pwc-data-mesh.pdf

Data Mesh
Just another buzzword
or the next-generation
enterprise data platform?
www.pwc.de

Foreword
Enterprises across industries want to generate more value
from data. This might be by digitalising and automating
processes, introducing business analytics and AI,
implementing new data-driven business models or even
directly monetising data. To do this, accessible, high-
quality data and effective data governance are essential.
The growing number and scale of data use cases across
enterprises are also making it vital for companies to share
and reuse data across organisational boundaries.
Because of this, many enterprises have invested in
modernising their data management and analytics
systems and have ramped up their in-house capabilities
for data engineering and data science. Centralised
data platforms (data warehouses and data lakes)
and centralised teams (e.g. data labs and centres of
excellence) have been the main paradigm for making
effective use of the limited personnel capable of working
with new technologies. However, centralisation by design
can also create organisational difficulties. As business
units develop more and more data use cases and
implement them at scale to achieve growth and efficiency
ambitions, these difficulties are becoming increasingly
noticeable at the business unit level. Therefore,
decentralising data platforms and data engineering
capabilities is now a hot topic for data and analytics
leaders in many organisations.
In line with the idea of decentralisation, the data mesh –
a new data architecture paradigm – is currently getting a lot
of attention in the global enterprise data management and
analytics community. But is this just another buzzword,
or does it offer a next-generation data architecture for
success over the next ten years?
In this study, we take a critical look at the data mesh theory
based on our practical experience from implementing data
platforms and developing data and analytics capabilities.
We complement our own experiences with the findings of
our latest survey among German IT managers and take
a closer look at a well-known company that is currently
implementing a data mesh.
Centralized data
platforms by design
create potential
organizational
bottlenecks.
Does the idea of
decentralisation, the
data mesh, offer a
next-generation data
architecture or just
another buzzword?
Data Mesh 2

Data Mesh 3
Contents
A Introduction............................................................................4
B The journey to the next-generation data platform..........12
C How to build the next-generation data platform............ 20
D
Why and how is IKEA implementing
its data mesh?..................................................................... 29
References..................................................................................................... 32
PwC Data Mesh Research 2022.................................................................... 32
Authors........................................................................................................... 33

Data Mesh 5
1
Evolution and current state of
enterprise data platforms
Ever since the rise of data warehouses in the 1990s,
enterprises have used specific data platforms to manage
data for analysis, with the aim of gaining insights for
business decision-making. Today, data warehouses are
an integral part of corporate IT systems. In our survey
of German IT managers, 84% confirmed to have a data
warehouse in use within their organisation.1
Although data
warehouses have undergone continuous technological
development – in particular, the shift to cloud-based
systems – their main purpose is and has always been
to support managerial decision-making by providing a
“subject oriented, non-volatile, integrated, time variant
collection of data”2
. Information is provided in reports
and on dashboards. These are built on top of data
warehouses, which have traditionally been implemented
using relational database management systems.
In response to the growing volume, velocity and variety
of data3
, many companies have added a data lake to their
enterprise architecture. In our survey, 78% of IT managers
stated that they are using or implementing a data lake in
their organisation. In contrast to data warehouses, data
lakes are designed to store any kind of data. This means
that data in a data lake can, amongst others, be:
• High-volume – e.g. data produced by wind turbines,
power generators and smart meters, which can easily
exceed 500TB of new data per day.
• High-velocity – e.g. streams of real-time sensor data from
manufacturing equipment in modern automotive plants.
• High-variety – e.g. documents, images, audio and video
managed by online media companies.
Data lakes also enable advanced analytics use cases
such as training machine learning algorithms, which need
large amounts of raw data to detect patterns and generate
new business insights.
Early adopters of data lakes in the early 2010s started with
their own on-premises implementations based on Hadoop
technology. Today, however, most companies have
shifted to managed big data services on Microsoft Azure,
Amazon Web Services (AWS) or Google Cloud Platform
(GCP), as these services are more cost-effective and offer
better scalability.
As the results of our study show, data warehouses and
data lakes coexist in most enterprises and are used for
different purposes. Some companies have multiple data
warehouses and data lakes in place due to complex
corporate structures, but the strategic goal for companies
is typically to consolidate their data assets in one central
platform. This facilitates access to data and simplifies
data platform governance.
To better integrate data warehouses and data lakes,
companies are increasingly adopting the data lakehouse
architecture, which involves stacking the data lake and
the data warehouse. This means that the data lake stores
any kind of raw data needed for analytical purposes, and
a subset of that raw data is then processed in the data
warehouse. A data lakehouse thus delivers integrated,
high-quality data and applies standardised KPIs to
business intelligence and reporting use cases.
Due to the increasing amount of data stored in and
processed through distributed data repositories, data
lakes or data lakehouses, many companies have recently
started implementing a data catalog to keep track of data’s
location and movement across the system landscape
based on its metadata. Additionally, in the light of emerging
data management concepts like data fabric4
, data catalogs
are increasingly being extended with AI capabilities to
automatically detect relations and dependencies between
data objects and maintain them in a knowledge graph that
facilitates data discovery and insights generation. However,
from our perspective such active metadata management
practices are still in their infancy today.
Although the technology used in enterprise data platforms
has become more advanced over the years, there remain
some limitations to using data in enterprises. Based
on our own research5
, 3 out of 4 C-level executives
consider data transparency, reliability and trust as critical
improvement areas.
The biggest challenges around data and analytics – as
assessed by these C-level executives – are the lack of data
literacy, low data quality, limits of legacy technology, and
unclarity about how to operationalise developed use cases.6
1, 5, 6

PwC Data Mesh Research (2022).
2
Inmon, W.H. (1992). Building the Data Warehouse.
3
Laney, D. (2001). 3d data management: Controlling data volume, velocity and variety.
4
Gartner (2021). Data Fabric Architecture is Key to Modernizing Data Management and Integration.

Our own experience from various projects is in line
with these findings. In recent years, we have seen
many companies develop data and analytics use cases
as small-scale proof-of-concepts or pilot projects;
however, only a few of them have been successfully
operationalised. This is due to a lack of trust in the
data, lack of skills among business users to work with
new data-driven solutions, and ongoing challenges in
quickly and reliably extracting data from old transactional
source systems.
1
The competence to work with and understand data.
2
Old IT systems.
Fig. 1 The biggest challenges around data and analytics for C-level executives
Lack of data literacy1
54%
54%
Low data quality
54%
Limits of legacy technology2
54%
Unclear use case
operationalization path
50%
Complex new technology
50%
Limited data availability
50%
Use-case ideation
32%
Insufficient data
analytics expertise
25%
Siloed data
Many companies develop
data and analytics use cases
as pilot projects; however,
only a few of them have been
successfully operationalised.
Data Mesh 6

Many enterprises have recently implemented self-service
analytics tools and processes to create more value
from data by making the data in their data platforms
more accessible to a broad range of users across the
organisation. While the adoption of self-service analytics
is increasing, only 8% of the surveyed companies have yet
reached a stage where everybody can conduct ad-hoc
data analysis – i.e. everyone has access to relevant data
and tools to perform analytics independently.7
As a result, only a small number of companies have so
far achieved their ambition of turning data into tangible
value for the business. More and more enterprises are
questioning their current approaches to data management
and are looking for better ways to organise their data
platforms and teams to create more value from data.
7
Fig. 2 Implementation of self-service analytics in German companies1
1
Percentages do not add up to 100% due to rounding.
8%
Everyone in the enterprise
can conduct ad-hoc
analytics independently.
There is an analytics team
in each business unit
that provides analytics
solutions to business
users on demand.
51%
There is a central analytics
organisation in place that
handles requests for new
analytics solutions.
39%
We don't have/offer any
self-service analytics.
1%
Data Mesh 7

Data Mesh 8
2
Three pillars for increasing
value with data
Data can be turned into business value by enabling
improved decision-making, business process optimisation
or automation, data-driven and AI-enabled products
and services, as well as new data-driven business
models. All these data use cases involve both costs and
benefits. On the one hand, they involve one-time costs for
developing new data pipelines, analytical models or data
visualisations, as well as recurring costs for operation.
On the other hand, these use cases deliver new insights
which can drive business growth, increase efficiency and/
or reduce risk.
Simply put, more use cases with a positive business case
will create more value for the business. If use cases are
deployed on a larger scale – for example, to more clients,
products or users – they will deliver higher value. And if
use cases can be readily scaled up, they will yield even
more value. For use cases that aim to create a competitive
advantage, time to market plays a crucial role, in turn
giving a good reason for companies to aim for quick
implementation.
In order to do this, it is vital for companies to have
sufficient skills and capacity in their data engineering and
data science teams. But this on its own is not enough: the
data itself is another critical factor.
As described by the widely used concept of “garbage in,
garbage out”, an analytical model or insight can only be
as good as its underlying data. And experienced business
decision-makers will only adopt and operationalise a
data-driven solution if they trust the underlying data –
i.e. if the results of data analyses are comprehensible and
show numbers that are consistent with other sources of
information.
Therefore, the first pillar for increasing
value with data is to enable rapid
development and operationalisation of
data use cases at scale.
Therefore, the second pillar for
increasing value with data is to improve
the quality and trustworthiness of data.
Therefore, the third pillar for increasing
value with data is to efficiently govern
data and use cases.
Some data quality issues can be tackled by data
engineers when preparing specific datasets for analysis,
but this is often a time-consuming activity which only
addresses the symptoms, rather than dealing with the
root causes. To sustainably resolve this challenge, it
needs to be tackled in the operational systems where the
data is entered or created. The scope of a true data and
analytics-driven system is therefore much broader than
just a data warehouse or data lake: it requires involvement
of multiple business and IT stakeholders to holistically
manage data.
Many companies have already launched initiatives to
address these pillars, but only a few have achieved
the desired results. As a result, new data management
methods are getting a lot of attention in the press and
social media – in particular, data mesh. Data mesh
proposes a new architecture paradigm to address our
three pillars, and is currently one of the most discussed
topics amongst data and analytics architects and
practitioners around the world.
As in any initiative, the more people are involved and
the more spread out they are across the organisation,
the harder it becomes to orchestrate activities, create
synergies, and ensure compliance with regulations and
corporate policies and standards. In fact, we see many
enterprises increasingly spending on data governance
and use case portfolio management to keep control over
data-related activities – which reduces the contribution
margin of data use cases.

Data Mesh 9
8
Dehghani, Z. (2019). How to Move Beyond a Monolithic Data Lake to a Distributed Data Mesh.
9
Dehghani, Z. (2020). Data Mesh Principles and Logical Architecture.
10
Ford, N., Richards, M., Sadalage, P., Dehghani, Z. (2021). Software Architecture: The Hard Parts.
11
3
Data mesh – the cure for
enterprise data management
challenges?
In 2019, Zhamak Deghani introduced a new data platform
architecture paradigm known as data mesh. Her initial
article argues that centralised, monolithic data platform
architectures – such as data warehouses and data lakes –
have failed. Particularly in enterprises with a wide variety
of data domains, data sources and data consumers.8
Often managed by centralised teams, centralised data
platforms do not scale well across organisations, and are
thus incompatible with our first pillar introduced above.
Hence, Deghani argues that data platforms should be
decentralised.
However, decentralisation involves certain risks, as explained
above in relation to our third pillar. Deghani emphasises
that decentralisation of data platforms should not result in
a landscape of fragmented data silos across the business;
instead, it should create a “harmonised, collaborative and
distributed ecosystem” which she refers to as a data mesh.
In 2020, Deghani proposed four fundamental principles for
a data mesh, which she refined in 2021 to give the following:
1. “Domain ownership of data”, which calls for the holders
of domain knowledge to enter into direct exchange
with data consumers – i.e. without having to go through
intermediaries and without interference from others.
2. “Data as a product”, which stresses a mindset of
sharing data and seeks to channel limited resources
to the uses most valued by data consumers.
3. “Self-service data platform”, which aims to lower the
barrier to entry and increase the utility of the data mesh
for data producers and consumers alike.
4. “Computational federated governance”, which
emphasises the need for automated, non-intrusive
control and enforcement of norms and rules.9,10
On paper, the data mesh concept seems to perfectly
address all three of our pillars, and it is resonating well
with IT managers. According to our survey, half of them
expect data mesh to improve the quality of data analytics
use cases, as well as to increase data reliability and
trust.11
However, few companies have fully implemented
a data mesh so far, so their benefits are yet to be proven
in practice.

Data Mesh 10
Fig. 3 Expected benefits of data mesh
Improved transparancy on
available data
5.
43%
Faster time to market for data
and anayltics use cases
6.
41%
Stronger collaboration
between teams
3.
47%
Increased synergies in data
and analytics operations
4.
45%
No benefits expected
7.
2%
Higher data quality and
analytics use cases
1.
55%
Increased data
reliability and trust
2.
51%
Many enterprises are aspiring to develop their data
platforms towards a data mesh, as shown by our survey:
93% of surveyed IT managers said that they had already
discussed about data mesh within their organisation,
and 83% of them want to adopt this concept. However,
more than a third (37.5%) of companies that want to adopt
the concept lack a data mesh strategy, which C-level
executives consider a major challenge.12
12

Data Mesh 11
To provide clarity on how to improve current data
platforms and organisational structures, and to help
you answer the question of whether a data mesh is the
right approach for your company, we embarked on an
expedition to a fictitious galaxy where an enterprise has
already established a next-generation data platform.
A data platform influenced by data mesh principles
and complemented with further best practices and
perspectives from our experts.
The next chapter describes the building blocks of this
next-generation data platform and explains how it works
in practice.
Fig. 4 Popularity and adoption of data mesh concept
36%
Our organization has discussed Data Mesh and we are planning to adopt this concept
Our organization is on the way of implementing certain elements of the Data Mesh concept
32%
Our organization implemented and adopted most elements of the Data Mesh concept
15%
Our organization has discussed Data Mesh but we are not planning to adopt this concept
10%
Our organization has not yet discussed the Data Mesh concept
7%

Data Mesh 12
B The journey to the next-generation data platform

Fig. 5 A new world: a data mesh-based enterprise
On our journey to a galaxy far away, we find an enterprise
that has implemented the three pillars and thus
successfully used data to increase business value. If we
look at its data platform and organisation, we can see
both centralised and decentralised capabilities effectively
working together. The decentralised capabilities enable
fast and scalable development and operationalisation
of use cases and ensure high-quality data (pillars 1
and 2), while the centralised capabilities enable efficient
governance of data and use cases (pillar 3).
More precisely, there are eight capabilities that differentiate
this data platform (and the organisation and processes
that go with it) from the typical data warehouses and data
lakes seen on Earth today. Four of them are decentralised
capabilities and four of them are centralised capabilities.
Decentralised
1. Data is owned, processed, and used decentrally
in domains.
2. Each domain creates and shares valuable data
products across the enterprise.
3. Data products are created and operated by
autonomous, interdisciplinary teams.
4. Data product teams have professional DataOps practices.
Centralised
5. All metadata about data and use cases is managed
in a central data catalog
6. Data products can easily be found and accessed
through a data marketplace.
7. Access to data products is mostly virtualised.
8. Data governance rules are centrally enforced for all
data products.
A look through our telescope shows the interplay of
these eight capabilities more clearly.
The following sections examine each of the eight
capabilities in more detail.
Data Mesh 13

Data Mesh 14
1
Data is owned, processed, and
used decentrally in domains
Every business function in the enterprise generates,
collects and uses data through its business processes.
For instance, the finance department records transactions
in the ledger system, the sales and marketing
departments collect data on their customers, and the RD
department creates data on new product features.
The enterprise has established several data domains
based on bounded business contexts such as accounting,
marketing or logistics (not necessarily corresponding
to the enterprise’s organisational structure). These data
domains have ownership of certain data objects, such
as financial transactions, customers or shipping orders.13
Ownership for any given data object is allocated to a data
product owner in the domain where that data object is
created and predominantly used. For example, customer
demographic data is created and heavily used in the
bounded context of marketing, so it is allocated to a data
product owner in the marketing data domain.
Each data domain has a data architect who properly
defines the logical data model and its key data objects
within the data domain based on functional business
knowledge and in accordance with enterprise guidelines
(e.g. regarding naming conventions or unique identifiers).
This allows effective communication between data
producers and consumers using a common language.14
13
Vernon, V. (2013). Implementing Domain-Driven Design.
14
Evans, E. (2003). Domain-Driven Design: Tackling Complexity in the Heart of Software.
Despite being owned by a single data domain, data
objects are used by multiple domains. For example,
a marketing analyst might want to conduct a price
sensitivity analysis based on sales transactions (owned by
the accounting data domain) and customer demographics
(owned by the marketing data domain). Therefore, data
domains are responsible for sharing their data with other
data domains in a usable way. This is where the “data
product” principle of the data mesh becomes important.
2
Each domain creates and shares
valuable data products across
the enterprise
A data product is not just a dataset containing relevant
information on a specific bounded context; it also
contains all necessary connectors to source systems,
transformation steps and access rights management. It is
provided to people across the organisation in an readily
consumable way – for example, using governed application
programming interfaces (APIs). There are different types of
data products to serve different purposes.
For instance, a dataset containing customer master data
such as name, address and demographic information is
a data product that is highly relevant for many functions,
such as sales, marketing, customer service and accounting.
This data product is created by the sales data domain in
the customer relationship management (CRM) system.
To ensure data quality, it might then be enriched with
data from an external service provider which checks
that postal addresses are correct. Internal data quality
checks are also carried out – for example, to identify and
remove duplicates. To ensure data consistency across
use cases, there is only one version of this data product
in the enterprise, and everybody requiring customer
demographic data uses this single data product as the
basis of their analysis. This is an example of a source-
oriented data product, whose main purpose is to
maximise the reuse of a specific dataset by a broad range
of data consumers across the enterprise. It demonstrates
how data products can become a vital part of a master
data management (MDM) implementation that reaches
beyond transactional systems. This broadened scope
increases the benefits offered by analytical use cases.
Data domains are responsible
for sharing their data with
other data domains in a usable
way. This is where the “data
product” principle of the data
mesh becomes important.

Fig. 6 Source-oriented and consumer-oriented data products
API
API
Input
data
Data
cleansing
Resulting
data
Metadata META
Consumer-oriented data product
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
Source-oriented data product
Published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
Source-oriented Data Product
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
API API
Input
data
Data
cleansing
Resulting
data
Metadata META
published
Published
Data Mesh 15
A marketing analyst might want to extend this customer
master dataset with information about customer
purchasing habits – for example, average monthly
spending over the last year. To that end, the analyst
combines the existing data products for customer master
data and sales transactions to create a new data product,
which can then be reused by other analysts across the
enterprise. This is an example of a consumer-oriented
data product, whose main purpose focuses on the needs
of a specific group of data consumers.

Data Mesh 16
15
Dehghani, Z. (2019). How to Move Beyond a Monolithic Data Lake to a Distributed Data Mesh.
3
Data products are created
and operated by autonomous,
interdisciplinary teams
The data product teams within each data domain work
autonomously to develop and operate their data product
within the guardrails set by the data domain architect.
This means that the teams are not dependent on external
resources, enabling them to work quickly and flexibly to
meet the needs of the business. The data product teams
are decentralised, being located in each data domain – a
fundamental difference to centralised data warehouse and
data lake teams. However, the various data product teams
still work on a shared platform to leverage key underlying
services, such as computing and storage infrastructures
or identity and access management. These services are
provided by a central platform team, but the data product
teams also have the freedom to independently deploy
additional services and resources on the platform to serve
the specific needs of their data product. For example, if
a data product team needs a graph database to model
their data product, they can deploy it on the platform
themselves, making them responsible for maintaining this
new component on the platform.
A data product team is composed of the data product
owner, additional business experts who understand specific
business processes and associated data, data engineers
who build scalable data pipelines and data models, and –
if required – data scientists who train advanced analytical
models and integrate them into the data pipelines.
Customer data includes personally identifiable information
(PII). As the enterprise is operating in the EU, the General
Data Protection Regulation (GDPR) dictates that this
information must be protected. Therefore, data product
owners are required to set access permissions for
each data product, and they also offer anonymised or
pseudonymised versions of the data products that can
be used for statistical analysis.
These examples show that data products have to meet
many different requirements. Therefore, the enterprise
has defined a list of criteria (inspired by Deghani’s first
article) that every data product needs to fulfil.15
Data product owners are responsible for ensuring that all
criteria are met at all times. Owners are offered incentives
in line with the quality and level of usage of their data
products across the organisation, as well as the business
value of the use cases they enable.
Data product owners are business experts rather than
technical data specialists, so they are supported by a
data product team.
Addressable
Valuable Trustworthy Interoperable
Discoverable Self-describing Secure
Fig. 7 Data product characteristics

Data Mesh 17
Planning and funding for data product teams are arranged
so that the teams can work on their data products
continuously (not just temporarily until a certain milestone
is reached) to ensure that data products are maintained
once development is complete. This also ensures
continuity of the data product at the agreed service level,
and allows individuals with very different backgrounds
and skills to become a high-performing team that delivers
value for the organisation. Adopting efficient and effective
work procedures is a key success factor of data product
teams.
4
Data product teams have
professionalised DataOps
practices
All data product teams work according to a set of
technical practices, known as DataOps, that ensure
efficient data operations and high-quality data products.16
DataOps applies the principles of Agile and DevOps
to data engineering and analytics. Data product teams
working based on DataOps practices do the following:
• Organise their work in agile sprints, quickly delivering
value and flexibly reacting to new or changed needs
in the business.
• Manage all their code (for data pipelines, schemas,
documentation etc.) in a shared repository to control
code merges from all team members.
• Create and reuse data pipeline templates to increase
efficiency in the development process.
• Use their own isolated development environments,
which reduce technical dependencies during
development of new data pipelines or models.
• Automatically test data and code before merging and
releasing any changes.
• Automate deployment procedures for all software
components such as environments, tools, data
pipelines, schemas and analytical models.
• Continuously monitor data pipelines and models to
detect any quality issues or anomalies.
• Collect feedback from data consumers for learning and
continuous data product improvement.
• Identify inefficiencies in team development and
operations procedures to continuously improve
performance.
By applying these practices, decentralised data
product teams within data domains can quickly provide
high-quality data products to the organisation, and
continuously improve them to maximise value for the
business.
The enterprise has established additional centralised
capabilities to ensure effective reuse of decentralised
data products across the organisation, and to ensure
compliance with regulations and internal data policies.
5
All metadata about data and
use cases is managed in a
central data catalog
To create visibility for all available data – i.e. raw datasets
as well as data products – the enterprise has compiled
a data catalog, which forms a central component of
the shared data platform. The catalog serves as a
technical backbone for the data marketplace and the
virtual data integration layer, which will be described
below. It automatically collects metadata about the
source systems, attributes, relationships, owners, quality
etc. of data, and stores these details in a structured
inventory. For data products, the catalog also collates
documentation and definitions of semantic data that
enable data consumers to better understand the data and
how to interpret it.
By visualising the metadata as a graph and applying
semantic data modeling and machine learning techniques
to the collected metadata in the data catalog, the
enterprise maintains a knowledge graph that detects and
explains relationships between data objects and uncovers
new patterns that may be relevant for data analysts when
using the data to create new business insights.
Tech-savvy data platform users, such as data engineers
or data scientists, leverage the data catalog to get a full
view of the available data and to trace data flows. This
enables them to identify the right system from which
to extract the input data for their data products, and to
identify which data product owner they need to contact
to request access to the data. These users can also
check whether the input data for their data product is
already being used by another data product. This opens
up opportunities to build on that data product, or at least
leverage existing data transformation logic.
16
Liebmann, L. (2014). 3 reasons why DataOps is essential for big data success.

Data Mesh 18
Furthermore, the data catalog allows tech-savvy users
to understand how their data product is being used
and further transformed by other data product teams,
right up to being released to data end consumers. This
can be achieved using methods such as data lineage
visualisation. Finally, analytics are applied to the data
catalog knowledge graph to discover new relationships
between data points, thus providing new metadata
insights for data engineers and data scientists on potential
ways of combining data for their use cases.
The data catalog is a centrally managed component of the
shared data platform, but its content is provided by the
decentralised data domains, their data product teams and
the data consumers. For example, a data consumer can
add a comment on a specific entity, attribute or dataset
and ask questions about how to correctly interpret this
data. A notification about this comment is sent to the
data product owner, who can then either post another
comment to directly answer the question, or improve the
metadata based on the data consumer’s feedback to
improve data usability.
Since the data catalog can only serve its purpose if all
data domains provide their metadata in the same form,
metadata collection is automated as far as possible.
Automated governance is also applied to manual
metadata entries to minimise the risk of error.
The data catalog holds all metadata and serves as a
valuable tool for data experts to find relevant data and
understand how to best query it. However, less technically
minded data consumers may want a more intuitive
interface to explore and request available data products –
such as the data marketplace described below.
6
Data products can easily be
found and accessed through
a data marketplace
The data marketplace makes data products accessible
to data consumers. This marketplace is another central
component of the shared data platform and works like an
online shop. It builds on the data catalog and provides
an easy-to-use interface which anyone in the enterprise
can use to search for data products and get immediate
access to them, or request access if restrictions are in
place due to confidentiality concerns. The marketplace
tracks consumption of data products and ranks them by
popularity, and recommends data products to consumers
based on the data products that they already use. It also
shows the service level and quality of data products so
that consumers know what to expect and which risks
and limitations to consider when using the data product.
Consumers can rate data products and provide feedback
to the product owner for further improvements. Data
products approved for external usage are also visible
and accessible to external data consumers such as
customers, suppliers or the authorities.
As well as providing data visibility and facilitating data
access, the data marketplace also handles internal cost
allocation between data domains based on data product
consumption. For example, the marketing domain might
provide many high-quality data products which are
heavily used by the sales domain, while the sales domain
provides fewer data products used by data consumers
from the marketing domain. Based on this, a portion of the
data product costs from the marketing domain is internally
allocated to the sales domain. External data consumers
can also be charged for using data products, creating a
new revenue stream for the enterprise.
To provide access to data products, the marketplace
leverages the platform’s data integration layer.

7
Access to data products is
mostly virtualized
To keep things simple for data consumers, the enterprise
has established a central data integration layer as part of
the shared data platform, which streamlines the process
of integrating decentralised data products with the central
data marketplace. As a result, data consumers don’t need
to worry about where exactly a data product is managed,
where the data is stored or how to query it on a technical
level. Consumer queries in the data marketplace are
translated to technical queries by the data integration
layer, using the metadata stored in the data catalog. The
data integration layer is capable of querying data from a
variety of data stores and interfaces. To avoid redundant
data storage in a central data repository, most queries
from the data marketplace are executed on demand on
the data product interfaces. This creates a direct flow
of data from the data product-specific data pipeline
and/or store to the data consumer’s data repository or
analytics tool. The service-level agreements (SLAs) for
the data products ensure good performance of the data
interfaces at scale. For some data products that have a
less scalable interface for ad-hoc data queries, the data is
pushed from the data product’s data pipeline into the data
integration layer, where it is stored for queries from the
data marketplace.
To enable automated querying of decentralised data
products using the data marketplace and data catalog,
common governance rules are needed.
8
Data governance rules are
centrally enforced for all data
products
The organisation has a central data governance officer,
who enacts common data policies for the organisation
in collaboration with data domain architects and data
product owners. The common data policies are kept to a
minimum to give data domains and data product teams
enough freedom to rapidly develop their products as
needed by consumers. The policies cover compliance
with regulations (e.g. for privacy), industry standards (e.g.
IT security), and some enterprise standards (e.g. data
quality metrics) that facilitate cross-domain collaboration
and use of data products.
Rather than being manually controlled within every data
domain and data product team, the data policies are
enforced by design through the shared data platform.
For instance, every data product is required to provide
its metadata to the data catalog. Additions to the data
catalog are automatically checked for sensitive data
and to ensure that appropriate access restrictions are
in place. Similarly, any registered data product API is
checked to ensure that it is compliant with corporate API
security policies and that API documentation is available.
Data products and APIs can only be deployed into the
productive environment of the shared data platform once
they have been verified as compliant, preventing non-
compliant data products from being made available on the
data marketplace.
Data Mesh 19
Data products are
consumed via a marketplace
and with common enterprise
governance rules.

Data Mesh 20
C How to build the next-generation data platform

Data Mesh 21
1
Critical reflection on the data
mesh concept
The principles underpinning the data mesh are the fruits
of hard-learned lessons in building complex transactional
systems over the last two decades. The shift towards
loosely coupled systems and autonomous agile teams has
led to significant improvements in software development
practices, such as faster time to market, superior
performance, better scalability and easier-to-maintain
products. It therefore makes sense that enterprises
would want to leverage similar best practices for data
engineering and analytics, as proposed by the data mesh
theory. As outlined above, it is clear that many enterprises
are struggling with organisational difficulties caused by
monolithic data platform architectures and centralised
data engineering and analytics teams. In addition to this,
the gap between decentralised business experts and
centralised technical specialists is growing ever larger as
enterprises move towards more advanced big data and
analytics use cases. These problems with the current
paradigm create the risk that data-driven use cases will
not deliver actual value for the business.
As a result, shifting towards more decentralised data
management with combined business and technical
expertise is a move in the right direction. Nevertheless,
this movement is still in its infancy, and businesses are
facing considerable challenges on their way towards
decentralised data management at scale.
Some of these challenges are technical in nature. Many
building blocks that are beneficial for a data mesh –
such as self-service technology offered through public
cloud systems and sophisticated distributed systems
for scalable data processing – are widely available and
reasonably well understood. However, there is still a lack
of dominant, open standards for practicable interfaces,
which severely hampers the interoperability of the nodes in
a data mesh. The emergence of RESTful interfaces using
HTTP and JSON as a standard has been a catalyst for
the microservices architecture style, but there is currently
no real equivalent for data exchange in decentralised but
highly connected data architectures, and an equivalent
is unlikely to emerge any time soon. The forces that have
driven enterprises to adopt data lakes – i.e. the volume,
variety and velocity of data for analytics use cases – make
it hard to create a small, unified set of mechanisms that
allow uncomplicated and far-reaching system integration.
In a microservice architecture, a well-designed service is
characterised by high internal cohesion and little relation
to things outside of its context, making interactions across
services rare and minimising the breadth of interfaces.
Analytics use cases, on the other hand, often need data to
be combined from different domains and data products,
which is complex on both the technical and organisational
levels. Such interactions across bounded contexts
are common in a data mesh, and data consumers are
dependent on interfaces with data products in much
broader and more complex ways. And while the challenge
of versioning a RESTful API has already been solved, this
is often not the case for the interfaces exposed by current
data management technologies.
As described above, the next-level enterprise data
platform is built on eight key capabilities that address
the three pillars of using data at scale to increase value
for the business. These eight capabilities are strongly
influenced by the data mesh principles, along with
further emerging concepts such as data fabric and other
best practices for data integration, governance and
monetisation. Taken together, these enable enterprises
to accelerate development and operationalisation of data
and analytics use cases, while also ensuring the required
level of governance.
But what does this mean for today’s enterprises aiming
to establish a data platform that will enable scaled
value generation from data over the next decade?
Is implementing a data mesh the right choice? And
where do you start?
This final chapter critically reflects on the data mesh
concept and its applicability for enterprises. Based on
this reflection, we outline a number of recommendations
to help enterprises to become truly data driven.

Data Mesh 22
17
Chee and Sawade (2021). HelloFresh Journey to the Data Mesh.
18
Schulze and Wider (2020). Data Mesh in Practice.
Additionally, data products in a data mesh will very likely
face similar challenges to those that are present in large
microservice architectures: performance will suffer when
data is pulled from multiple data products into a new
one, the complexity of the system will increase, and the
number of dependencies between data products will
increase with it. Finding and isolating errors in the pipeline
across multiple data products will be difficult.
Other challenges to implementing a data mesh are
organisational. Many software engineers do not possess
the skills and knowledge to work effectively on data-
driven use cases, while many data professionals lack
fundamental software engineering skills such as writing
automated tests, building continuous integration and
delivery into their development processes, or ensuring
proper observability in their solutions. Without the
combined skills of both worlds – which we refer to as the
DataOps engineering skillset – data product teams will not
succeed. Implementing a data mesh therefore requires
major upskilling and a strong collaborative mindset in the
organisation to break up both data silos and knowledge
silos.
In addition, decentralising data management will not
reduce the number of skilled people required. In fact,
companies may even require significantly more personnel
to get a data mesh started because a certain critical
mass is needed within each data domain to effectively
provide useful data products. For instance, as outlined
in a recent blog post, German meal kit company
HelloFresh realised that it had a significant knowledge
gap when implementing the cornerstones of a data
mesh organisation and had to drastically increase its
headcount.17
Our observations across industries confirm
that finding and hiring suitable people to drive the
development of data products is a major difficulty, as
this requires domain knowledge, product thinking and
technical expertise.
Finally, shifting from a centralised to a decentralised
organisation is a big change for today’s corporate data
management teams. The new ways of working needed for
a data mesh will initially cause friction, which will conflict
with the goals of rapidly developing and operationalising
data products and use cases. This means that there is a
risk of enterprise leaders becoming disillusioned in the
early phases of implementing a data mesh, leading them
to abandon their strategy before it starts creating value.
However, some early adopters of data mesh have already
made significant progress towards a decentralised data
management organisation. According to a presentation
of German online fashion retailer Zalando at the Data +
AI Summit, 40 of their 200 technical teams are already
actively sharing their data through the new data platform
based on data mesh principles. And over 100 teams are
already using central data platform infrastructure for their
decentral data engineering activities.18
Despite these challenges, we still believe that data
mesh principles are fundamentally the right ones.
However, immediately abandoning current data platform
approaches is not the right choice for every company,
and enterprises need to be properly prepared before they
implement a data mesh.

Data Mesh 23
2
Enterprise readiness for
data mesh
As mentioned above, most companies already have data
warehouses and data lakes in place. According to a study
conducted by BARC in April 2021, 63% of companies
follow a centralised approach to data provisioning.
However, two thirds of companies with centralised data
provisioning only use it for some of their data.19
Our recent
survey found that 51% of companies have moved to a
federated model which has some decentralised data and
analytics teams in the business domains in addition to
the company’s central data organisation, while 39% of
companies are still fully centralised. However, our survey
also revealed the implications of this shift towards more
decentralisation: among IT managers in companies with
a federated model, half are unsatisfied with synergies in
data and analytics operations, which are obviously easier
to generate with a centralised organisational model.20
As described above, decentralisation creates challenges
as well as benefits: data leaders need to understand
these challenges and address them.
Talking to our clients, we have seen that enterprises need
a certain level of maturity to successfully implement a
data mesh. There are many criteria for measuring the
maturity of data organisations, but the most important
ones – from our perspective – are the following:
• Clear data strategy with commitment from senior
management – i.e. from business leadership as well as
from IT leadership
• Sufficient number of highly skilled data architects and
data engineers in the organisation
19
Bloemen and Grosser (2021). Data Black Holes.
20
Organisational model for data management
Level
of
data
management
maturity
High
Low
Decentralised Centralised
Fig. 7 Approaches to data management for companies with different levels of maturity in data and analytics
Data Warehouse
Central data management system for
core business reporting on structured
data using relational database models
and business intelligence technologies
1
(Cloud) Data Lake
Central data management
system for analytics on
any kind of data using big
data methods and cloud
technologies
2
Data Mesh
Decentralised data management systems
distributed over data domains, but central
governance and shared infrastructure
3
Uncoordinated
0
• Adequate level of basic self-service data analytics
skills and data literacy (i.e. understanding of data as a
valuable asset that should be shared across domains)
across the entire organisation
• Adoption of Agile and DevOps principles from software
engineering to data
• Familiarity with the driving forces behind modern
architecture principles, such as cloud computing,
microservices, distributed data processing and
metadata management
• Breadth and depth of data analytics use case portfolio,
with a significant potential value for the business
The maturity of a data organisation determines whether
a more centralised or decentralised approach is suitable,
as shown above.

Fig. 8 The four phases of implementing a data mesh
Get senior manage
-
ment support and
involvement
Lay the foundations
for a data mesh
Professionalise
your capabilities
Phase
03 Democratise
and virtualise
Phase
01
Phase
02
Phase
04
If your company has a lower level of maturity, we believe
that a centralised approach to data management is more
appropriate, given that you will most likely need to keep
your few data engineering experts close together and
centrally funded in order to create momentum for using
data and analytics in the enterprise. This momentum can
be generated through a few strategic use cases that will
create significant value, even when only operationalised
on a small scale. Insufficient data engineering capacity
and low data literacy within the business domains mean
that shifting to a decentralised approach at this stage
would create a significant risk of failure. You can, of
course, start to prepare for future implementation of a
data mesh – for example, by launching broad data literacy
initiatives and establishing an enterprise data catalog,
which is a valuable development for your current data
platform setup as well as for a future data mesh. As you
grow your data and analytics capabilities and successfully
operationalise more use cases over time, organisational
difficulties will become even more apparent – but you will
also have increased your readiness for decentralisation,
thus reducing your risk of failure when you implement a
data mesh.
If your company has achieved a higher level of maturity,
we firmly believe that a more decentralised approach
to data management will help you with developing and
operationalising more use cases and scaling them
better across the enterprise. If you are in this position,
you should therefore consider adopting the data mesh
principles and implementing the eight capabilities that
we observed in our far-off galaxy. However, as described
above, this transformation won’t be a trivial task.
Embarking on this journey needs strong leadership to
drive the organisational change required, but you also
need a pragmatic approach that allows you to try out new
practices, fail fast and learn quickly. The next section
introduces our recommended path to implementing
the data mesh for enterprises with higher levels of data
management maturity.
3
How do you start implementing
a data mesh?
Based on our experience from introducing new data
management practices in enterprises and supporting
them on their journeys towards a data mesh, we
recommend starting with the fundamental capabilities
you need and allowing enough time to get used to
the new decentralised approach, before going for full
implementation of the eight capabilities described
above. As the data mesh concept is still evolving, your
initial implementation roadmap should not be set in
stone. Instead, it should serve as a strategic guide
which you should review, refine and flesh out after each
implementation phase. We recommend a four-phase
approach to implementation, as illustrated below.
Data Mesh 24

Data Mesh 25
Data domains (5%)
Data products (5%)
Interdisciplinary teams (5%)
DataOps (5%)
Data catalog (5%)
Data marketplace (5%)
Data virtualisation (5%)
Automated governance (5%)
Implementing a data mesh is a transformational change
for your enterprise’s data architecture and technology, but
even more so for your organisation and processes – not
just in IT, but across the entire business. This means that
implementation cannot be driven solely by IT: it needs to
be co-led by business and IT leaders.
We recommend carefully planning this transformation
and ensuring strong sponsorship and support from senior
management. This can be achieved by formulating a clear
data mesh strategy that outlines
a. the aspiration, i.e. what you want to achieve and why
a data mesh is the right approach that will create
more value,
b. the use cases which you can enable and improve to
achieve the aspiration,
c. your target vision for the data mesh and its impact on
people, processes and technology; and
d. the investment plan and high-level roadmap to achieve
the target vision, with a detailed plan of your next steps
as described for phase 2.
The maturity of the eight capabilities is not very high in
this first phase: the journey has only just started, and your
focus needs to be on getting leadership commitment.
Once proper change management mechanisms are in
place and your leaders are all on the same page, have
committed to the strategy and have communicated it
throughout the organisation, it is time to launch phase 2.
Phase 01
Get senior management support and involvement
Fig. 9 Maturity of capabilities in phase 01

Data Mesh 26
Data domains (75%)
Data products (50%)
DataOps (25%)
Data catalog (75%)
Since the greatest changes brought about by a data mesh
transformation will be in the organisation, we recommend
starting by delineating your data domains and anchoring
them in the business. This will create the foundations
for decentralising data management. It may take some
time for the business to truly take ownership of the data
domains, and coaching may be necessary. We suggest
taking an agile approach with the data domains – i.e.
taking the key data domains and then starting with a
smaller bounded context within these domains to develop
your first data products and learn how the concept works
from a data architecture point of view. This will also help
ensure adequate staffing of your interdisciplinary data
product team despite the capacity limitations that you
may have to deal with among experienced data engineers
and architects. Once the first data domains are firmly in
place, further data domains can be established using the
lessons learned from the first ones.
Phase 02
Lay the foundations for a data mesh
While your decentralised data product teams are
developing their first data products, your central platform
team should establish the data catalog to provide
transparency on the data products and their usage, and
to centrally collect metadata from the very beginning.
The central platform team should also start to introduce
automated governance rules to serve as guardrails for the
data product teams – otherwise, you may find that your
data mesh becomes a data mess.
With these fundamental capabilities in place, the data
mesh can be launched and will grow over time as more
and more domains join with their data products. Once
the data mesh has gained traction in the business and is
being filled with some mature data products, it’s time to
move on to phase 3.

Data Mesh 27
To strengthen your data mesh, data product teams should
gradually adopt and professionalise DataOps practices.
This makes these teams more efficient in managing their
data products and allows them to gradually extend and
enrich these products. Depending on the level of usage
of data products, capacity may also be shifted to data
products which have the highest levels of adoption and
create the most value. For these products, long-term
budget planning for your data product teams is very
important: this will allow your teams to be set up for the
long term and will ensure reliable operation of their data
products.
Building on the data catalog, metadata management
should also be combined with analytics to automatically
and continuously detect new patterns and relationships
between available data points (e.g. using a knowledge
graph). This will suggest more or better ways to model
and integrate data pipelines and will provide data
consumers with a more integrated data landscape for
better business decision-making. To do this, the data
mesh should be complemented by a data fabric, which
takes your data platform to the next level in terms of
automation and intelligence.
Phase 03
Professionalise your capabilities
Data domains (100%)
Data products (100%)
DataOps (75%)
Data catalog (100%)

Data Mesh 28
Data domains (100%)
Data products (100%)
DataOps (100%)
Data catalog (100%)
To drive higher reuse of data products, the data
marketplace should be introduced as the new landing
page for data consumers to intuitively explore and
access available data products. You can launch your
data marketplace with a few data products that already
have a broad reach, and then continuously expand it with
more and more data products as your data mesh grows.
Ultimately, the data marketplace model will change the
budgeting process of your data product teams, as it
allocates budgets to the most frequently used and most
business-critical data products on the marketplace.
Although its first and most important role is to promote
data products within the enterprise, it can also be
subsequently used to share and monetise data externally,
thus creating new business opportunities through
industry-wide data ecosystems.
Access to data products should also be virtualized as much
as possible where this makes sense. This will allow you to
reduce central data repositories and fully embrace the data
mesh vision. You may find that you need to complement
traditional data storage and processing technologies with a
virtual data querying and processing engine that connects
directly to the decentralised data repositories of the data
products through their defined interfaces.
To conclude our exploration of the data mesh paradigm, we
decided to take a look at IKEA – a company that is currently
implementing a data mesh – to gain some insights and
lessons learned from the company’s data mesh journey.
Phase 04
Democratise and virtualise

Data Mesh 29
D Why and how is IKEA implementing its data mesh?

Inter IKEA (hereafter “IKEA”), an international
conglomerate that designs, manufactures and sells home
furniture, kitchen appliances and other home accessories,
is currently on an analytics journey led by senior
management. The goal of this journey is to transform
the organisation and get it ready for the next era, in line
with the IKEA vision: “To create a better everyday life for
the many people”. At the centre of this transformation
lies the IKEA data and analytics platform, which has
been designed in collaboration with PwC based on the
federated approach of a data mesh. We spoke about the
data mesh with Jathisan Chandrasegaram, leader of the
data and analytics platform at IKEA.
IKEA is a complex organisation, with an end-to-end value
chain spanning multiple companies and covering all
areas from sourcing materials to distributing and selling
products. This means that some parts of the organisation
are more data-focused and mature in using analytics
than others, resulting in a mix of technologies: these
range from Excel and Access databases to distributed
cloud-based data processing, machine learning and
natural language processing. There are already many
analytics use cases and a lot of positive momentum, but
the organisation as a whole is still in the early stages of its
journey towards democratised data and analytics. IKEA
hopes to use its data mesh to accelerate this journey.
Data mesh is therefore not a target in itself, but a stage
on IKEA’s strategic roadmap to achieve the company’s
desired position in the field of data and analytics.
When the data mesh concept first appeared, Jathisan felt
that the idea was a natural fit for IKEA, given the company’s
highly federated structure and entrepreneurial spirit, which
are deeply embedded within the organisational culture
across all business functions. Although the architecture
pattern of a data mesh does generally fit IKEA’s goals and
needs, Jathisan and his team made some adjustments
to the concept to better reflect the company’s
requirements – for example, by adapting the data domain
model to handle IKEA’s core business organisations and
to suit the specifics of a franchise business.
Data Mesh 30
When implementing the data mesh architecture, the
biggest challenge for IKEA is to bring business users
closer to the technology and enable them to use it
effectively. The company’s data and analytics community
addresses this challenge on four different levels:
Digital leadership: bringing people with different
business and technical backgrounds into the
same room to share viewpoints and work out
how to improve IKEA.
1
Changing the development model: moving from
the traditional split between development and
operations funding to DevOps funding for data
products.
2
Clear demand management: following a defined
process that prioritises and promotes the right
use cases to maximise value using existing
resources.
3
Increasing data literacy: investing in upskilling
people and helping them to prepare for the new
data-driven era.
4

Data Mesh 31
The results have been overwhelmingly positive. Many
teams have been onboarded since the release of the
platform’s first capabilities in the autumn of 2021, as it
enables teams to quickly get up to speed and seamlessly
apply data processing and analytics to data products.
The platform enables the business domains to accelerate
their journey to becoming more data-driven and reduces
bottlenecks in the data engineering and analytics process –
all enabling Jathisan and his team to live up to their
ambition to “make it easy for the many platform users by
tackling the complexity within the core platform team”.
It is important to emphasise that the data mesh alone
does not solve all data management challenges. As
a result, IKEA has introduced additional initiatives
alongside its platform, such as data catalog, master data
management, data lineage and data quality initiatives.
Some of these initiatives are driven at the domain level,
while others are driven at the enterprise level, depending
on the type of challenge they address. In general,
technological challenges can mostly be solved at the
enterprise level, while challenges around people and
processes are better solved at the domain level.
Based on his own experience, Jathisan suggests that
organisations thinking about implementing a data mesh
should consider the following questions:
• Is your organisation large enough to need a data mesh?
• Is your organisation federated?
• Do you have a culture of entrepreneurship and people
committed to the data vision?
• Is your organisation invested in making people more
data-driven?
• Do your senior managers know what journey you want
to take as a company?
If the answer to most of these questions is yes, then a
data mesh could be a good fit for your organisation.
If you decide to implement a data mesh, Jathisan has
three recommendations:
Build a strong team first, before starting the
technical implementation.
1
Find the right partners that can help you to
manage the unknown.
2
Don’t rush the development process: make
sure you consider the community’s designs and
interpretations of the data mesh, which could
save you a lot of time.
3
We would like to extend our thanks to Jathisan for sharing
his thoughts and experiences to help other companies on
their journeys to becoming data-driven enterprises.

References
Bloemen, J. and Grosser, T. (2021)
Data Black Holes. Available at https://pages.barc.de/hubfs/Marketing/Reports/Data%20Black%20Holes_Report.
pdf?utm_medium=email_hsmi=166767220_hsenc=p2ANqtz--f-MeOWIQB1oyEpQwgmPe9hJMpWfVD3fR-
GbTrj5_Flt5EOlkcA_RjyHOyVJyC_GRAnBUGyB52RZ70Sy49M9UtDDaLvP88hOsWOj7uiMk2ZSjVC0Yutm_
content=166767220utm_source=hs_automation.
Clemece, W. and Sawade, C. (2021)
HelloFresh Journey to the Data Mesh. Available at https://engineering.hellofresh.com/hellofresh-journey-to-the-data-
mesh-7fe590f26bda.
Dehghani, Z. (2019)
How to Move Beyond a Monolithic Data Lake to a Distributed Data Mesh. Available at https://martinfowler.com/articles/
data-monolith-to-mesh.html#TheNextEnterpriseDataPlatformArchitecture.
Dehghani, Z. (2020)
Data Mesh Principles and Logical Architecture. Available at https://martinfowler.com/articles/data-mesh-principles.html.
Evans, E. (2003)
Domain-Driven Design: Tackling Complexity in the Heart of Software (1st ed.). Boston, MA: Addison-Wesley.
Ford, N., Richards, M., Sadalage, P., Dehghani, Z. (2021)
Software Architecture: The Hard Parts. Sebastopol, CA: O’Reilly, Inc.
Gartner (2021)
Data Fabric Architecture is Key to Modernizing Data Management and Integration. Available at https://www.gartner.
com/smarterwithgartner/data-fabric-architecture-is-key-to-modernizing-data-management-and-integration.
Inmon, W.H. (1992)
Building the Data Warehouse. New York, NY: John Wiley Sons, Inc.
Laney, D. (2001)
3d data management: Controlling data volume, velocity and variety. Gartner (Ed.).
Liebmann, L. (2014, June 19)
3 reasons why DataOps is essential for big data success [Blog post]. Available at www.ibmbigdatahub.com/blog/3-
reasons-why-dataops-essentialbig-data-success.
Schulze, M. and Wider, A. (2020)
Data Mesh in Practice at Data+AI Summit 2020. Available at https://databricks.com/session_eu20/data-mesh-in-
practice-how-europes-leading-online-platform-for-fashion-goes-beyond-the-data-lake.
Vernon, V. (2013)
Implementing Domain-Driven Design (1st ed.). Boston, MA: Addison Wesley.
Data Mesh 32
PwC Data Mesh
Research 2022
PwC Germany carried out online market research
in collaboration with Statista from December 2021
to January 2022. This research involved surveying
152 IT decision makers (i.e. C-level executives, division
managers and team leads) at enterprises in Germany
with at least 500 employees, across multiple industries
(i.e. manufacturing, retail, financial services, IT services
and public services). The survey featured multiple-
choice questions focusing on data and analytics
maturity, current challenges, adoption of the data
mesh concept and expected benefits.

Data Mesh 33
Authors
About us
Our clients face diverse challenges, strive to put new
ideas into practice and seek expert advice. They
turn to us for comprehensive support and practical
solutions that deliver maximum value. Whether for a
global player, a family business or a public institution,
we leverage all of our assets: experience, industry
knowledge, high standards of quality, commitment to
innovation and the resources of our expert network
in 158 countries. Building a trusting and cooperative
relationship with our clients is particularly important to
us – the better we know and understand our clients’
needs, the more effectively we can support them.
PwC. More than 11,000 dedicated people at 21
locations. €2.2 billion in turnover. The leading auditing
and consulting firm in Germany.
Andreas Odenkirchen
Andreas is a Senior Manager in PwC’s
technology consulting practice in
Germany, where he leads the data
and AI strategy service offering. He
supports consumer-focused and
industrial enterprises throughout
their transformations into data-driven
enterprises, from strategy to execution. Prior to joining PwC,
Andreas worked as a data scientist and strategy consultant
at IBM’s global center of competence for advanced analytics.
Oliver Krukow
Oliver is a Senior Associate in PwC’s
Germany. His work covers all aspects
of designing and building data-intensive
systems, ranging from business
intelligence applications to industrial
IoT solutions. He loves bringing modern
software development practices and agile methodologies
to the world of analytical systems just as much as focusing
on the nitty-gritty details of the technologies.
© May 2022 PricewaterhouseCoopers GmbH Wirtschaftsprüfungs-
gesellschaft. All rights reserved.
In this document, “PwC” refers to PricewaterhouseCoopers
GmbH Wirtschaftsprüfungsgesellschaft, which is a member firm
of PricewaterhouseCoopers International Limited (PwCIL). Each
member firm of PwCIL is a separate and independent legal entity.
Agnes Heringer
Agnes is a Manager in PwC’s
Germany, where she focuses on
data and analytics strategy and
designing data and analytics operating
models. Agnes has more than seven
years of professional experience in
digital strategy, data-driven marketing and IT-related
environments. As an experienced data and analytics
consultant, she helps clients to unlock business value from
data, and to derive actionable insights through analytics.
Martin Whyte
Martin is a Partner within PwC’s
Germany, where he leads the digital
operations analytics team. He
focuses on envisioning, designing and
implementing data-driven solutions
in manufacturing, supply chain
and procurement. Prior to this, Martin held a variety of
positions at other technology consulting firms, but he has
always had a focus on data analytics solutions.
Stephan Bautz
Stephan is a Senior Manager in
PwC’s financial services technology
consulting department, where he
leads the data architecture offering.
With experience from several
projects in various industries and a
team of experts to support him, he
uses today’s best data platform architectures to enable
analytical use cases. He is also a board member in
Bitkom’s “Big Data Advanced Analytics” workstream.
Contributors
Sebastian Müller
Antonios Marinos
Jathisan Chandrasegaram, Inter IKEA Systems B.V.
Maximilian Franke
Tamás Barát
Acknowledgements

pwc-data-mesh.pdf

More Related Content

What's hot (20)

Similar to pwc-data-mesh.pdf (20)

Recently uploaded (20)

pwc-data-mesh.pdf