S P O R TR A DA R W HITEPA PER 202 3

DON’T GET LEFT BEHIND:

HOW AI-POWERED
SPORTSBOOKS
ARE SHAPING THE
SPORTS BETTING
INDUSTRY

1
2
TABLE CONTENTS

0 EXECUTIVE SUMMARY 5

1 THE RISE OF AI 6

3 AI AND ML IN THE SPORTS BETTING INDUSTRY 8

3.1 Risk management 10

3.2 Computer vision 11
3.3 Player acqusition and retention 13
3.4 Integrity 15
3.5 Business intelligence 16

4 DEVELOPING AI AND ML TECHNOLOGIES 18

4.1 The challenge of in-house development 19

5 MYTHS AND MISCONCEPTIONS ABOUT THIRD-PARTY TECHNOLOGIES 21

5.1 Privacy 22
5.2 Integration 23
5.3 Dependence 23

6 A CLOSE LOOK AT SPORTSBOOK MODELS 24

6.1 Risk and liability management models 25

6.2 Player acquisition and retention models 30
6.3 Integrity 33

7 FUTURE OUTLOOK 35

8 KEY LEARNINGS FOR OPERATORS 36

9 OTHER WHITEPAPERS 37

10 REFERENCES 38

3
 76.75 BIL
Is the current valuation of sports
betting in 2021, according to Data

167.7 BIL Bridge Market Research forecasting.

The global sports betting

market is expected to reach
USD 167.66 billion by 2029,
more than doubling its
current valuation at 76.75
10% 37%
billion in 2021 — register- In 2015, just 10% of By 2019, that number
ing an impressive CAGR of organizations reported more than tripled.
10.26% during the period. that they already used AI or Reaching more than 37%
would so in the future. of organisations.

27% 79% 40%+ 50%

Of orgs. say Of organisations Just a singular Organsations
that at least report that AI mention of AI with a
5% of their adoption has on an earnings mature AI
earnings decreased call has been infrastructure
before taxes operational shown to experience
are attributable to AI. costs in 2021, some by as generate a 40% greater 50% greater revenue
much as 19%. change that share prices growth on average than
would increase. their competitors.

LIST OF ABBREVIATIONS
• AI: Artificial Intelligence • DL: Deep Learning
• AML: Anti-Money Laundering • LDO: Liability Driven Odds
• BI: Business Intelligence • LTD: Live Time Delay
• CAGR: Compound Annual Growth Rate • LTDO: Live Time Delay Offset
• CCF: Customer Confidence Factor • ML: Machine-learning
• GDPR: General Data Protection Regulation • MTS: Managed Trading Services
• CPA: Costs Per Acquisition • UFDS: Universal Fraud
• CRM: Customer Relationship Management Detection System
• CV: Computer Vision

4
1 | EXECUTIVE SUMMARY
Sportsbook operators have a lot to be Forward-thinking bookmakers are already
happy about these days. Having been valued leveraging AI to their advantage on a daily
at USD 76.75 billion in 2021, the global basis. Top to bottom, complex AI models
sports betting market is expected to reach are increasingly streamlining sportsbook
USD 167.66 billion by 20291 — registering an operations, driving bookmakers’ businesses
impressive CAGR of 10.26% during the period. and changing the game for every player in the
market. But what exactly are these technologies
The increasing popularity of mobile betting, capable of doing? How can you take advantage?
the commercialisation of sporting events What does it take to develop and apply them?
and the rise of the U.S. market, are among the This paper will aim to answer these
factors driving the industry’s growth. That said, questions and illustrate the importance
capitalising on these important opportunities of artificial intelligence to sportsbook
isn’t an easy win, even for the large, globally management. From its earliest iterations to its
acting operators. Beneath the surface, there’s newest innovations, let’s explore the impact of
another force driving the future of the sports AI and what bookmakers need to prepare for
betting industry: the rise of artificial intelligence. the future.

The global sports betting market is increasing in popularity and expected to double

10% $167.66B
CAGR

$76.75B

2021 2029

5
2 | THE RISE OF AI
Despite its humble beginnings, the operations. From supply chain management
growth of AI has rapidly accelerated in recent and cloud security to customer service and
years. Today, its many subsets, including facial recognition, AI has the potential to
machine learning, deep learning and redefine what it means to be an efficient and
computer vision, are leveraged by industries innovative organisation.
of all shapes and sizes. Artificial intelligence has grown rapidly in
recent years. In 2015, just 10% of organisations
When you imagine artificial intelligence, reported that they either already used AI or
you might imagine a far and distant future, would be doing so in the near future. 3 By 2019,
like something out of science fiction. But in that number more than tripled. Reaching more
reality, AI isn’t just where the world is headed than 37% of organisations, AI surged 270% in
— it’s the here and now. Already, businesses just four years.
are implementing algorithms into their daily

At its most basic level, AI aims to mimic the problem-solving and decision-making processes of the human
mind using computers and machines. First hypothesised by Alan Turing — the “father of computer science” —
in 1950, AI has transcended theory and is now very much a reality. 2 In the decades since, several subsets of AI
have emerged, such as:

Machine learning: Machine learning (ML) algorithms provide computers with the ability to
 automatically ingest historical data and improve performance. In other words, they learn from
experience without needing to be explicitly reprogrammed. By automatically analysing data, ML
algorithms can identify patterns in information and produce valuable insights.

Deep learning: Deep learning (DL) models are a subset of machine learning. DL algorithms can
 learn by ingesting unstructured data, such as text or images, and automatically determining
characteristics that distinguish one category of data from another. The breakthrough of deep
learning has eliminated much of the manual intervention required to train certain AI systems.

Computer vision: Computer vision (CV) is one of the fastest growing fields of AI to date. CV
 algorithms enable computers to see and understand visual inputs, such as digital images or videos.
Whether it be the swing of a bat or the distance of a throw, CV models analyse visual data and
produce insights faster than any human possibly could.

6
 AI surged 270% in just four years

2015 10%
Organisations
using AI

2019 37%
Organisations
using AI

Now, with adoption only accelerating, the • Just a singular mention of AI on an

global market for AI is expected to balloon at a earnings call has been shown to generate a
CAGR of 38.1% through 2030.4 One of the many 40% greater chance that share prices would
factors propelling this growth is the continued increase.6
realisation of AI’s substantial business potential. • Organisations with a mature AI infrastructure
Regardless of industry, enterprises are seeing the experience 50% greater revenue growth on
impact of AI on their bottom line. average than their competitors.
• From tech, automotive and transportation
• 27% of organisations say that at least 5% of their to manufacturing, advertising and finance,
earnings before taxes are attributable to AI.5 countless industries are leveraging AI to their
• 79% of organisations report that AI advantage. For example, Metro de Madrid
adoption has decreased operational costs in — one of Europe’s oldest urban rail systems
2021, some by as much as 19%. — uses AI algorithms to optimise electricity
consumption, thus reducing annual energy
intake by 25%.7

7
3 | AI AND ML IN THE SPORTS
BETTING INDUSTRY
With AI growing at breakneck speed, its increasingly diverse applications, AI
forward-thinking bookmakers began technology is poised to become an integral
implementing the technology into their part of any sportsbook operator’s playbook.
regular operations. Now, considering

Operators’ opinions of whether AI and ML will be game changers for their business

3% 3%

23% Yes
Will AI and ML be a Not yet, but in the future
game changer for
business? No
I don't know
71%

Source: Sportradar 2022 Industry Survey

8
 There’s no reason to believe that the respondents still aren’t implementing artificial
success of AI’s many applications can’t be intelligence in their operations. The reason
translated to the data-rich field of sportsbook for this is debatable, but one thing is for sure:
management. In fact, bookmakers are already Operators who don’t keep up with technological
beginning to realise the potential of AI. developments risk lagging behind the
According to a Sportradar survey of global increasingly competitive market. To illustrate
bookmakers, 71% of operators believe AI their importance, it’s vital to understand how AI
and ML technologies will be game changers algorithms are disrupting the industry.
in the betting industry. However, half the

Operators’ replies to whether they are already using AI and ML technology within their organistions

7%

Are you already 43% Yes

using AI and ML
tech with your No
organisations? I don't know
50%

9
3.1 | RISK MANAGEMENT
According to Sportradar data, operators Managing Director of Managed Trading Services
rank risk management as the area of their at Sportradar. “You’ll be more exposed to risk
business with the greatest potential for using and volatility and you’d be less efficient in your
AI and ML. For example, these technologies transaction handling and how much profit you
allow operators to create sharper, more take out of each transaction.”
accurate odds in real time. Self-learning AI
analyses historical data using sophisticated
algorithms, thus giving sportsbooks a better Darren Small,
sense of how likely an event is to occur. Managing Director of Managed
Trading Services at Sportradar
In turn, operators can set odds accordingly
and reduce their exposure to risk. Not only does
this help bettors and operators win more often,
but it also increases betting activity overall. As
a result, the operator can offer better pricing to
“If you have more than 12 million
its customers and maximise its margins. active accounts a month that
AI technologies are also helping operators you have to manage, it would
identify sharp bettors through Automated be near impossible to profile all
Player Profiling. By analysing player data, AI can of those accounts in a real-time
automatically generate an accurate profile of a
bettor’s behaviour and determine how much of
environment and do that process
a risk they are to the bookmaker. These insights manually. You’ll be more exposed
empower operators to protect their business to risk and volatility and you’d be
without needing to comb through data by hand. less efficient in your transaction
“If you have more than 12 million active handling and how much profit
accounts a month that you have to manage,
it would be near impossible to profile all of
you take out of each transaction.”
those accounts in a real-time environment and
do that process manually,” says Darren Small,

Areas where operators see greatest potential to apply AI and ML

Trading & Risk Management 29%

Acquisition, retention, bonusing 26%

Business Intelligence 25%

Customer Support 18%

Others 2%
Source: Sportradar 2022 Industry Survey

10
3.2 | COMPUTER VISION
And with the automated, frictionless and With this information, you may detect
non-invasive data collection made possible patterns in performance.
by computer vision, sportsbooks will “Skilled punters would maybe see this
continue to optimise risk tolerance. situation,” says Pataky. “If they understand
it correctly the operator would potentially
When it comes to CV, it’s not about using the be exposed to more risk without that same
technology directly in trading — it’s the output information.”
of CV that can be of great benefit. According to
Luka Pataky, Chief Product Officer of Computer
Vision at Sportradar, this additional information Luka Pataky,
can be used by risk managers to trade better. Chief Product Officer of
“For example, we might be observing a Computer Vision
tennis match that enters a deciding game,” he
says. “If we only look at past performance and
scores we would conclude that whoever was the
favourite will win, right? But perhaps there is “It’s not about using the
another angle to it.”
In other words, CV can spot more latent technology directly in trading —
data points that could impact the outcome of it’s the output of CV that can be of
the match, such as an injury, illness or fatigue. great benefit.”

11
We are focused on delivering innovation and quality solutions. Check out this case study example.

12
3.3 | PLAYER ACQUISITION
AND RETENTION
Operators rank player acquisition and 2 – Player potential
retention as the second biggest area of their
business where AI and ML technologies If you want to maximise player acquisition
could potentially have the greatest and retention, you need to understand your
impact. According to Jay Kanabar, COO of customer at the most granular level. In other
VAIX — a pioneer in the development of words, you need to leverage as much insight
AI in the gaming industry — there are two into their individual journey as possible. In
applications that stand out above the rest. combination with personalisation models,
player profiling and lifetime value algorithms
are enabling bookmakers to make accurate
1 – Personalisation predictions about their customers.
“We’ve classified personalisation as the More importantly, they’re allowing them
sexy product because no one’s really doing it, to forecast player value across multiple
but everyone’s been doing marketing or CRM time-frames simultaneously. Knowing
since day one,” says Kanabar. “I often say that how much a player might be worth during
the quick wins are optimising your marketing different milestones in the first 12 months of
budget or retention strategy, which is where their activity in comparison to what they’re
customers often overlook what they’ve been worth today changes how the operator
doing and don’t change things.” might approach that specific player. In turn,
AI-driven personalisation models are the bookmaker can further optimise their
enabling operators to do exactly that. marketing strategies.
Personalisation, states Kanabar, extends “Our Deep Learning models have more
the bettor’s discovery process by allowing than 50 different inputs that go into it, and
bookmakers to target customers with the right that could be age, location and the various
message at the right time. These models work parameters of your depositing and betting
by detecting patterns in player behaviour and pattern,” says Kanabar. “Looking at the
using that information to target messages, different matrices of who you are allows us to
bonuses and bets more effectively. predict the player value.”
Bet recommendation models, for instance,
enable a more personalised betting experience.
If a player tends to bet on underdog teams, the
Jay Kanabar,
AI might send a special promotion around those Chief Operating Officer
events. Or, if a player prefers to bet on a specific & DPO at VAIX
sport, programmatic display models can ensure
the most relevant matches are at the top of the
page where the bettor can easily access them.
“The more interesting content you see, the
likelier you come back,” says Kanabar. “If you “Our Deep Learning models
see college football, you might bet twice a week have more than 50 different
rather than once a week on NFL. And that’s how inputs that go into it, and that
it increases retention — by showing you content could be age, location and the
that you want to see before you even know
it existed.” various parameters of your
depositing and betting pattern.
Looking at the different matrices
of who you are allows us to
predict the player value.”

13
 These deep learning algorithms generate However, operators also depend on
several valuable insights that bookmakers receiving these insights as fast as possible.
can use to optimise their retention strategies, Acquiring a prediction is one thing, but
including the likelihood a player will churn, how acquiring an accurate one quickly is entirely
much they might spend per bet and whether or another. Sophisticated models, like those
not they’re likely to become a VIP bettor. provided by VAIX, can produce proper
predictions within just two or three days of
player activity.

14
3.4 | INTEGRITY
AI has become commonplace in fraud Most significantly, though, this
detection in industries such as banking, advancement in the application of AI to the
insurance and eCommerce and is firmly detection process has enabled the human
in the adoption phase across the betting operation to narrow their focus on matches
industry – particularly in helping operators with suspicious activity – as opposed to those
with their Anti-Money Laundering (AML) that generate false positives.
activity and supporting social responsibility With match-fixing accounting for
in identifying problem gambling.8 approximately €165 million of betting profit in
2021, upholding integrity is more important
However, the early adoption of AI in than ever.11 AI-driven fraud detection services
Sportradar’s Universal Fraud Detection are helping operators eliminate fraud and
System (UFDS)9 has become fundamental maintain a clean betting environment.
in the Integrity Services’ ability to detect
and report suspicious activity to sports
federations. The reality is the application of
this technology within the UFDS is a key reason
why Sportradar is now able to offer its core
monitoring service for free to over 150 sports
partners. The successful launch of Sportradar’s
Integrity Exchange is also underpinned by this
commitment and investment.10

15
3.5 | BUSINESS INTELLIGENCE
In the increasingly crowded sports “I think the core thing about AI,” says VAIX
betting market, every advantage matters. COO Jay Kanabar, “is the speed, accuracy and
To stay ahead of the competition, operators the volume of data that one can consume.
should be leaning on the insights provided by There’s only so much that I can do in Excel.
artificial intelligence. There’s only so much I can do in Sequel or in a
Python script.”
Whether it be ML, deep learning or
computer vision, sophisticated AI models
are enabling sportsbooks to level the playing “I think the core thing about AI
field and translate large volumes of data into
meaningful business intelligence. Bookmakers
is the speed, accuracy and the
are applying these insights throughout every volume of data that one can
area of their business, from budgeting and consume. There’s only so much
trading to marketing and forecasting finances. that I can do in Excel.”
Critically, these are insights that don’t need
to be gleaned using manual processes — AI
automates their collection, translation and
~Jay Kanabar
visualisation into a comprehensive display.

16
 AI not only streamlines this process, “It’s the ability to get data faster and
but also does so without bias. According to more accurately than any human being could
Kanabar, data scientists often go in with an deliver to you,” he says. “You want to be
opinion before they’ve done the analysis, proactive rather than reactive, especially in
whereas AI doesn’t. It simply looks at the an industry where there’s easy access to your
data and makes a decision based on the input competitor next door. You want to be a step
provided to the model. ahead of that competitor.”

What Operators say:

In the survey Sportradar carried out, the company asked operators, what they think might
be the biggest benefits of using AI and ML for their operations. Considering the option to
give multiple answers, the results are ranked by importance.

Improved customer insights and analytics

Improved operational processes

Creation of more reliable odds

Improvements in customer acquisition and retention activities

Potential creation of new markets and promotion options

Improved business security

Others benefits not mentioned above

17
4 | DEVELOPING AI AND ML
TECHNOLOGIES
Often times, organisations who choose to The results of the Sportradar survey
develop their own proprietary technologies indicate that the majority of operators are well
don’t realise the uphill battle they’re about aware of how important AI will be in the sports
to begin. Before embarking on that journey, betting industry. In fact, 70% of respondents
it’s wise to consider what it truly takes to claim that AI and ML will be game changers.
create a high-quality and reliable AI system. However, just 43% of respondents say they’re
using these technologies in their operations.
In a perfect world, operators could As a key competitive advantage in an
pause time and capitalise on each and every increasingly crowded market, it’s vital that
opportunity that’s thrown their way. But organisations strategize how to best approach
in reality, the pace of technological change AI adoption. For many, proprietary technologies
demands immediate action — and there’s no are an enticing path forward, but ultimately
technology with their foot so firmly on the gas also involve several substantial challenges that
as artificial intelligence. ought to be addressed beforehand.

Operators’ replies to whether they develop AI technology in-house and/or use 3rd party technology

Source: Sportradar 2022 Industry Survey

23% 27%

As an operator It’s developed in-house

will you develop
AI tech in-house It’s third party technology
or use a 3rd party
technology Both, in-house and third party

18
50%
4.1 | THE CHALLENGES OF IN-
HOUSE DEVELOPMENT
27% of Sportradar’s survey respondents Ecosystem
say they develop AI in-house, as opposed
to sourcing it from a third-party provider. The next biggest challenge bookmakers
In doing so, however, these operators also face when developing proprietary AI is usually
choose to persevere through the many their ecosystem. As Personeni explains,
challenges this process entails. you need to have a data-scientist-friendly
infrastructure in place to attract the right type
Paolo Personeni, Sportradar’s Managing of talent. However, this challenge isn’t exclusive
Director of Managed Betting Services, identifies to the sports betting sector. In truth, many
three primary obstacles when it comes to organisations don’t have the in-house expertise
developing proprietary AI technology. The first needed for the job.
one, he says, is size. According to Kumar Abhishek, a machine
learning engineer at Expedia, the lack of
skilled personnel is a major barrier to entry.
Size “Organisations need to have staff with the
“The minimum size at which a bookmaker necessary skills to develop and operate AI and
or a gaming operator can maintain their ML technologies,” he says. “This can be difficult
own sophisticated platform or invest in AI is to find, as the field is relatively new and there is
increasing,” says Personeni. “And for some, they a shortage of qualified workers.”
just don’t have it.” In fact, 64% of global IT executives say that
That said, Personeni adds that the size of a lack of skilled talent is their biggest barrier to
the bookmaker isn’t necessarily indicative of adopting emerging technologies like artificial
their ability to develop technologies effectively. intelligence.12 Without expert programmers,
Larger operators may have access to more engineers and analysts, operators simply can’t
resources, but more resources don’t guarantee get their projects off the ground.
success.
“A lot of very large and established
bookmakers are lagging behind others in terms
of the adoption of artificial intelligence,” he Paolo Personeni
explains. “It needs to be embedded in the Managing Director of Managed
platform into technical tools and sometimes Betting Services
operators have very old legacy tools that just
can’t technically do what they would like to do.”
For this reason, some younger, smaller
operators are in a better position to adopt AI “If you don’t have data you can’t
sooner and more efficiently than their larger
counterparts.
feed the algorithms.”

19
Access to data “What’s important is not just the quantity of
data,” says Andrej Bratko, head of technology
Finally, Personeni identifies data as perhaps for AI and BI at Sportradar, “but also the
the biggest challenge to in-house development. diversity of data that you have.”
High-quality data, he says, is fundamental to The golden rule is typically that the amount
any AI initiative. You can have money, you can of data required is 10 times the number of
have a good team, but if you don’t have data parameters included in the algorithm. In sports
— the “oil of the 21st century,” he adds — you betting, whose models need to account for an
can’t feed the algorithms. In turn, you can’t get extremely high volume of factors, acquiring
anything out of the process. And if you do, it enough data is exceedingly difficult. Even
won’t be anything meaningful. for operators who have a sufficient internal
“Quality data is necessary for a viable AI/ database, they may still lack the diversity
ML initiative, in the same way fresh water is needed to cover the entire market the same way
necessary for a healthy garden,” says James a global provider can.
Caton, head of AI partnerships at the SAS
Institute. “AI and ML initiatives will wither away
slowly without good data prep, which ensures
a steady flow of fresh, nutritious data for the
data-hungry models.” Andrej Bratko
Caton’s advice to operators is this: Head of Technology for AI and BI
Don’t start with how you want to apply the at Sportradar
technology or the problem you want to solve.
Instead, start with finding your best source of
data in the organisation, validate it and identify
a problem you can address with that dataset. “What’s important is not just the
“That is your quickest path to success,” quantity of data, but also the
adds Caton. “Starting with a problem, and then
hoping to find complementary, complete and diversity of data that you have.”
high-quality datasets is a much tougher path to
take.”
However, this is easier said than done for
the average bookmaker. Data may be a dime a
dozen, but high-quality datasets are far and few
between. For some operators, access to quality
data is an expensive prohibitive factor.

20
5 | MYTHS AND MISCONCEPTIONS
ABOUT THIRD-PARTY TECHNOLOGIES
For bookmakers who struggle to algorithms, third-party technologies offer
overcome the obstacles of internal a more accessible approach to artificial
development, third-party providers intelligence. Engaging a third-party provider is
supply an ideal alternative. ultimately a smarter, more sustainable long-
term solution. However, some operators still
Rather than investing in proprietary have their reservations.

When asked why they might be reluctant to leverage external AI technology,

operators listed the following three points as most important:

The possibility of a provider Integrating third-party The possibility they’ll

gaining insight into sensitive technologies might require become too dependent on
areas of the company too much effort an external provider

Additional answers included amongst others:

Lack of knowledge No trust in the Potential loss of Using AI/ML as

to apply the effectiveness of autonomy, self- black box would
technology external AI/ML sustainability and not help company’s
technology long-term process understanding of
stability problem

21
5.1 | PRIVACY
“One thing we always say is we never “We are ISO-certified and GDPR
want personally identifiable information,” compliant,” he explains.
says VAIX COO Jay Kanabar. “Give us your
data at the player level, where we don’t “We don’t merge data between
know who’s betting or how much they’re different operators and we
betting — it’s just a number to us.” have separate instances.
We’re a serious organisation that
Although data processing agreements won’t sell your data without your
are a necessary part of the partnership,
that doesn’t mean technology providers are permission.”
accessing sensitive corporate information. At
the end of the day, Kanabar says, it all comes ~Jay Kanabar
down to trust.
“We are ISO-certified and GDPR
compliant,” he explains. “We don’t merge
data between different operators and we
have separate instances. We’re a serious
organisation that won’t sell your data without
your permission.”

22
5.2 | INTEGRATION
Ultimately, integration isn’t about fitting
a square peg in a round hole. Instead, third-
party providers are focused on empowering
operators to succeed and integrating their
technologies as seamlessly as possible. In
short, it’s about being just as much of a
partner as you are a provider.

“We go in asking what their problems are,”

explains Darren Small, Managing Director of
Managed Trading Services. “‘What do you see
your future being in the next three years? What
would you need to support your growth and
success in the future?’ And then we look at our
product range to see how we can underpin
some of their questions, their thoughts and
how we can help them to grow in those areas to
support what they need to do.”

5.3 | DEPENDENCE
“I think there is a point when ownership For this reason, Small argues that
is a big question,” says Darren Small. “Is it bookmakers shouldn’t be afraid of losing
better that I own that technology and can autonomy by engaging a third-party provider.
take it down a road I want to go down? Do I External solutions can be customised to
want to be able to own the destiny and the meet the specific needs of any size operator.
final destination of this particular piece of Whether they want to retain full control over
technology?” the technology or lean on the dedicated
infrastructure of third-party experts, the
In reality, most operators lack the right operator is always behind the scenes calling
combination of size, talent and access to the plays.
data. Because these factors are such a rare
commodity, proprietary development isn’t
often realistic.

23
6 | A CLOSER LOOK AT
SPORTSBOOK MODELS
Thus far, this paper has spoken generally To demonstrate the real potential of
of AI applications and how sportsbooks artificial intelligence and its several subsets,
might leverage them throughout their the following section will examine models
operations. However, it’s difficult to truly developed by two of the industry’s leading
understand the power of AI without concrete technology providers: Sportradar and VAIX
examples of algorithms in action. (which got recently acquired by Sportradar).

24
6.1 | RISK AND LIABILITY
MANAGEMENT MODELS
Automated Player Profiling In contrast, Automated Player Profiling
can automatically assess a player’s risk and
Without the use of AI, player profiling boost operator profits compared to traditional
is a manual, time-consuming and error- profiling. By automatically performing millions
prone process. If player behaviours aren’t of player checks every day, the bookmaker can
appropriately analysed, the bookmaker can’t simplify risk management and gain a clearer
sufficiently determine risk tolerance per bettor. picture of their exposure — thus allowing them
Consequently, risk exposure increases and to maximise revenue.
turnover goes down.

Automated Player Profiling — Efficiency Study

Simulation Analysis

Understanding how players behave let operators determine the level of risk they can
tolerate per player, shop or terminal. Which means bookmakers can protect their
business while offering the best possible betting experience.

The analysis above shows a simulation of player profiling versus no profiling (which
means all players are equal). Whether using fully automated profiling (where AI does all
the work) or semi-automatic (where traders use the AI to inform their choices) there’s a
clear boost to both turnover and profit.

You can also see that semi-automated profiling consistently performs the best, with an
additional accepted YOY cumulative turnover of over €363 million for 2022, resulting in a
plus of over €39 million in profits.

25
 Furthermore, the technology works best Sportradar’s Alpha Odds deploys its
in cooperation with operators’ own trading Liability Driven Odds (LDO) model, which allows
teams, as it allows them to intervene manually operators to automatically manage odds in
at any time. They can either have self-created real time. The algorithm works by constantly
values checked or fine-tune automatically adjusting probabilities according to the current
created profiles in order to better match them liability situation. Odds that would increase
to specific business strategies. liabilities are automatically lowered to deter
business, whereas odds that would decrease
Alpha Odds liabilities are raised to attract bettors.
As a result, bookmakers can boost their
Keeping a balanced sportsbook is no easy margins and gain a more profitable, less risky
task, especially when operators are doing it by operation. In fact, analysis of the LDO’s effect
hand in a real-time environment. To keep up on 2.7 million bets and more than 10,000
with the pace of the action, sportsbooks need events indicates a positive impact on margin
a model that can continuously update odds performance of approximately 2%. Across all
according to their risk exposure and liabilities. markets, LDO increased margins and generated
a substantial boost to turnover.13

26
Alpha Odds — Efficiency Study

Graph Analysis Summary

The simulation over the second half of 2022 shows the impact Alpha Odds had on live bets
on soccer compared to the standard live odds models. Alpha Odds have created consistently
better margins and a cumulative uplift in turnover.

Cumulative Efficiency Percentage Different — Efficiency Study

Graph Analysis Summary

In our analysis of live soccer bets, we calculated the margins on actual Profit and Loss/
Turnover, and the efficiency on expected Profit and Loss/Turnover. Alpha Odds has achieved
a higher cumulative efficiency than standard Live Odds model, resulting in an overall
cumulative uplift of 8 percentage points over the course of six months.

27
Live Time Delay Offset As a result, it automatically decides in real
time and on a bet-by-bet basis whether to
Sports scouts aren’t always at the venue. apply the default live betting delay an operator
The bettors who are in the arena have an unfair has set up, to skip it or even increase it if
advantage: they’re seeing the action before advisable. Thus, operators can provide a more
anyone else. The standard way to mitigate this personalised and improved betting experience
risk is a validation delay on all live bets. But and increase their profits in the long term.
that slows things down for everyone, and puts
people off betting in the first place.
Sportradar’s Live Time Delay Offset model
is trained to evaluate any incoming ticket
based on the player’s profile, the current match
situation and the betting market wagered on.

Avg. Bet Validation Delay Per Ticket — Efficiency Study

Graph Analysis Summary:

Statistics for trailing 12 months from January to December 2022 — considering all
MTS tickets with at least one in-play selection on soccer, basketball or tennis and
no live selections on other sports:

Bet validation delay reduced by 1.8 seconds on average

88 years of time saved for bettors waiting for bet acceptance

LTD skips were applied for 26.8% of all bets amounting to 421 million tickets

Profit margin of 11.3% achieved on bets with LTDO applied vs. 9.7% on bets
without LTDO model applied

28
Live Time Delay Offset — Efficiency Study

Summary:

Our Live Time Delay Offset (LTDO) model uses AI to check each bet request in milliseconds, working out if the
standard bet validation delay should be applied, skipped or even extended. Over the course of 2022, the LTDO
model resulted in an extra accepted turnover of over €50 million and constantly higher margins.

Comparison Achieved Margins — LTDO vs. no LTDO

Summary:

Furthermore, the LTDO model saves time for players placing bets. By working out
which default delays can be skipped, the average bet acceptance delay for all bets will
go down significantly. All those saved seconds add up – giving an overall KPI of nearly
90 years. That makes life easier for the operator, and betting faster for their customers.

29
6.2 | PLAYER ACQUISITION AND
RETENTION MODELS
Player Lifetime Value Prediction 85%, as compared to an industry average
of 55%.14 In addition to lifetime value, these
When it comes to optimising your models also predict player churn, as well as
acquisition and retention strategies, it pays which players are likely to become a VIP.
to know as much about a player as possible. Most importantly, it can deliver these
Without technology, operators would have to insights in just two or three days of player
manually sift through masses of player data and activity. In other words, operators can leverage
activity just to generate a single prediction. information faster and with more confidence,
AI offers a much faster, more efficient and allowing the marketing team to change
more accurate solution. VAIX’s Player Lifetime direction as necessary and optimise their
Value model can predict a player’s value across strategic efforts.
multiple time-frames with an accuracy of over

Churn Prediction — Case Study

KPI Operator Benchmark CRM AI-Powered CRM

Email open rate 9.47% 16.78%

Email click rate 0.59% 2.64%

Users remained active 28.8% 39.5%
Wager uplift - 29.7%
Deposit amount uplift - 32.6%

The AI targeted a list of top 10% wagering players that are predicted to churn.
The players were split into two groups: Control lead by operator decisions vs Test
lead by AI decisions.

30
Sports Personalisation that targets specific players with the most
relevant information possible. By analysing
In the world of modern marketing, player data, the model can recommend deeply-
personalisation is the name of the game. relevant bets or events, which can help a player
Bettors are increasingly seeking out content organically discover content they’d likely be
tailored to their specific needs, interests interested in betting on.
and betting preferences. AI-driven player A Personalisation engine such as VAIX’s can
retention modules are supplying sportsbooks boost player activity by up to 8% and increase
with exactly the tools they need to deliver a player retention rates by as high as 60%.15
personalised experience. Simply put, it captures your player’s attention
For example, a Sports Personalisation and drives engagement to your services.
model can generate hyper personalised content

Recommended Bets — Case Study

An A/B test was run over the course of 6 weeks with over 40,000 active users. The AI-
powered personalization was compared to the operator’s rules-based algorithm with the
following results:

2-3x 43% 115% 17%

increase in click increase in increase in turnover increase in ACCA
through rates number of bets and 147% in profits length

Player-personalized Site — Case Study

The A/B test was run over the course of 3 months with over 220,000 active users. The
benchmark was set up on the operator’s popularity-based algorithm versus the AI-driven
personalization algorithm. The results found were:

>8% +8% +9% +8%

increase in click increase in number increase in tickets increase in
through rates of bets placed created turnover

Programmatic Acquisition AI technologies offer operators a way to

grow their customer base without increasing
Business leaders know that the cost of cost per acquisition (CPA). Programmatic
acquiring a new customer is often much Acquisition models, for instance, allow
greater than that of retaining an existing bookmakers to personalise outbound
one — and sportsbooks are no exception. In communications and deliver a more targeted
turn, attracting new bettors as effectively and message. By hitting prospective customers with
efficiently as possible is a constant challenge for the right message at the right time, operators
bookmakers of any size. can optimise acquisition while also reducing
CPA by as much as 40%.16

31
We are focused on delivering innovation and quality solutions. Check out this case study example.

32
6.3 | INTEGRITY
Match-fixing and fraudulent betting Furthermore, modelling derived from
stand in stark opposition to what the sports Sportradar’s MTS account-level data now
betting industry is all about. Yet, it’s a enables enhanced risk management through
reality that bookmakers are dealing with the removal of continuously offending teams
on a constant basis. To uphold the integrity and individuals from betting products, which
of their services and provide players a safe ultimately reduces the exposure of match-fixing
and legitimate experience, operators need a to betting partners.
way to detect suspicious activity across all of This elevates the service UFDS can provide
their events. across the entire ecosystem to the benefit of all
stakeholders. Importantly, it is additional bet-
Universal Fraud Detection monitoring intelligence that sometimes cannot
be detected through odds movements alone.
Sportradar’s Universal Fraud Detection Tom Mace, Director of Global Operations
System (UFDS) has historically monitored Integrity Services, firmly believes this type of
billions of odds changes and account level technology is the benchmark for current and
transactions year-on-year on a global scale, future monitoring: “The threat of match-fixing
with analysts reviewing anomalous activity is ever present and evolving all the time. We
through automated alert generation – in the are seeing record numbers year-on-year in
process, a vast historical dataset of both the amount of matches we detect, and with
suspicious and non-suspicious activity is more and more matches to monitor, we are
now readily available for training supervised continuously investing in technology to improve
machine-learning models. our quality and our efficiency – it’s absolutely
As a result, since 2018, AI has been essential.
processing hundreds of thousands of UFDS “Not only has AI enabled us to monitor more
alerts to support the identification and matches annually, it’s allowed us to do this in
escalation processes conducted by expert a cost-effective manner, which is important to
betting analysts – increasing not only the all of the stakeholders we service – particularly
amount of activity detected, but the speed at post-pandemic.
which it is.

33
 UFDS predicts in real-time the probability of a match being suspicious on a continuous
basis – from markets opening to the end of the event - and simultaneously provides feature
contributions of the model, essentially narrating to analysts which elements of the data are
potentially suspicious.

“However, one thing that mustn’t be take more and more responsibility, it must be
forgotten is the people in our Integrity unit remembered our highly skilled Integrity betting
driving the product and service forward every experts are the bridge between our technology
year. A diverse workforce unlocks innovation and our partners.”
in every area, and while AI will continue to

34
7 | FUTURE OUTLOOK
With adoption quickly accelerating, and “AI technologies will continue making
new developments happening every day, AI sportsbooks more and more efficient,”
technologies will reach nearly every corner according to Bratko. “And I think the net effect of
of the industry in the near future. that, for players, is that they’ll get a better price
and less friction in their betting activity going
Looking ahead, the global market for forward.”
artificial intelligence will grow at an annual Other subsets of AI are also paving the way
rate of nearly 40% between 2022 and 2028.17 for significant advancements in sportsbook
As one of the most impactful and disruptive technology. For instance, as venues continue to
technologies in recent years, it’s likely that more install sophisticated cameras, computer vision
industries will continue adopting AI and taking models are poised to deliver unprecedented
advantage of its vast capabilities. And now that insights into player behaviour.
sportsbooks are beginning to understand the One of the biggest industry trends,
value that AI and ML models bring to the table, according to Luka Pataky, Sportradar’s Chief
it’s unlikely that the sports betting industry will Product Officer of Computer Vision, is that
ever be the same. there’s a rising demand for deeper data. It’s
According to Andrej Bratko, almost all through this additional information, he says,
areas of sportsbook management will soon be that bookmakers acquire their competitive edge.
impacted by artificial intelligence.

“Almost all areas of sportsbook “If you don’t have information,

management will soon be you will not be competitive. And I
impacted by artificial intelligence”. really think computer vision is key
to acquiring that data, which means
~Andrej Bratko you’ll need to have CV to compete.”

~Luka Pataky

35
8 | KEY LEARNINGS FOR OPERATORS
In our increasingly data-driven world,
AI is the wave of the future: it’s hard to imagine the rise of artificial
intelligence slowing down anytime soon.
71% of surveyed operators agree that When it comes to the sports betting sector,
AI and ML technologies will be game AI-driven sportsbooks are no longer the
changers in the industry moving forward. exception — they’re the expectation.

And while AI and ML technologies are still a

relatively untapped resource by bookmakers,
Few operators are prepared to their substantial benefits continue to make
develop in-house technologies: waves across the industry. The faster that
operators leverage their capabilities, the sooner
Proprietary AI models not only they can leapfrog their competition and expand
require high-quality, deep and diverse their business.
datasets, but also size and skill — That being said, it’s also becoming
three critical resources that even the abundantly clear that not every operator
largest bookmakers may not have at has what it takes to develop sophisticated AI
their disposal. technologies on their own. Whether it be due
to a lack of size, skill, data or a combination
of the three, most operators will find internal
Third-party AI offers a better development a much too risky, time-consuming
long-term solution: and expensive initiative. Entrusting an external
partner doesn’t mean sacrificing autonomy or
Operators who engage with a third-party risking interference; it means leaning on their
provider gain the advantage of well- expertise and leveraging it to your advantage.
developed, flexible and cost-effective That’s why Sportradar is committed to
technologies that offer faster time to empowering operators with its own proprietary
value than proprietary developments. technologies. Using its experience as an
This approach also eases the challenge of industry-leader in the development of AI,
adapting models to fit different content, Sportradar offers bookmakers a chance to
sports or betting markets when a fully capitalise on the AI opportunity without the
agnostic system cannot be deployed. burden of doing it on their own.

36
9 | OUR OTHER WHITEPAPERS

How live streaming can supercharge your betting business

See content

Building the sportsbook of tomorrow. Today.

See content

How to control customer acquisition costs in sports betting

See content

How strong partnerships can future-proof your betting business

See content

Tackling data for the knowledge economy

See content

The value of international sports to U.S. sportsbooks

See content

37
10 | REFERENCES
1. Data Bridge Market Research, Global Sports Betting Marketing – Industry Trends and Forecast
to 2029, July 2022.
2. IBM, “Artificial Intelligence,” June 3, 2020.
3. Gartner, “Gartner survey 37 percent of organizations have implemented AI in some form,”
2019.
4. Grand View Research, “Artificial Intelligence Market Size Report, 2022-2030,” 2022.
5. McKinsey, “The state of AI in 2021,” December 8, 2021.
6. Accenture, “The art of AI maturity,” 2022.
7. Ibid.
8. https://swisscognitive.ch/2021/02/23/ai-and-problem-gambling/
9. https://sportradar.com/integrity/bet-monitoring-detection/ufds-premium-services/
10. https://investors.sportradar.com/news-releases/news-release-details/sportradar-launches-
exchange-further-engage-bookmakers-anti
11. Sportradar, “Match fixing on the rise as global sports betting turnover surpasses €1.45 trillion
for the first time,” 2022.
12. Gartner, “2021-2023 Emerging Technology Roadmap for Large Enterprises,” 2021.
13. Ibid.
14. Betradar, “Marketing Services from ad:s.”
15. Ibid.
16. Ibid.
17. Bloomberg, “$422.37+ Billion Global Artificial Intelligence (AI) Market Size Likely to Grow at
39.4% CAGR During 2022-2028,” 2022.

38
 In closing, our successful 2021 and excitement for 2022 are tempered by the senseless loss of
life we’ve seen recently in Ukraine. My thoughts are with those who have been impacted by the
heartbreaking events, and our top priority has been to help ensure the safety of our employees
and their families in the region. Through our emergency relief program, Player Assist, we have
been able to extend financial and other forms of support to our extended Sportradar family who
are directly impacted by this crisis. We stand in solidarity with all those who wish for peace and
will continue to support the people hurt by these tragic events.
In 2021, our team grew both through acquisitions of companies that are well matched to ours
and organically in existing and new markets. We also welcomed our teams back to our offices,
reminding us of the value of a culture rooted in community and collaboration. Three thousand
employees across the globe have been the pillar behind our achievements in 2021. I want to
thank them for their tireless efforts to enable our strategic growth and drive our future success.
I also want to express my gratitude to our customers, partners and shareholders who are key to
our long-term growth. We look forward to a long and prosperous relationship.

Carsten Koerl,
Chief Executive Officer

39