eBook

A Primer on
Using Data + AI
for Fraud Prevention
Abstract
A Primer on Using Data + AI for Fraud Prevention
Fraud is a costly and growing problem. For every $1 of fraud, companies pay 3.36x in chargebacks and
replacement and operational costs. Identity fraud losses soared to $56 billion in 2020. Fraudsters continue to
increase the scale, speed and sophistication of attacks — from 2019 to 2020, fraud grew at an average rate of
33% — threatening the revenue and growth of companies. This eBook explains how data and machine learning
are ideal applications for fraud prevention — especially in the face of modern threats — and how to future-
proof your organization using the Databricks Lakehouse Platform. It also explains how financial services leaders
are using data and AI to combat fraud and includes two Databricks Solution Accelerators with easy-to-use
and best-practice notebooks so you can get a jump start on fraud detection.

A Primer on Using Data + AI for Fraud Prevention 2

Contents
1 . The growing cost of fraud 4

2. Why machine learning and AI are ideal

applications for fraud prevention 5

3. Data and ML challenges in fraud prevention 7

4. How a lakehouse approach to data and AI

simplifies fraud prevention at scale 9

5. Databricks Fraud Prevention Solution Accelerators 14

6. Customer Story: FINRA 16

7. Customer Story: Coins.ph 17

8. Conclusion 19

A Primer on Using Data + AI for Fraud Prevention 3

The growing cost of fraud
Fraud is a costly and growing problem. For every $1 of fraud, companies pay 3.36x in chargebacks and
1
replacement and operational costs.1 Identity fraud losses soared to $56 billion in 2020, and with more
businesses and end users keeping their sensitive information online, bad actors are devising new tactics to
defraud them both. In the past few years, we’ve seen huge data breaches that have flooded the dark web with
sensitive personal and financial information that can be used for account takeover and identity theft. The
For every $1 of fraud,
unfortunate truth is: Fraud will affect your industry.
companies pay 3.36x
in chargebacks and • $300 billion, or up to 10% of total healthcare expenditure, could be attributed to healthcare fraud in
replacement and
the U.S.2
operational costs
• $1.45 trillion lost turnover as a result of financial crimes, representing 3.5% of global turnover3

• $150 billion lost by government agencies to a wide range of fraud, representing 10% of all fraud4

Fraud continues to grow at 33% year over year, threatening the revenue and growth of companies. It impacts
customer satisfaction, loyalty and the bottom line.5

Moreover, fraudsters’ methods are becoming more technologically advanced. The Association of Certified
Fraud Examiners (ACFE) and PwC have shared how organized crime and large-scale fraudsters are increasing
Fraud continues to grow
at 33% year over year, the scale, speed and sophistication of fraud attacks.6 One of the newest methods involves using machine
threatening the revenue learning (ML) and other automation techniques to commit fraud that legacy approaches can’t detect, including
and growth of companies phishing emails, identity theft and forgery, phone and location spoofing, and the emulation of user behavior.
Experts and industry leaders are now looking to machine learning and AI to understand and get ahead of fraud.
Fraudsters are using AI. Why aren’t you?

1
CNP Fraud Costs U.S. Merchants $3.36 for Every $1 of Direct Fraud Loss
2
U.S. sentences 14 to combined 74 years in prison for healthcare fraud
3
Refinitiv Survey Report: Revealing the true cost of financial crime
4
McKinsey on Government Perspectives: Adopting AI, automation, and advanced analytics in governments
5
Fraud rate rises 33% during COVID-19 lockdown
6
ACFE and ABFA Fraud Resources

A Primer on Using Data + AI for Fraud Prevention 4

Why machine learning and AI are
ideal applications for fraud prevention 2
Fraud detection is the task of predicting unauthorized access to or use of accounts, credit cards and other
sensitive information. One salient method of recognizing fraudulent activity is to leverage machine learning
models to dynamically detect fraudulent transactions. ML models can be trained on a set of data — for example,
account history, credit card transactions, personal information and more.

With the availability of data and with advances in machine learning, fraud prevention is a key area in
which machine learning is changing both workflows and outcomes, allowing organizations to stay ahead of
criminals who are only growing more technologically advanced. Today’s businesses are facing an increasingly
sophisticated enemy that attacks, responds and changes tactics extremely quickly. With data analytics and
machine learning, companies can get ahead of threats. Below, we discuss the main reasons why machine
Fraudsters constantly learning is especially well suited for taking on fraud.
adapt their tactics,
making them difficult to Fraud hides under massive amounts of data
detect by humans
The most effective way to detect fraud is to examine the overall behaviors of end users. Looking at transactions
or orders is not enough — we need to follow the events leading up to and after the transaction. This culminates
in a lot of structured and unstructured data, and the best way to detect fraud in such huge volumes of data is
with machine learning and AI.

Fraud happens quickly

When a machine learning system gets updated in real time, this knowledge can be used within milliseconds to
update machine learning models and prevent an attack.

Fraud is always changing

Fraudsters constantly adapt their tactics, making them difficult to detect by humans — and impossible to
detect by static rules-based systems, which don’t learn. Machine learning, however, can adapt to changing
behavior.

A Primer on Using Data + AI for Fraud Prevention 5

 Fraud is unnoticeable on the surface
To the naked eye, fraudsters and good users don’t appear any different from each other. Machine learning
has a deeper and more nuanced way of viewing data, which helps avoid false positives.

Machine learning uses statistical models to look at past outcomes and anomalies to predict future
outcomes. A machine learning system can learn, predict and make decisions as data arrives in real time.

The benefits of data and AI for fraud prevention

• Highly accurate results: Advanced and data-driven behavioral analysis means fewer false
positives

• Less need for manual review: Machine learning automates processes in which behaviors can be
learned at the individual level and anomalies can be detected

• Ability to prevent fraud without impeding the user experience: AI brings automation to the
Coupled with human talent process seamlessly and prevents fraud in advance without burdening users
and experience, data and AI
work together to constantly • Lower operational costs than other approaches: With less manual work and more automation,
learn and adjust to new user data and AI require fewer resources and preempt losses associated with fraud
behaviors and trends
• Frees up time for teams to focus on more strategic tasks: Most companies are not in the
business of fraud detection, and a machine learning fraud prevention process can help them
focus on core activities

• Adapts quickly: Coupled with human talent and experience, data and AI work together to
constantly learn and adjust to new user behaviors and trends

When it comes to operationalizing data and AI to build customer relationships and drive higher returns
on equity, fraud should be considered a top priority. Curbing fraudulent or malicious behavior — from
fraudulent securities trading to money laundering — is key to mitigating negative revenue impact.

A Primer on Using Data + AI for Fraud Prevention 6

Data and ML challenges
in fraud prevention 3
We’ve established that data analytics is the best way to prevent fraud. But there are challenges when it
comes to execution, especially regarding the data. As the amount of data increases, so does the complexity
of understanding it all. Where will all this data be stored, and in what format? How will it be retrieved and used
again? How will the data be prepared for BI and machine learning? These issues are easier to solve when
data is limited vs. when it scales by 100x or more. Companies that want to build their own machine learning
infrastructure need to think from the beginning about supporting massive data growth — and, along with it — an
increasingly complex surface area of problems to tackle. Here are a few objectives to keep in mind when trying
to leverage data and AI for fraud prevention:

• Getting high-quality, clean data and maintaining a rich feature store for ever-evolving fraud patterns
Companies that want
to build their own ML • Using multiple vendors that are siloed in capabilities — for example, ETL, data science and scaling on
infrastructure need to demand
think about supporting
massive data growth • Enabling multiple data teams to scale their data pipelines and collaborate in the cloud

• Meeting ML challenges by training complex models with hundreds of features on gigabytes of data

A Primer on Using Data + AI for Fraud Prevention 7

 With its high operating costs and lack of agility, a disjointed operating model cannot successfully
combat fraud. In order to succeed in the journey of digital transformation, financial services
institutions (FSIs) must adapt to today’s agile and data-driven times while modernizing their
approach to data and analytics.

Today, the fraud ecosystem is disjointed

Data Warehousing Data Engineering Streaming Data Science and ML

SILOED DATA TEAMS DECREASE PRODUCTIVITY

Data Analysts Data Engineers Data Engineers Data Scientist

DISCONNECTED SYSTEMS AND PROPRIETARY DATA FORMATS MAKE INTEGRATION DIFFICULT

Amazon Redshift Teradata Hadoop Apache Airflow Apache Kafka Apache Spark Jupyter Amazon SageMaker
Azure Synapse Google BigQuery Amazon EMR Apache Spark Apache Flink Amazon Kinesis Azure ML Studio MATLAB
Snowflake IBM Db2 Google Dataproc Cloudera Azure Stream Google Dataflow Domino Data Lab SAS
Analytics
SAP Oracle Autonomous Confluent TensorFlow PyTorch
Data Warehouse Tibco Spotfire

SILOED STACKS INCREASE DATA ARCHITECTURE COMPLEXITY

Analytics and BI Transform

Data Machine
science learning

Data marts Extract Load Real-time

database
Data prep

Data warehouse Data lake Streaming data engine Data lake

Structured data Streaming data source

Structured, semi-structured Structured, semi-structured

and unstructured data and unstructured data

A Primer on Using Data + AI for Fraud Prevention 8

How a lakehouse approach to data and AI
simplifies fraud prevention at scale
The Databricks Platform leverages the openness and simplicity of the lakehouse — a new, open architecture
4
that combines the best elements of data lakes and data warehouses. The lakehouse system design has similar
data structures and data management features to those in a data warehouse, but directly leverages the kind of
low-cost storage used for data lakes.

A lakehouse architecture is well suited for fraud prevention because it provides financial services companies
with a fully managed cloud platform that accelerates innovation by unifying data engineering, data analysis and
data science with the rest of the business. And a lakehouse environment can easily extract and manage the
massive, real-time, changing and differing data types that are needed to detect fraud.
A lakehouse architecture
is well suited for fraud
prevention because it Late 1980s 2011 2020
provides a fully managed
cloud platform that unifies D ATA WA R E H O U S E D ATA L A K E LAKEHOUSE
data engineering, data
analysis and data science
with the rest of BI Reports Data ML Real-Time Reports BI Streaming Analytics Data Science
Science Database
the business
BI ML
Data Marts ETL Data Marts
Data Prep
and Validation

ETL
0 1 0 1 01 10 1 0 0 1 0 11 01 0 0 1 0 1 00 10 1 0
Data Lake

External Data Operational Data Structured, Semi-Structured, and Unstructured Data Structured, Semi-Structured, and Unstructured Data
SOURCE: DATABRICKS

The Databricks Lakehouse Platform combines the best elements of data lakes and data warehouses — delivering the data
management and performance typically found in data warehouses and the low-cost, flexible object stores offered by data lakes

A Primer on Using Data + AI for Fraud Prevention 9

 Unifying batch and streaming pipelines is key to ML success
One of the most common problems FSIs face is dealing with massive volumes of data — both batch and
streaming — across disparate sources. Many FSIs turn to data lakes to aggregate their big data cost-
effectively, but this creates a new set of challenges around data management and governance.

Databricks enables organizations to overcome these challenges with a lakehouse architecture powered
by Delta Lake. An open source technology, Delta Lake is natively integrated within Databricks to provide
reliability and performance. It acts as a storage layer on top of your data lake that enforces data quality
with ACID transactions. FSIs can ingest structured, semi-structured and unstructured data, both in batch
and streaming, into a single Delta Lake to ensure that the supply of data is clean and usable. The scalability
of Databricks enables organizations to then process and query this data for near real-time insights.

A Primer on Using Data + AI for Fraud Prevention 10

 How cross-team collaboration boosts AI innovation
Fraud is not solely addressed by AI — it requires domain expertise (e.g., rules) combined with AI-generated
insights — so business analysts have to visually inspect patterns and rules efficacy through MI/BI
dashboards. To combat fraud, business analysts and data scientists must have the same set of data, which
is possible in a lakehouse. Because Databricks’ approach to fraud detection often involves a combination of
rules and ML, it’s well suited to meet the needs of a diverse set of personas required to create rules-based
ML models.

The ability for users to collaborate across multiple workspaces while providing isolation at the user level
is critical in financial services. Databricks’ fraud solutions address the key areas of scalability in the cloud,
fraud prevention workflow management and production-grade open source ML frameworks. They allow
organizations to build and maintain a modern fraud and financial crimes management infrastructure by
increasing alignment between different internal teams.

The Databricks Lakehouse Platform unifies data teams so they can collaborate across the entire data and AI workflow

A Primer on Using Data + AI for Fraud Prevention 11

 The power of business insights for fraud analytics
Databricks centralizes your data for easy access. You and your team of data analysts can easily and
directly connect and query your most complete and recent data in the data lake — using Delta Lake and
Databricks SQL to discover anomalous behaviours and perform root cause analysis. Connectors with
popular BI tools like Tableau and Power BI allow your analysts to use their preferred BI visualization and
reporting tools for real-time insights that can help stop a malicious attack.

Unlock the potential of your data on a single platform

Financial services institutions can minimize losses and improve customers’ trust with fewer false declines.
Fraud has diversified rapidly and requires an agile approach. There’s no one type of fraud, and fraudsters
are constantly changing techniques. We need a platform that allows real-time access to new data and
faster delivery to insights and new rules. The Databricks Lakehouse is a high-performance query engine
designed for agility and performance on big data workloads and machine learning.

Additionally, any AI-based model must comply with strict regulatory requirements, necessitating as-of
code, data and models. By combining MLflow and Delta Lake capabilities, models and rules can be created
with a high degree of governance and trust in your data. An independent team of experts can ensure the
right parameters are used against the right set of data, resulting in the right outcome.

A Primer on Using Data + AI for Fraud Prevention 12

 Unlock the value of data lakes for BI and ML
Databricks provides a Lakehouse Platform that helps financial services institutions democratize data for downstream fraud
analytics and AI — minimizing risk while accelerating transformative innovation.

All your financial services data Reliable, real-time processing Analytics capabilities for every use case

Banking and Structured batch

investments
Structured batch
Customer data
PRODUCTION MACHINE LEARNING
Unstructured batch Risk Fraud Anomaly
Credit
DATA LA K E H O U SE management detection detection
Unstructured stream
Market ticker
Process, manage BI REPORTING AND SCORECARDING
Unstructured stream and query all
Social media Scorecards Dashboards
your data
Unstructured batch
Web scrapes

DATA CHALLENGE THE DATABRI CK S L A K E H OUS E F OR

FI N AN C I AL FRAUD P R E VE N T I ON

DATA INGEST: Processing batch and streaming Connect traditional data with alternative data insights
data can be slow and error-prone, impacting
downstream analytics

DATA LAKE MANAGEMENT: Data silos Easily handle large volumes of data from multiple sources
can limit the ability to gain a complete view of (transactions, geospatial, demographics, etc.) built on a strong
the customer privacy foundation

DATA QUERY: Fragmented, siloed and Ability to rapidly and inexpensively experiment, manage and
inconsistent data sources for BI and data science push out at scale from a single platform

A Primer on Using Data + AI for Fraud Prevention 13

Databricks Fraud Prevention
Solution Accelerators 5
Based on best practices from our work with leading brands, we’ve developed Solution Accelerators for fraud
data analytics and machine learning use cases to save weeks or months of development time for your data
engineers and data scientists.

Both accelerators demonstrate the innovative use of anomaly detection with geospatial analysis in fraud
prevention, and show how easy it is to get started.

How to build a rules-based AI model to combat financial fraud

Preempt fraud with rule-based patterns and select ML algorithms for reliable fraud detection. Use anomaly
detection and fraud prediction to respond to bad actors rapidly.

• Build trust and loyalty among your customers

• Respond to fraudulent activities fast and reduce operational costs

• Eliminate data complexity and implement fraud prevention at scale

A Primer on Using Data + AI for Fraud Prevention 14

 How to identify credit card fraud through AI and geospatial analysis
Learn how geospatial analysis and a lakehouse framework enable organizations to better understand
customer spending behaviors and detect abnormal card transaction patterns using machine learning
in real time. Geospatial data can enhance fraud prevention to avert losses and build customer trust.

Geospatial clustering helps identify customer spending behaviors

• Leverage geospatial data and rules-based AI to combat financial fraud

• Identify abnormal behavior and detect threats to better protect users

• Personalize the banking experience with granular, customer-centric insights

Solution Accelerators are an easy and simple way to approach fraud.

Visit our Solutions Accelerator page to get started.

A Primer on Using Data + AI for Fraud Prevention 15

 6
Customer Story:

The Financial Industry Regulatory Authority (FINRA) is an independent, nongovernmental organization

responsible for protecting investors by ensuring that the U.S. securities markets operate in a fair and honest
manner. FINRA oversees 12 markets and exchanges, 3,700 firms and more than 600,000 brokers. FINRA deters
misconduct by enforcing rules, detecting and preventing wrongdoing in the U.S. markets, and disciplining

“
members who break the rules.

With Databricks, we have Challenge

one cohesive end-to-end • Disjointed systems, workflows and teams caused FINRA to struggle with leveraging big data analytics to
process with one single detect fraud and protect investors
unified team working on
protecting the securities Use case
markets.” • Data ingest and ETL

• Leverage machine learning to detect fraudulent securities trading

— Saman Michael Far
Senior Vice President Why Databricks?
of Technology at FINRA
• Fosters collaboration among data scientists, engineers and analysts to boost innovation

• Provides a fully managed platform that removes infrastructure complexity so they can focus
on the data rather than DevOps

With Databricks, teams can quickly iterate on ML models and scale detection efforts to hundreds of billions of
market events per day. As a result, FINRA has significantly improved fraud prevention, leading to a more secure
financial future for investors in the U.S.

Impact
• Analyze 100 billion stock market events per day to identify fraud and wrongdoing

A Primer on Using Data + AI for Fraud Prevention 16

Customer Story:

7
As a key digital payments platform in the Philippines, Coins.ph needs to be able to perform accurate, insightful
financial audits and prevent fraud — in real time. With more than 10 million customers accessing digital payment
services for local and international remittances, bill payments and online shopping, Coins.ph needed to find a
way to move beyond development operational processes and address more advanced business challenges.
With Databricks, Coins.ph was able to harness richer data insights to deliver ML-powered fraud detection and

“
anti-money-laundering solutions at greater speed while optimizing financial reconciliation.

With the Databricks Challenge

unified data analytics • Legacy analytics system built on EMR struggled to turn massive data into meaningful insights,
platform, our operational requiring a significant effort from their data team to maintain and use the platform for development
costs have been reduced operational tasks
by about 70% due to
Use case
more efficient cluster
management.” • Data ingest, ETL and machine learning

• Develop ML experiments and prototypes to address and deploy fraud detection

— Dmitry Ustimov, Data
Architect at Coins.ph • Use the Databricks Platform to enable standardized analytics and create new rule sets for anti-
money-laundering compliance

Why Databricks?
• Provides a unified platform for data teams to collaborate on data preparation and analytics and to
prototype new models

• Delta Lake ensures consistent data pipelines that feed data downstream for ML

• MLflow provides easy development and tracking of new ML models

Using Delta Lake to ingest large volumes of data in real time, data engineers at Coins.ph are able to bring greater
reliability to the data lakes and get up-to-date insights to develop more robust and scalable data pipelines for

A Primer on Using Data + AI for Fraud Prevention 17

 fraud detection. MLflow simplifies and streamlines the ML lifecycle, allowing data teams to easily track ML

70%
experiments and quickly develop new prototypes to address fraud detection.

Impact
• 14x reduction in complaints received

• 70% operational cost reduction in compute costs

operational cost reduction
in compute costs • 50% infrastructure cost reduction

A Primer on Using Data + AI for Fraud Prevention 18

Conclusion 8
When financial services institutions are able to predict fraudulent activity, they can greatly mitigate revenue loss
and customer churn. While machine learning has come very far and is impacting our lives in countless ways, it’s
still just getting started. Soon, the question of whether to leverage machine learning for fraud won’t be a question
but a necessity. In the meantime, the lakehouse is an imperative first step to take in fraud prevention. Learn more
about Databricks’ capabilities and the ways in which the lakehouse can take on fraud.

Stay up to date on the latest events, case studies and

solutions for financial services at dbricks.co/fiserv.

A Primer on Using Data + AI for Fraud Prevention 19

Databricks is the data and AI company. More than 5,000 organizations worldwide — including Comcast, Condé Nast, H&M, and
over 40% of the Fortune 500 — rely on the Databricks Lakehouse Platform to unify their data, analytics and AI. Databricks is
headquartered in San Francisco, with offices around the globe. Founded by the original creators of Apache Spark™, Delta Lake
and MLflow, Databricks is on a mission to help data teams solve the world’s toughest problems. To learn more, follow Databricks
on Twitter, LinkedIn and Facebook.

© Databricks 2021. All rights reserved. Apache, Apache Spark, Spark and the Spark logo are trademarks of the Apache Software Foundation. Privacy Policy | Terms of Use