As many industries, banking is undergoing a fundamental change because of the software revolution. No longer are banks competing only on interest rates and having the best traders, these days customer experience and having the best engineers are the focus. In this changing world, banks compete with new start-ups, the so-called Fintechs, and with large platform organisations such as Google, Facebook and Apple. At ING, we believe that staying ahead of the game means changing how we interact with our customers, no longer a traditional model of waiting for the customers to come to the bank through our website or apps, but to actively reach out to the customer with information that is relevant to him or her in order to make their financial life frictionless. Many of these changes are driven by reacting to all events that are relevant to the customer, and using streaming analytics to be able to reach out to the customer in milliseconds after the event occurs. Apache Flink is key for ING to achieve this. This presentation addresses how ING approaches the challenge, the role that Apache Flink plays, and the consequences regulations have on how we work with Open Source in general, and with Apache Flink (and data Artisans) in particular. This keynote takes place at Kino 3.

1. Taking away customer friction
through streaming analytics
FlinkForward Berlin
Ferd Scheepers. Chief Information Architect ING.
How ING uses data in real time with Apache Flink to enable it’s data driven journey
Sept. 2017

2. Market leaders Benelux
Growth markets
Commercial Banking
Challengers
The world of ING- The best global bank in the world
according to
Global Finance magazine
2
Customers
37 Million
Private, Corporate and
Institutional Customers
Countries
more than 40
In Europe, Asia, Australia,
North and South America
Employees
52,000

3. 3
1. Earn the primary relationship
2. Develop analytics skills to understand our customers better
3. Increase the pace of innovation to serve changing customer needs
4. Think beyond traditional banking to develop new services and business models
Empowering people to stay a step
ahead in life and in business.
Simplify &
Streamline
Operational
Excellence
Performance
Culture
Lending
Capabilities
Purpose
Customer
Promise
Strategic
Priorities
Enablers
Creating a differentiating customer
experience
Clear and Easy Anytime, Anywhere Empower Keep Getting Better

4. Trends in the banking landscape continue to evolve, our
strategy is there to adapt and come out on top
4
Customer
Behaviour
Competitive
Landscape
Technology
Fintech
SocietyRegulation
People’s time is precious;
they don’t want to spend it
on finance
Products have become commoditised; the only
way to differentiate is through the experience
Regulatory uncertainty
continues
Digitalisation is erasing
borders
Ability to leverage new
technologies will define
future competitive
advantage

5. “ING is an IT company with a banking license” – Ralph
Hamers
5

6. In 2017, a few companies are making huge amounts of money
with data. The data world is maturing.
6
The Economist: “The world’s most valuable resource is no longer oil, it’s data”

7. 5 years ago, everybody needed data scientists, now we need
data engineers, we are out of experimentation only.
7

8. Our data driven journey started around 5 years ago, even
before the CEO vision. Enter the ING Data Lake architecture.
8
Governance, Risk and
Compliance Platform
Events to
evaluate
Information
Service Calls
Data Load
Data
Feeds
Information
Service
Calls
Data
Export
Search
Requests
Report
Queries
Understand
Information
Sources
Understand
Information
Sources
Deploy
Decision
Models
Deploy
Real-time
Decision Models
Understand
Compliance
Report
Compliance
Information
Service
Calls
Data
Export
Advertise
Information
Source
Information
Owner
Information Systems
System of Record
Applications
Distribution Layer
Customer
Centric Core Layer
EnterpriseServiceBus
New Sources
Third Party Feeds
Internal Sources
Fulfilment Layer
Generic & Support
Services Layer
Data Lake
INFORMATION WAREHOUSE
DEEP DATA
Advanced Data
Provisioning
Catalog
Interfaces
Other
Data Lakes
Other
Data Lakes
Inter-lake
Exchange
Line of
Business
Interaction
Data Marts
Search Data
Information
Integration &
Governance
INFORMATION
BROKER
OPERATIONAL
GOVERNANCE
HUB
CODE
HUB
Decision Model
Management
10001
01011
01101
Line of Business
Insight
Simple,
Ad Hoc
Discovery
and
Analytics
Reporting
DataLakeRepositories
Shared
Operational
Data
ASSET
HUB
ACTIVITY
HUB
OPERATIONAL
STATUS
Descriptive
Data
INFORMATION
VIEWS
CATALOG
Deposited
Data
Harvested
Data
DEEP DATA
INFORMATION WAREHOUSE
Data Lake Operations
MONITOR WORKFLOW
Self-service
Data Access
Deploy
Real-time
Decision
Models
STAGING AREAS
Enterprise IT
Interaction
Information
Ingestion
Publishing
Feeds
Real-time
Interfaces
Generic
Real-time
Analytics
Decisions
STREAMING
ANALYTICS
Notification
Events

9. In 2015, we started to work on the Touch Point Architecture,
loosely based on platform thinking in the automotive
industry.
9

10. We identified the essential components of the new modular
bank, focussing on the customer and customer interaction
10
Authentication/Security provides the components and their interactions that facilitate a shared
security architecture, covering Authentication, Security Tokens, Authorisation and Risk Engine
interaction. Authentication/Security is based on the principles of ‘Stateless Design’, ‘Use of industry
standards’, ‘Channel-agnostic Authentication means’ and ‘Means-agnostic Business API’s’.
Notification brings the bank to its Customers. A push mechanism shifts ING towards becoming a
proactive bank versus a reactive bank. We will provide meaningful information to our Customers at all
time. We know the customer extremely well and add value by using pro-active notifications.
Standardised Global Party, Product & Arrangement Management (PPAM) defines a common data
exchange model and a set of interaction patterns that support globally reusable components with
regards to parties, products, agreements and mandates. All PPAM processes, exchange models and
data models are aligned in a standardised Global PPAM. Global PPAM has a standard interaction
pattern and a single API, with no boundaries between segments and countries.
Next to authentication, there is a specific Authorization workflow to request explicit consent of a
customer (by ‘signing’) for executing actions like a payments or changing an agreement.
Customer Context is about what we already know about the customer (previous locations,
arrangements, his financial position, etc.), about the environmental context (world news, news about
ING, what happens in my region) and what we know about the current customer’s touchpoint
(location, device, time). Events are about a specific touchpoint, and as such influence the current
interaction, but can also update person context as well as environment context for future analytics.

11. 11

12. Offering our customer actionable insights is the foundation of
the new banking experience
1212
Relevant
Actionable
Real Time
Personal

13. To deliver on the promise of TPA, we combined some
workstreams, and started to look at solutions already in
place.
13

14. Except for Coral, none of the existing solutions really suited
the architecture, and Coral turned out to be hard to develop.
14
Streaming Computing
Web, Apps and Services
Event Bus
Big Data (Not Realtime)
other data sources,
Lakes, SoR and services
REST
API
STREAM
API
OLTP
API
Data Science Scenario testing
● Low Latency Data store
● Streaming computing on event Bus
● In-Memory Big Data computing
Data Store
Mobile and Web

15. As we started looking at patterns for streaming analytics, they
always looked alike. Fraud and customer use cases are
identical.
15
Producer Raw event What is it Relevant event
What to
do with it
Outcome
Producer
Producer
Producer
CEP
Filtering
Enriching
Rules
Model scoring
Artificial Intelligence
Machine Learning

16. To re-use all the analytics, and to share state, we decided on
one global platform for all streaming analytics in ING.
16
And we did a beauty contest for the technology choice

17. The result is an Event based architecture, with Apache Kafka
and Apache Flink at the heart of it. Delivered as one platform.
17

18. CCN
The Operating Model for all TPA platforms is global, one tribe,
one product owner, a platform squad, potentially satellite
squads
18
Platform
tribe
Flink Platform
Squad
SAS platform
squad
X
Z
Y
Feature
squad1
Feature
squad 2
Satellite
Satellite
Satellite
Organization • Platform team builds the core
• Satellite teams across different regions do first line support
• Feature teams build on top of the platform
Code base • Maintained by Platform team
Deployment • Handled by platform team
• Multiple instances in different regions under same Life Cycle
Management
Use case jobs • Build and maintained by feature team
Support/
Monitoring
• Platform squad and satellite squads will both perform monitoring.
• First-line support will be done by the satellite squads. Second-line
support will be done by the platform squad.

19. Multiple startups have declared ETL dead, and that streaming
will take over. Is batch just a slower version of real time?
19

20. But I have a different opinion on where we are in this journey
20
Ferd Scheepers, Chief Information Architect ING

21. Choosing how to address the needs of the enterprise will
determine who will succeed in this space.
21

22. 22
ING is championing an Open Metadata and Governance that
will allow…
… metadata to be captured when the data is created, moved with the data and
be augmented and processed by any of the vendor tools.

23. All enterprises have a heteregeneous data landscape, and the
need for managing meta data across technology boundaries.
23

24. •
24
The Open Metadata and Governance initiative will allow easy
implementation of meta data in any platform
• Common base:
• Automate the collection, management and use of metadata across an enterprise
An enterprise catalog of data resources that are transparently assessed,
governed and used in order to deliver maximum value to the enterprise.
an open set
of APIs
an open set
of metadata
types
an open set of
exchange
protocols
Governance Access
Frameworks
Discovery

25. 25
We will deliver through Apache Atlas both a meta data
highway, and an implementation that can be used by all
Open and
Unified Metadata
Metadata
repository
Apache Atlas
Metadata
repository
IBM
Metadata
repository
Flink
Open Metadata Repository Service
OMRS
Open Metadata Access Service
OMAS
Components defined
and being developed
by Open Metadata &
Governance project
Meta data
highway

26. 1. Tink, a temporal graph analytics library for Apache Flink.
13-9 at 14:00 – Palais Atelier
2. Fast Data at ING – building a streaming data platform with Flink and Kafka
13-9 at 15:20 - Kesselhaus
Drop me a note at Ferd.Scheepers@ing.com
Or: @Ferdscheepers
Before I sign off, there is more from ING tomorrow:
26