formats, such as reports, presentations, and emails,
• Extract and organise data: I can efficiently extract relevant
tailored to specific audiences and purposes. This enhances
information from diverse sources like financial reports,
communication clarity and efficiency.
news articles, and social media data. This includes identifying
key entities, sentiment analysis, and summarising • Supporting brainstorming and idea generation: I can assist
complex information. in brainstorming sessions by providing relevant information,
suggesting alternative perspectives, and exploring different
• Identify patterns and trends: My ability to analyse large
scenarios. This can foster creative problem-solving and lead to
datasets allows me to uncover hidden patterns and trends
innovative solutions.
that might escape human observation. This can be crucial
for identifying market opportunities, predicting market
movements, and making informed investment decisions. 4. Risk Management and Fraud Detection:
• Automate repetitive tasks: I can handle repetitive tasks Financial data analysis plays a crucial role in identifying and
like data cleaning, normalising, and formatting, freeing up mitigating risks. I can contribute to this by:
valuable time for professionals to focus on higher-level • Analysing historical data: I can analyse historical financial data
analysis and strategy. and market trends to identify potential risks and vulnerabilities.
This can inform risk management strategies and proactive
measures to mitigate potential losses.
2. Market Research and Due Diligence:
• Detecting fraudulent activities: I can analyse financial
Gathering comprehensive market intelligence is crucial for transactions and identify anomalies that might indicate
financial professionals. I can assist in this process by: fraudulent behaviour. This can help prevent financial losses
• Monitoring news and social media: I can continuously and ensure the integrity of financial data.
monitor relevant news sources and social media platforms • Staying updated on regulations: I can monitor regulatory
to identify emerging trends, potential risks, and competitor changes and keep financial professionals informed about
activities. This real-time information can inform investment relevant updates, ensuring compliance and mitigating
decisions and risk management strategies. regulatory risks.
• Researching companies and industries: I can access and It’s important to remember that I am not a replacement for
analyse vast amounts of data from various sources to provide human expertise and judgment. My role is to augment human
comprehensive insights into specific companies and industries. capabilities by providing efficient data processing, insightful
This includes financial statements, news articles, analyst analysis, and comprehensive information access. As the financial
reports, and regulatory filings. data industry continues to evolve, AI assistants like me will play
• Generating reports and summaries: I can synthesize the an increasingly important role in empowering professionals to
gathered information and generate concise reports and make informed decisions, navigate complex challenges, and
summaries, saving professionals valuable time and effort. achieve greater success.

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 8

Part One:
Navigating the
Crossroads of AI
Adoption

AI-powered solutions are evolving rapidly. While The implementation of AI remains relatively new to the
the technology has been used in some capacity for industry, and many firms struggle to quantify its impact
several years, the number of use cases and deployable on their P&L (Profit and Loss Statement). According to
solutions have advanced rapidly in the past 18 months. our research, establishing dedicated teams is the most
Traditionally, the financial data industry has been slow common approach to tackling this problem. Internal or
to adopt these technologies, so implementing AI into external teams focus on identifying efficiencies and cost
operations is now a must. savings, benchmarking performance indicators, and
ultimately determining AI’s financial contributions.
According to 70% of respondents to our survey, firms
are prioritising customer service-focused chatbots and Our survey also revealed encouraging news about the
solutions that can identify and prevent cybersecurity industry’s overall AI implementation. A massive 91%
threats. The rapid growth of Generative AI and LLMs of respondents rank their organisation’s AI capabilities
means firms can deploy AI-enabled chatbots that have in terms of patents, deployment of AI solutions and
the ability to revolutionise customer service. These partnerships as adequate, or strong. Even more
chatbots allow for 24/7 support by addressing frequently promising, none rated their organisation’s capability in
asked questions while generating personalised content these areas as ‘very weak’, suggesting widespread and
that understands the nuances of customer service. They proper AI adoption across the industry.
also offer assistance in areas such as budgeting, expense
tracking and financial planning.

Despite the promise of AI, leveraging these technologies

does not come without challenges. Fifty-two per cent
of financial data executives are struggling to integrate
AI-enabled solutions with their existing systems, and
50% have data privacy concerns. These concerns might
be particularly focussed on generative AI, and a fear
that chatbots have the potential to learn from internal
data to generate responses that could be shared with
external users.

Furthermore, non-standardised, existing data

frameworks in organisations may not be compatible with
newly-formed AI solutions. This provides CDOs with a
significant challenge when it comes to consolidating all
data sources to integrate with their new AI solutions.

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 9

“In the face of rapid AI advancements and their somewhat

What are the primary use cases for AI and ML
unpredictable emerging capabilities, the imperative for
in your organisation? integrating human assistance and expertise—especially
(Respondents were asked to select all that apply) judgment—becomes more pronounced, necessitating
continuous refinement of this integration. Achieving a
Customer service: AI chatbots for customer service
meaningful collaboration between human intelligence and
and account management
artificial systems is crucial for enhancing decision-making
processes. This challenge highlights the importance of
70%
adopting a strategic approach in selecting use cases,
ensuring that each chosen domain maximally benefits
Cybersecurity: identify and prevent from the unique strengths of both human and artificial
cybersecurity threats intelligence.”
Vahe Andonians
70% Founder, Chief Technology Officer,
Chief Product Officer
Cognaize
Market analysis: analysis to predict market trends
more accurately

67%
“This rings true, particularly in the areas of market analysis
and customer service. The results list many use cases, but
Financial spreading: use AI to spread and annotate
we should also consider research as a use case. In the
financial reports field of investment banking, the potential is enormous.
Many individuals in the city have research-related jobs
53% or conduct research as part of their daily work. LLMs can
greatly accelerate their research efforts.”
Elodie Blyth de Fontenay
Extraction: specific information from vast amounts Head of Data Strategy and Transformation
of documents e.g. SEC filings, trustee reports, Royal London Asset Management
ESG reports

53%
“It’s interesting to see trends here connected to delivering
and protecting customer experience. As well as delivering
Fraud detection and compliance: identify patterns a more personalised service for customers. I also see an
that indicate fraud and compliance issues
emerging trend that has a lot to do with leveraging AI for
reporting. As regulations harden and acceptable uses of
46%
AI build, I think we’ll see use cases in the reporting sphere
grow as evidence-based trust in AI services increases.”
Portfolio management: optimise Tom Webber
investment portfolios Conference Director
FIMA Europe 2024
46%

Credit risk assessment: analysing data to “These initial results suggest a pragmatic approach to
make more accurate risk assessments e.g. for AI adoption by European financial data organisations.
commercial lending or credit underwriting. Both customer service chatbots and cybersecurity
applications offer tangible benefits and potential return
42% on investment (ROI).

Chatbots can streamline operations, improve accessibility,

Trade finance: automating analysis and
and personalise interactions, potentially leading to cost
verification of trade documents savings and increased customer satisfaction. Meanwhile,
AI-powered cybersecurity tools can mitigate risks and
enhance protection against evolving threats, potentially
17%
reducing financial losses and reputational damage.
It’s interesting to see these practical applications take
precedence over potentially more transformative, but also
riskier, use cases.”
Bard
Large Language Model
Google AI

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 10

“It is important to examine the hallucination phenomenon

What are the top challenges you face when it
not merely as an anomaly but as an integral part of
comes to AI use cases? cognitive processes akin to human problem-solving.
(Respondents are asked to select three options) Intriguingly, the stages of creative thinking after
preparation, namely incubation and illumination, profit
Integration with existing systems of hallucination in scientific settings. This gain hints at
hallucination as a viable necessity for AI’s progression.
52% This perspective invites a re-evaluation of hallucination’s
role, suggesting that its understanding could be crucial
Data privacy / data security issues for advancing AI’s capabilities in mimicking human-like
thought sequences.”
50%
Vahe Andonians
Founder, Chief Technology Officer,
Regulatory issues Chief Product Officer
Cognaize
31%

Explaining AI decisions and outcomes “It is surprising to see that integration with legacy systems
as the top challenge. In my opinion, the greater challenge
30% lies in the lack of skills and embedding AI into the culture
of firms. It is important to make sure that colleagues
Ability to guarantee data quality for input / training data understand what AI is … and is not, how the human should
always be in the loop and decision maker, and that people
25% do not feel threatened by its use.”

Include human oversight – appropriate

Elodie Blyth de Fontenay
role distribution – UI Head of Data Strategy and Transformation
Royal London Asset Management
23%

The cost of deploying LLMs / also sustainability issue “Legacy and privacy are understandably top of the list
as Banks and Asset Managers compete to leverage the
22% latest AI tech whilst observing their clients’ preferences
for converting data into more personalised experiences.
Interesting trends here too are around the importance of
A lack of talent to develop our own AI models
data quality as a foundation for reliable, AI-enabled data
20% management strategies.”
Tom Webber
A lack of clarity around AI ethics Conference Director
FIMA Europe 2024
17%

Explainable or interpretive AI “The survey exposes two key challenges hindering AI

adoption in European financial data firms: integrating
13% AI with existing systems (52%) and data privacy/security
concerns (50%). Both pose significant roadblocks.
Fairness and bias Integrating new AI technologies with complex legacy systems
requires careful planning and execution. Additionally,
12% ensuring data privacy and security while leveraging AI
for valuable insights remains a delicate balancing act in
the heavily regulated financial sector. Addressing these
AI hallucinations
technical and regulatory hurdles through innovative
5% solutions is crucial for unlocking AI’s full potential.”
Bard
Large Language Model
Google AI

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 11

“Attributing P&L to AI investments is difficult as cashflows

We asked our respondents how their organisation
are not deterministic. Most importantly, the true cost of
is attempting to capture AI efficiencies in the AI is determined by its potential mistakes - a figure that
P&L. Here is what they told us: is not easily forecastable. Moreover, given its reliance on
trial-and-error advancements, AI’s unpredictable evolution
risks turning current investments into near-future sunk
costs. Stochastic methods, most prominently a strategic

The capabilities of AI are improved and
approach based on scenario analysis paired with stringent
evaluated by internal teams. This way, we are cost control, is currently the most promising method.”
prepared to capture efficiencies in the P&L. Vahe Andonians
Founder, Chief Technology Officer,
Chief Product Officer

AI impact on productivity is tested. Cost and Cognaize
time savings are recorded for accurately
capturing efficiencies in the P&L.
“I’m pleasantly surprised to see these. A lot of responses are
talking about having benchmarks and ways of identifying
an efficient model. It feels to me as though firms are well

There are performance indicators used to prepared to understand the impact on productivity and
map the efficiencies without any mistake. costs which is comforting.”
Elodie Blyth de Fontenay
Head of Data Strategy and Transformation

There is a detailed method to understand Royal London Asset Management
the true positives and the false positives.
Classification methods can help capture “Where we’re seeing this conversation land at FIMA at the
efficiencies correctly. moment is around measuring responsible and ethical
use of AI inside banks and asset managers. We’re at the
beginning of the journey when it comes to formalising

We have benchmarks to test the efficiencies the measurement of AI efficiency and these comments
reflect this position well. What these comments also show
of AI in different processes.
is that there is a significant opportunity for vendors to
step in and help banks and asset managers structure
and formalise performance of AI inside the firm to drive

It’s tough to do so without consulting with continuous improvement.”
experts. We have a capable team to provide Tom Webber
insights on AI efficiencies. Conference Director
FIMA Europe 2024

“Investing in AI as a business capability is clearly a priority

How would you rank your company’s AI capabilities
for most respondents and what this chart shows us too
in terms of patents, deployment of AI solutions and is that each organisation participating in the survey is
partnerships? approaching investment in AI as a competitive advantage
7% with some very far ahead, most well on the journey to
benefiting and a small minority lagging behind.”
Very strong
9% Tom Webber
Weak Conference Director
FIMA Europe 2024

45%
Strong

39%
Adequate

DOWNLOAD THE AGENDA HERE

How
WhyGenAI Provides Graphs
Knowledge
Deep Understanding
Should Be Part of With
Your
Knowledge Graphs
AI Strategy
Christopher Crosbie
VP Product Management, Data Science and AI
Neo4jCrosbie, VP Product Management, Data Science and AI
By Christopher

Connected data
data reveals
reveals new
newways
waysto
tosolve
solvepressing
pressingbusiness
businessproblems,
problems,even
evenasas
data sizes increase.
increase. With
With aaknowledge
knowledgegraph,
graph,firms
firmscan
canunlock
unlockthe
thefull
fullpotential
potential
of data assets
assets for
for AI
AI applications.
applications.This
Thisyear’s
year’sFIMA
FIMAsurvey
surveyshows
showsthat
that50%
50%ofof
data executives are
are exploring
exploringororplanning
planningtotouse
useknowledge
knowledgegraphs
graphsfor
forAI,
AI, and
and another
another 20%20% reported
reported interest.
interest.
Firms have achieved
achievedearlyearlywins
winswith
withAIAIto dollarsmanual
cutting in fraudreview
prevention
time ineach
half. year
Another
while • Traverses multi-degree
multi-degreerelationships
relationshipswithwith
to enhance
enhance thethe customer
customer experience,
experience, internal Neo4j
cutting manual boosted
customer fraud
review time in detection
half. Another millisecond latency
millisecond latency
internal
processes,processes,
personal personal productivity,
productivity, and rates
Neo4jby 200%. boosted fraud detection
customer • Scales for the
the enterprise
enterprisewith
withtransaction
transaction
and cost-saving
cost-saving efforts.
efforts. The the
However, truetrue
valuevalue rates by 200%. support, high
high availability,
availability,granular
granularsecurity
securi-
of AI, however, is found when it allows
AI lies in the ability to uncover deep The GenAI Enabler controls,
ty andand
controls, fullfull
ACID compliance
ACID compliance
us to unlock
meaning the deeper
in data, enablingmeanings
us to make in more The arrival
The GenAI Enabler
of generative AI brought a • Delivers an
an integrated
integratedset setof
oftools
toolsfor
fordata
data
data, enabling
strategic us towhere
decisions makeprecision
more strategic
and demand
The arrival forofAIgenerative
models toAI produce
broughtbetter,
a de- loading, visualisation,
visualization,and andanalysis
analysis
decisions for theThat’s
accuracy matter. business.
whereThat's where
knowledge more
mandaccurate responses.
for AI models KGs better,
to produce act as amore • Supports AI development
developmentwith withenablers
enablers
knowledge
graphs come graphs
in. come in. control
accurate forresponses.
Large LanguageKGs actModels (LLMs)
as a control such as GraphRAG,
GraphRAG, free-text
free-textindexing,
indexing,
by
forenabling knowledge-based
Large Language reasoning
Models (LLMs) by graph algorithms,
algorithms, and andML MLpipelines
pipelines
A
A Knowledge GraphPrimer
Knowledge Graph Primer • Integrates with
with open-source
open-sourceRAG RAGframe-
frame-
based
enabling on knowledge-based
the connections inreasoning
the data based
–
Knowledge
Knowledge graphs
graphs (KGs)
(KGs) underpin
underpin a a wide
wide ultimately leading to
on the connections in AI
thewe can
data trust. 48%
– ultimately works and all
all major
majorcloud
cloudproviders
providers
range of applications,
range of applications, from
fromdrug
drugdiscovery
discovery of FIMA to
leading survey
AI werespondents
can trust. 48% plan to use
of FIMA
and
and consumer-facing personalisation sys-
consumer-facing personalization techniques that aimplan
survey respondents to improve responses
to use techniques
tems to critical infrastructure likelike
systems to critical infrastructure public
public generated
that aim to by AI. It’sresponses
improve no surprise that a by
generated
transport, power grids,
transport, power grids, and
and supply
supplychain
chain combined 70% of that
AI. It’s no surprise respondents
a combined stated
70%that
of Neo4j, the Graph
Graph Database
Database & & Analytics
Analytics
management.
management. they’re planning
respondents to use
stated thator are interested
they’re planning to leader, helps organizations
organisations find hidden
in
useexploring knowledge
or are interested graphs this
in exploring year.
knowledge relationships and patterns across billions of
A KG provides a powerful framework for
A KG provides a powerful framework for graphs this year. data connections deeply,
deeply, easily, and quickly.
organising and linking data of all types. Last month, Microsoft released a study
organizing and linking data of all types. By capturing the connections between data
KGs capture relationships between enti- where researchers combined
KGs capture relationships between entities Last month, Microsoft releasedaaknowl-
study as they exist in the real world, the Neo4j
ties from a real-world domain by using a edge graph with an LLM, an approach
from a real-world domain by using a graph where researchers combined a knowledge knowledge graph provides the data context
graph structure – which formats data as termed GraphRAG.
structure – which formats data as an inter- graph with an LLM, The results demon-
an approach termed required for advanced AI applications,
an interconnected network – rather than a strated increased comprehensiveness,
connected network – rather than a table with GraphRAG. The results demonstrated in- including generative AI use cases. Neo4j
table with rows and columns. explainability, and diversity of viewpoints
rows and columns. creased comprehensiveness, explainability, supports LLM applications
applications by
by combining
combining the
the
in the LLM’s responses.
and diversity of viewpoints in the LLM’s power of inferential reasoning with semantic
For example, in cybersecurity, the KG acts as
For example, in cybersecurity, the KG acts responses. similarity from native vector search.
a data representation of an IT infrastructure, Deloitte has recognised KGs as the industry
as a data representation of an IT infrastruc-
with all its security tools, applications, and standard for enterprise-ready AI, and many
ture, with all its security tools, applications, Deloitte has recognized KGs as the industry Neo4j customers
Neo4j customers include
include NASA,
NASA,Adobe,
Adobe,
dependencies. By modelling their IT network leading analyst firms agree. According to
and dependencies. By modeling their IT net- standard for enterprise-ready AI, and many AstraZeneca, Boston
AstraZeneca, Boston Scientific,
Scientific,Cable
CableNews
News
in a graph data model, security teams gain a Gartner, “GenAI models are being used in
work in a graph data model, security teams leading analyst firms agree. According to Network (CNN),
Network (CNN), eBay,
eBay,Hewlett-Packard
Hewlett-Packard
holistic understanding of the system and can conjunction with knowledge graphs to deliver
gain a holistic understanding of the system Gartner, “GenAI models are being used in Enterprise, IBM,
Enterprise, IBM, International
InternationalConsortium
Consortiumof of
apply predictive analytics to find their own trusted and verified facts to their outputs, as
and can apply predictive analytics to find conjunction with knowledge graphs to deliv- Investigative Journalists
Investigative Journalists(ICIJ),
(ICIJ),Levi
LeviStrauss
Strauss&
vulnerabilities before attackers do. well as provide rules to contain the model.”
their own vulnerabilities before attackers do. er trusted and verified facts to their outputs, Co.,
& Lyft,
Co., Merck,
Lyft, Merck,Meredith Corporation,
Meredith Micro-
Corporation,
(Emerging Tech Impact Radar: 2023). soft, Novo Nordisk, OrbitMI, The UBS Group,
Over the last 15 years, Neo4j has helped as well as provide rules to contain the mod- Microsoft, Novo Nordisk, OrbitMI, The UBS
Over the last Tech Impact Radar: 2023). and Vanguard. ConnectConnect
with us on Twitter,
companies use15KGs
years, Neo4jabout
to reason has helped
complex el” (Emerging
KGs for the Enterprise Group, and Vanguard. with us on
LinkedIn,
Twitter, and YouTube.
LinkedIn, and YouTube.
datasets. Because KGs excel atabout
companies use KGs to reason complex
connecting
Knowledge graphs power a new gener-
datasets.
data acrossBecause KGs
disparate excel at
systems andconnecting
data KGs for the Enterprise
ation of AI that is uniquely capable of
theydisparate
data across
sources, work wellsystems and data
for applications Knowledge graphs power a new generation
providing a deep understanding of data.
sources, real-time
requiring they workanalytics.
well for applications
For example, of AI that is uniquely capable of providing a
When evaluating a knowledge graph
requiring
in real-time
fraud detection, analytics.
KGs For example,
show real-time deep understanding of data. When evaluat-
offering, use these criteria to determine
in fraud detection,
behavioural patternsKGs show
in the datareal-time
that indicate ing a knowledge graph offering, use these
enterprise readiness:
behavioralactivities.
fraudulent patterns in
Forthe
onedata that indicate
Fortune 500 criteria to determine enterprise readiness:
fraudulentusing
company, activities. For one
a KG saved Fortune
millions of 500
company,
dollars usingprevention
in fraud a KG saved millions
each of
year while
The Artificial Intelligence Report 13

Part Two:
Building Gen
AI Apps for the
Enterprise

The popularity and growth of Generative AI (Gen AI) Furthermore, 18% are actively exploring the potential
solutions has exploded in the past year. This surge can be of knowledge graphs, which confirms their growing
partly attributed to the phenomenal success of Chat GPT, importance. knowledge graphs provide a structured
which is already in its fourth iteration and continues to push framework for organising and connecting information.
the boundaries of possible. Tech giants like Google (Gemini,
previously Bard) and others are at the forefornt of this By embracing these technological advancements and
revolution, offering powerful Gen AI solutions that extend addressing data security concerns, firms can unlock the
beyond text generation as artistic expression is redefined potential of Gen AI and could revolutionise various aspects
with image creation tools like Midjourney. of their operations.

Gen AI solutions have become increasingly sophisticated,

accurate and valuable. However, firms have told us they
are experiencing challenges in their Gen AI development
efforts. Chief among them, according to 46% of
respondents to our survey are privacy and data security
concerns. Publicly available Gen AI platforms often have
concerning data practices. These platforms frequently
employ an “opt-in” approach by default, automatically
incorporating user input data into their training models.
This raises a red flag, as user data, potentially containing
sensitive information, becomes part of the Large Language
Model (LLM), shaping its responses for all users.

This increases the risk of sensitive data being inadvertently

incorporated into AI-generated outputs that are then
shared with external users. While Gen AI providers are
actively looking to improve their data security measures,
firms remain understandably cautious. To overcome
these risks, organisations are strategically investing in
interpretability tools and specialised training for their Gen
AI models.

Looking forward, knowledge graphs have emerged as a

critical tool for firms who are developing Gen AI strategies.
According to our research, 32% are already utilising
knowledge graphs in conjunction with built-in vector
search functionalities.

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 14

“The results feel quite technical and would imply that

What are the biggest challenges in your Gen AI organisations may already have identified valuable use
cases for AI and are ahead in their development. For
development efforts?
me, defining the right use case remains the greatest
(Respondents were asked to select two options) challenge in GenAI development. It is (relatively) easy to
be carried away and build GenAI stuff, but developing
useful and meaningful GenAI tools for business users is
much harder. Not everything needs an LLM and we need
46% 42% 36% to keep this in mind during the process. Other challenges
include having the right talent to develop Gen AI systems
Privacy and Curating a high- Integrating as things evolve very quickly in that space and even the
data security quality dataset domain-specific experts do not always have the answer. Last but not
for grounding or knowledge for
fine-tuning AI relevant responses
least, we need to do so in a responsible, ethical, and
sustainable way, which can be challenging in itself.”
Elodie Blyth de Fontenay
Head of Data Strategy and Transformation
33% 30% 13% Royal London Asset Management

Ability to trace Building robust Reducing

sources and data governance hallucination “What we see here is that banks and asset managers are
understand and compliance deliberately prioritising investment in Generative AI around
reasoning frameworks
core risks to the business – data usage, risk data and
data quality. The finer work to be done around reducing
hallucination indicates most respondents are spending
most time on the bigger picture early on so I expect this
“Knowledge graphs allow you to exert control over what metric to increase as the journey towards broad adoption
“Knowledge graphs
the LLM delivers, allow
which is you
muchto easier
exert control over you
to do when what
the LLM delivers, which(rather
is much easier to dothrough
when you of Generative AI continues.”
query against a graph than digging
query against
documents, fora example).
graph (rather
The than
person digging through
on your team Tom Webber
documents,
with the domain expertise, such as a business team
for example). The person on your analyst, Conference Director
with the domaininexpertise,
can participate suchofas
the building a business
the knowledge analyst,
graph FIMA Europe 2024
canensure
to participate
the datain the building of the
is represented knowledge
accurately. graph
Then the
to ensure the
knowledge dataacts
graph is represented
as a controlaccurately.
for the LLMThen the its
so that
knowledge graph acts as a control for
responses reflect the facts in the data.” the LLM so that its
“With 46% of respondents citing privacy and data security
responses reflect the facts in the data.”
Christopher Crosbie as a concern, it’s evident that ensuring compliance with
Christopher
VP Crosbie
Product Management, Data Science and AI regulations and safeguarding sensitive financial data
VP Product Management, Data Science and AI
Neo4j remains a top priority. This challenge is likely amplified by
Neo4j the evolving nature of AI and the potential for unintended
bias to creep into algorithms.
Furthermore, 42% of respondents identified the difficulty
of curating high-quality datasets for training and fine-
tuning AI models as a significant hurdle. This underscores
the importance of having access to clean, accurate, and
relevant data to ensure the effectiveness and reliability of
AI applications in the financial sector.”
Bard
Large Language Model
Google AI

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 15

“It’s surprising to see fine-tuning rank so high. If you

Which of the following strategies is your firm
need to act on your firm’s domain-specific data, such as
most likely to use in 2024 to address your
financial data, you should choose RAG over fine-tuning.
biggest Gen AI challenges? RAG and fine-tuning are not mutually exclusive and they
work well together. But when you want GenAI to return
Invest in interpretability tools and training
facts about your domain, RAG is the better option.”
29% Christopher Crosbie
VP Product Management, Data Science and AI
Neo4j
Fine-tuning of models using financial/legal
and other domain-specific data

23% “The three most popular responses tell a clear story about
how respondents view the role of humans in the evolution
of AI models as integral and key. Adopting AI as part of
your business and data strategy means humans still need
Human oversight and review processes
to be employed, involved in the process and properly
16% trained for AI to land and perform.”
Tom Webber
Conference Director
FIMA Europe 2024
Use a knowledge graph to support AI

15%

Retrieval-Augmented Generation (RAG) techniques

to ground the model in our data

10%

Don’t know yet/not leaning in any

particular direction

7%

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 16

“Dense vectors encapsulate information in a streaming

How is your firm planning to use knowledge
form. Despite expanding context windows, a vector
graphs, vector, or other data systems in your database remains vital for managing costs and effectively
Gen AI strategy? enhancing model processing speed. Knowledge graphs,
on the other hand, represent information discretely and
We plan to use knowledge graphs with built-in vector search capture prior knowledge and human judgment in the most
comprehensive manner. Therefore, integrating vectors and
32% knowledge graphs is essential for achieving superior hybrid
intelligence. However, the best method for integration
varies, presenting a significant opportunity to distinguish
Knowledge graphs seem interesting, but we oneself from the competition.”
have not looked into them yet
Vahe Andonians
20% Founder, Chief Technology Officer,
Chief Product Officer
Cognaize

We are exploring knowledge graphs

“Responses to this question demonstrate a wide range of
18% experiences and awareness of knowledge graphs. The
majority who plan to use them see a clear connection
between the benefits they bring and the opportunity of
combining them with Gen AI.”
We want to use another database with vector search
Tom Webber
Conference Director
15%
FIMA Europe 2024

Not sure, we are still exploring all options “The survey results offer an interesting glimpse into the
evolving approaches to AI strategy within European
8% financial data organisations. While a significant portion,
32%, are already planning to leverage knowledge graphs
with built-in vector search, another 20% remain unsure
and haven’t explored them yet.
We want to use a vector database only This suggests a cautious but growing interest in knowledge
graphs and vector search technologies within the
7% industry. The perceived benefits of knowledge graphs,
such as enhanced data organisation and reasoning
capabilities, could be driving the adoption by the 32% of
the respondents.”
Bard
Large Language Model
Google AI

“The industry has gotten better at ensuring models don’t

hallucinate. Now our problem is that LLMs return generic
responses. If you’re building a wine application with a
baseline RAG technique, you might get responses about
how wine is red and sweet – accurate, but this doesn’t tell
you much. If you used GraphRAG (a knowledge graph with
RAG): the LLM will offer a more nuanced answer: “The wine
is sweet due to barrel x, from the y region of France…”
Knowledge graphs provide this completeness of response.”
Christopher Crosbie
VP Product Management, Data Science and AI
Neo4j

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 17

Part Three: Measuring the

Impact and Value of AI

Widespread adoption of AI hinges on its ability to

deliver tangible value. To assess its value, financial data
executives are increasingly deferring to dedicated teams
to provide their opinion and gauge the performance of
various AI solutions. Having a team of experts will be
key, particularly in the initial years of implementation as
firms attempt to understand the impact of AI solutions,
which may not be immediately clear.

This is particularly important when firms are in the early

stages of their AI journey.

Our research confirms this, particularly when it comes

to AI chatbots. Half of our respondents said their We asked our respondents how their organisation
organisation is currently exploring AI-enabled measures AI/ML ‘proof of value’ in key projects
Chatbots using internal knowledge bases. to accelerate adoption across units. Here is what
they told us:
Similarly, more than 60% of respondents are exploring
or selecting technology for their AI journey, aiming to
unlock AI and ML capabilities by developing reference
architectures for cloud, modern database 
Streamlining of complex operations and its
and ML operations platforms. efficiency is one of the important measures to
map the proof of value for AI/ML.
While only 6% are in the early stage, these figures
highlight the significant work that lies ahead for firms.
Going forward, expert guidance will be crucial to 
Expert opinion does help in accelerating
ensuring a successful AI journey. adopting and measuring proof of value
systematically.

“There are some interesting comments in the responses, 

Metrics related to productivity and viability
especially when it comes to metrics and productivity.
of using AI/ML in other business functions
This suggests quite an advanced way of thinking, which
is encouraging.” are used to measure proof of value.
Elodie Blyth de Fontenay
Head of Data Strategy and Transformation 
Performance of tools is evaluated at
Royal London Asset Management stages of deployment and through usage
systematically.

“Proving value is essential for any ongoing or new investment 

Adoption of AI/ML across units is finalised
including an AI or ML proof of concept. Given investment in based on initial reviews by internal and
a project has to be justifiable, measurable and competitive external personnel.
against all other projects vying for funding, these responses
demonstrate respondents can see and articulate the
benefits of AI and ML well but most agree that the benefits 
Sometimes, drastic changes are not noticed
are felt over a longer period, which suggests adopters need upon introducing AI/ML functions. But, it
to exercise significant patience as well.” does have a positive impact on projects
Tom Webber over time. Adoption of AI/ML is based on
Conference Director
the long-term impact it promises.
FIMA Europe 2024

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 18

At what stage is your organisation at with regards At what stage of your AI journey is your organisation
to AI Chatbot to further augment customer at regarding its efforts to unlock AI/ML capabilities
experience and agent efficiencies? through developing reference architectures for cloud,
modern database and ML operations platforms?

6%
32%
Early stage
8% Exploring
50% 24% Production
Technology
Stage

Exploring AI enabled Chatbot using traditional

Chatbot using internal search or decision tree
knowledge base
23%
Exploring
business
case 31%

18% 8% Selecting
Technology

Generative AI and LLM No Chatbot

already deployed for next
gen Chatbot “We are quite advanced ourselves because we have the
architecture in place, with modern cloud architecture and
very modern databases. I think these results give a true
representation of where businesses are at.”
“I’m not surprised to see that almost all respondents are
Elodie Blyth de Fontenay
exploring some sort of chatbot. It is a difficult thing to do
Head of Data Strategy and Transformation
it right and raises inevitable challenges, including how
Royal London Asset Management
to treat the output of the bot. A chatbot cannot make a
decision for you, for example, it can only give me some
useful information for you to make a decision. Exploring “Leveraging AI and ML and the Cloud in a single strategy is
chatbots is not just a technology matter, it should now an activity most banks and asset managers are invested
include the approach on how colleagues should use it, in. While there is still a range of maturity of adoption it is
when colleagues should intervene and all the associated clear all respondents are prioritising this strategy in one way
learning to do the right thing with it.” or another and the majority are at an advanced stage of
Elodie Blyth de Fontenay technology selection and deployment.”
Head of Data Strategy and Transformation Tom Webber
Royal London Asset Management Conference Director
FIMA Europe 2024

“Most respondents are actively using or planning to use

AI-enabled chatbots offering a clear view on a trend “These survey results showcase the diverse stages of AI
in banking and asset management to leverage this adoption within European financial data organisations. While
capability to deliver digital customer experiences.” a substantial portion (32%) are actively exploring potential
Tom Webber technologies, a similar number (31%) are already in the selection
Conference Director phase. This suggests a strong collective interest in harnessing
FIMA Europe 2024 AI’s potential, but also highlights the ongoing evaluation and
decision-making processes within many organisations.
A smaller group (6%) are still in the early stages, focusing on
building a foundation for AI implementation. Interestingly, a
limited group (8%) have already reached the production stage,
indicating successful integration of AI/ML capabilities within
their operations. These findings portray an industry actively
exploring and adopting AI, with many organisations taking
a thoughtful and measured approach to integrating these
transformative technologies.”
Bard
Large Language Model
Google AI

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 19

Conclusion

The financial data landscape is Ultimately, the widespread adoption of AI across the
undergoing a radical transformation, industry hinges on its ability to demonstrably create
driven by the relentless power of AI. value. Our research provides encouraging insights
in this regard, revealing that firms are increasingly
While these technologies are nothing new, what
turning to dedicated AI teams, both internal and
we have witnessed over the last 18 months is
external, to spearhead implementation efforts.
an unprecedented acceleration in its application
and deployment. This has forced even the most These teams play a critical role in driving AI adoption,
traditionally cautious sectors within finance to evaluating performance metrics, and quantifying
embrace AI and its many potential benefits. Our the tangible impact on the bottom line. Their work
research confirms this shift and highlights the highlights the crucial role of human expertise in
growing prominence of chatbots and AI-powered navigating the early stages of AI adoption within
cybersecurity solutions the industry.

However, integrating these solutions with existing While it is true that many firms are still in the early
technologies remains a significant hurdle that firms stages of their AI journey, the immense potential
must overcome. Financial data leaders are grappling of his technology to trasnform the financial
with data privacy concerns, particularly when it comes data industry cannot be denied. Challenges will
to the retention and use of sensitive information by undoubtedly continue to arise. But, with a continued
publicly available chatbots. Despite these challenges, focus on responsible development, data security,
the future appears to be bright. This is proved by and levering human expertise, AI has the power to
91% of respondents to our survey who expressed a completely revolutionise the financial data industry.
favourable view of their organisation’s AI capabilities,
suggesting a commitment to responsible deployment
and effective utilisation of AI solutions.
Generative AI (Gen AI) is another area witnessing
explosive growth within the industry. However,
data privacy is again a concern for many. Firms
are actively looking to address this challenge by
investing in interpretability tools that provide
greater transparency into how Gen AI models
function. Furthermore, knowledge graphs are rapidly
emerging as the preferred method for developing
Gen AI strategies. Knowledge graphs offer a
structured framework for organising and connecting
information, allowing Gen AI models to generate
more accurate, comprehensive, and contextually
relevant outputs.

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 20

Key Suggestions

1 Be wary of data privacy concerns

While the big chatbot providers are improving and addressing data
privacy concerns, firms should be wary of using public chatbots,
and the data that employees are giving them. Organisations have
been harmed by providing sensitive information. Financial data
executives should ensure employees are promoting data security
when engaging with public chatbots.

2 Lean on the experts

Many of the AI solutions that firms are beginning to explore
and implement are relatively new, and firms do not have
much experience with the technology. So, having a dedicated
team of AI experts will be required to ensure the proper and
appropriate deployment of these solutions.

3 Invest in training
Having teams of experts is crucial, but they should not be
relied upon exclusively for the long term. AI use will be
widespread across departments of all organisations before
long. Therefore, sufficient training processes should be put in
place to ensure all employees can use the AI technologies to
ensure maximum safety, efficiency and value.

4 Leverage chatbots
Our research reveals chatbots are providing the most value when it comes to AI use
cases. While concerns remain, particularly around data security, when leveraged properly,
chatbots can provide firms with incredible value. From enhanced customer service, as well
as assistance in budgeting, expense tracking and financial tracking, AI-enabled chatbots
should be used in some capacity by all firms going forward.

DOWNLOAD THE AGENDA HERE

The Artificial Intelligence Report 21

About

About Cognaize About WBR Insights

Cognaize was founded 2020 by Vahe Andonians, a We use research-based content to drive conversations,
banker, data scientist, AI-pioneer and senior lecturer at share insights and deliver results. Connect with our
the Frankfurt School of Finance & Management. Cognaize audience of high-level decision-makers in Europe and
specialises in hybrid intelligence to deliver on the promise Asia from industries including Retail and eCommerce,
of AI for financial services. Its AI-solutions are being Supply Chain and Procurement, Finance, as well as many
deployed by a growing number of financial services more. From whitepapers focused on your priorities, to
organisations, including the world’s largest banks, rating benchmarking reports, infographics, and webinars, we can
agencies, investment firms, and insurance companies. help you to inform and educate your readers and reach
your marketing goals at the same time.
Contact us to find out how your business could benefit
from:
• Year-round access to our network of decision-makers
and industry-leaders
About NEO4J • In-depth research on current fast-moving issues and
Neo4j, the Graph Database & Analytics leader, helps future trends
organisations find hidden relationships and patterns
across billions of data connections deeply, easily and • Lead generation campaigns that fit your priorities
quickly. Customers leverage the structure of their • Promoting your organisation as an authority in your
connected data to reveal new ways of solving their most industry
pressing business problems, with Neo4j’s full graph stack
To find out more contact:
and a vibrant community of developers, data scientists
and architects across hundreds of Fortune 500 companies. Martin Watchorn
Visit neo4j.com. Insights Sales Director
T: 01753 980796 Ext 1122
E: [email protected]

About Redis
Data is the lifeline of every business, and Redis helps
organisations reimagine how fast they can process,
analyse, make predictions, and take action on the data
they generate. Redis provides a competitive edge to
any business by delivering open source and enterprise-
grade data platforms to power applications that drive real-
time experiences at any scale. Developers rely on Redis to
build performance, scalability, reliability, and security into
their applications.
Born in the cloud-native era, Redis uniquely enables
users to unify data across multi-cloud, hybrid and global
applications to maximise business potential. Learn how
Redis can give you this edge at redis.com.

DOWNLOAD THE AGENDA HERE

 ©WBR Insights 2024. All rights reserved. This document and its content are proprietary to WBR Insights and may not be reproduced, republished or resold.
The information contained within is provided on an “AS IS” basis for information purposes only and WBR Insights makes no warranties of any kind including in
relation to the content or suitability. The name WBR Insights, the logo and any associated brand names are all trademarks of Worldwide Business Research.