Professional Documents
Culture Documents
com FA L L 2 0 1 7
TENhot
DATA
ANALYTICS
FROM IDG
TRENDS
HOW TO MINE
SOCIAL MEDIA FOR
BUSINESS GOLD 3
CONNECTING WITH
+FIVE
GOINGcold
CUSTOMERS ON
THEIR TERMS 6
ANALYTICS SUCCESS
Big data, machine learning,
STORIES: AN INSIDE data science — the data analytics
LOOK 18
revolution is evolving rapidly.
BY MARTIN HELLER
contents FA L L | 2017
COV E R A RT T H I N K S TO C K FA L L 2 017 | C I O . C O M 2
emerging TECH
How to mine
social media for
business gold
A strong social intelligence strategy can transform
social media data into actionable insights
— and help shape your company’s products,
advertising and customer experience.
BY SANDR A GIT TLEN
T H I N K S TO C K FA L L 2 017 | C I O . C O M 3
emerging TECH
FA L L 2 017 | C I O . C O M 4
emerging TECH
Smart
SECURITY
TM
ing or kept as a separate entity. “In exchanged for a quarterly newsletter ready for distribution to your
the early days, companies set up better decision employees—saving you precious time on employee
social media as a feed for listening, making.” education! The compelling content combines
personal and organization safety tips, making it
figuring they would find some- —JENNY SUSSIN applicable to many facets of employees’ lives.
thing that would lead to a million-
Security Smart has an easy-to-read design and clear,
dollar idea and make it all worth
engaging and entertaining articles so you are assured
it,” she says, adding, “That’s how with the governance applied to other
that your intended audience of employees—your
they got in trouble.” data management efforts such as
organization’s most valuable assets—will read and
For social intelligence to truly take CRM and ERP systems, according retain the information. Sign up today to start having
hold, it must be incorporated with to Sussin. “We need the social people this newsletter distributed as a key tool in raising
the organization’s business analyt- and the data management people to security awareness within your organization.
ics and intelligence infrastructure in talk to each other and understand
■ Most security breaches happen due to human error
Subscribe today!
IT. “Data leaders will say they have that this information can and needs To view a sample issue of
■ 4 out of every 5 data breaches caused by human
enough on their plate with bigger to be exchanged for better decision the newsletter, learn about
error are unintentional
data management problems and making,” she says. ♦ the delivery options and
■ 4 out of every 5 security events caused by insiders to subscribe visit:
don’t see social intelligence as their have a negative impact on their enterprise organi-
WWW.SECURITYSMART.COM
issue, but it is,” she says. Sandra Gittlen is a freelance writer and zation (including loss of confidential information,
critical system disruptions, reputational harm,
Social intelligence would do well editor in the greater Boston area.
lost customers and more)
Security Smart is published by CSO, A business unit of IDG Enterprise. | ©2015 CSO
FA L L 2 017 | C I O . C O M 5
transformation NATION
Connecting with
customers on
their terms
Food-service company HMSHost harnesses
cross-channel data to optimize dining
experiences for hungry travelers.
BRENDAN MCGOWAN
T
here are, perhaps, no
more vocal or assertive
customers than airline pas-
sengers between connect-
ing flights. Sometimes these
hungry and harried travelers just need a
hot coffee and croissant.
HMSHost, a leading firm in the food and
beverage space for travelers, is focused
T H I N K S TO C K FA L L 2 017 | C I O . C O M 6
transformation NATION
FA L L 2 017 | C I O . C O M 7
transformation NATION
FA L L 2 017 | C I O . C O M 8
cover story | ANALYTICS
hot
D
ata analytics
are fast becom-
ing the lifeblood
Big data,
machine data of IT. Big data,
machine learn-
analytics
ing, data science — the range of
learning, data technologies and techniques for
science — the analyzing vast volumes of data
data analytics
trends
is expanding at a rapid pace. To
revolution is gain deep insights into customer
behavior, systems performance,
evolving
going
and new revenue opportunities,
rapidly. your data analytics strategy will
BY MARTIN HELLER benefit greatly from being on
cold
top of the latest trends.
Here is a look at the data ana-
lytics technologies, techniques
and strategies that are heating
up and the once-hot trends that
are beginning to cool. From busi-
ness analysts to data scientists,
everyone who works with data is
being impacted by the analytics
revolution. If your organization
is looking to leverage data ana-
lytics for actionable intelligence,
the following heat index should
be your guide. >>
T H I N K S TO C K FA L L 2 017 | C I O . C O M 9
cover story | ANALYTICS
Self-service BI
Who: BI/BA pros, managers
behavior, systems performance, and new
W
ith self-service BI tools,
such as Tableau, Qlik
Sense, Power BI and Domo,
revenue opportunities, your data analytics
managers can obtain cur-
rent business information
strategy will benefit greatly from being
in graphical form on demand. While a certain
amount of setup by IT may be needed at the out-
on top of the latest trends.
set and when adding a data source, most of the
work in cleaning data and creating analyses can
be done by business analysts, and the analyses tions, and even the building of additional stores erance within minutes of the event; that’s easily
can update automatically from the latest data in under-served areas. done with an app that queries the QA database
any time they are opened. every minute, updates and displays a Shewhart
Managers can then interact with the analy- H E AT I N G U P: control chart, and optionally sounds an alarm
ses graphically to identify issues that need to Mobile dashboards when a line goes out of spec.
Who: BI/BA pros, managers, developers
I
be addressed. In a BI-generated dashboard or
“story” about sales numbers, that might mean n a world where managers are rarely at CO O L I N G D OW N:
drilling down to find underperforming stores, their desks, management tools need to Hadoop
Who: Data scientists
H
salespeople and products, present mobile-friendly dashboards to
or discovering trends in be useful and timely. Most self-service adoop once seemed like the
year-over-year same-store BI tools already have this feature, but not answer to the question “How
comparisons. These dis- every key business metric goes through a BI tool. should I store and process really
coveries might in turn For example, a manufacturing plant is likely to big data?” Now it seems more like
guide decisions about have a dedicated QA system monitoring all pro- the answer to the question “How
future stocking levels, duction lines. All plant managers need to know many moving parts can you cram into a system
product sales and promo- whether any of the lines have drifted out of tol- before it becomes impossible to maintain?”
T H I N K S TO C K FA L L 2 017 | C I O . C O M 10
cover story | ANALYTICS
From business analysts to data scientists, R is available as free open source, and is
embedded into dozens of commercial products,
everyone who works with data is being including Microsoft Azure Machine Learning
Studio and SQL Server 2016.
impacted by the data analytics revolution.
H E AT I N G U P:
S
modules: Hadoop Common (utilities), Hadoop
Who: Data scientists with strong statistics
D
Distributed File System (HDFS), Hadoop YARN ome of the most powerful deep
(scheduler) and Hadoop MapReduce (parallel ata scientists have a number of learning algorithms are deep neu-
processing). On top of or instead of these, people options to analyze data using ral networks (DNNs), which are
often use one or more of the related projects: statistical methods. One of the neural networks constructed from
Ambari (cluster management), Avro (data seri- most convenient and powerful many layers (hence the term “deep”)
alization), Cassandra (multi-master database), methods is to use the free R pro- of alternating linear and nonlinear processing
Chukwa (data collection), HBase (distributed gramming language. R is one of the best ways units, and are trained using large-scale algo-
database), Hive (data warehouse), Mahout (ML to create reproducible, high-quality analysis, rithms and massive amounts of training data. A
and data mining), Pig (execution framework), since unlike a spreadsheet, R scripts can be deep neural network might have 10 to 20 hidden
Spark (compute engine), Tez (data-flow program- audited and re-run easily. The R language and layers, whereas a typical neural network may
ming framework intended to replace MapRe- its package repositories provide a wide range of have only a few.
duce), and ZooKeeper (coordination service). statistical techniques, data manipulation and The more layers in the network, the more
If that isn’t complicated enough, factor in plotting, to the point that if a technique exists, characteristics it can recognize. Unfortunately,
Apache Storm (stream processing) and Kafka it is probably implemented in an R package. R the more layers in the network, the longer it
(message transfer). Now consider the value is almost as strong in its support for machine will take to calculate, and the harder it will be
added by vendors: Amazon (Elastic Map learning, although it may not be the first choice to train. Packages for creating deep neural net-
Reduce), Cloudera, Hortonworks, Microsoft for deep neural networks, which require works include Caffe, Microsoft Cognitive Tool-
(HDInsight), MapR and SAP Altiscale. Con- higher-performance computing than R cur- kit, MXNet, Neon, TensorFlow, Theano, and
fused yet? rently delivers. Torch.
FA L L 2 017 | C I O . C O M 11
cover story | ANALYTICS
H E AT I N G U P:
T
fundamental design errors. In some cases, the
Who: BI/BA pros, data scientists
T
smart devices only work if they are connected ensorFlow is Google’s open
he internet of things (IoT) may be to the internet and can reach the manufac- source machine learning and
the most-hyped set of technolo- turers’ servers. That becomes a significant neural network library, and it
gies, ever. It may also be the worst point of failure when the manufacturer ends underpins most, if not all, of
thing that happened to internet product support, as happened with the Sony Google’s applied machine learn-
security, ever. Dash and the early Nest thermometer. Includ- ing services. The Translate, Maps, and Google
IoT has been touted for smart homes, wear- ing a remote internet-connected server into a apps all use TensorFlow-based neural net-
ables, smart cities, smart grids, industrial inter- control loop also introduces a significant and works running on our smartphones. Tensor-
net, connected vehicles, connected health, smart variable lag into the control loop which can Flow models are behind the applied machine
retail, agriculture and a host of other scenarios. introduce instability. learning APIs for Google Cloud Natural Lan-
Many of these applications would make sense Even worse, in their rush to connect their guage, Speech, Translate and Vision.
if the implementation were secure, but by and “things” to the internet, manufacturers have Data scientists can use TensorFlow, once they
large that hasn’t happened. exposed vulnerabilities that have been exploited can get over the considerable barriers to learn-
T H I N K S TO C K FA L L 2 017 | C I O . C O M 12
cover story | ANALYTICS
R
research and production, auto-differentiation
of variables, and the ability to maximize perfor- unning batch jobs overnight to
mance by prioritizing GPUs over CPUs. analyze data is what we did in
the 1970s, when the data lived
lyze data is what
H E AT I N G U P:
on 9-track tapes and “the main-
frame” switched to batch mode
we did in the 1970s,
MXNet
Who: Data scientists
for third shift. In 2017, there is no good reason to when the data lived
M
XNet (pronounced “mix-net”)
settle for day-old data.
In some cases, one or more legacy systems on 9-track tapes and
is a deep learning framework
similar to TensorFlow. It lacks
(which may date back to the 1960s in some
cases) can only run analyses or back up their “the mainframe”
the visual debugging available
for TensorFlow but offers an
data at night when not otherwise in use. In other
cases, there is no technical reason to run batch
switched to batch
imperative language for tensor calculations that
TensorFlow lacks. The MXNet platform auto-
analysis, but “that’s how we’ve always done it.”
You’re better than that, and your management
mode for third shift.
matically parallelizes symbolic and imperative deserves up-to-the-minute data analysis. In 2017, there is no
operations on the fly, and a graph optimization
layer on top of its scheduler makes symbolic H E AT I N G U P: good reason to settle
Microsoft Cognitive
execution fast and memory efficient.
MXNet currently supports building and Toolkit 2.0 for day-old data.
Who: Data scientists
T
training models in Python, R, Scala, Julia, and
C++; trained MXNet models can also be used he Microsoft Cognitive Toolkit, directed graph. It has many similarities to Ten-
for prediction in Matlab and JavaScript. No also known as CNTK 2.0, is a uni- sorFlow and MXNet, although Microsoft claims
matter what language you use for building fied deep-learning toolkit that that CNTK is faster than TensorFlow especially
your model, MXNet calls an optimized C++ describes neural networks as a for recurrent networks, has inference support
back-end engine. series of computational steps via a that is easier to integrate in applications, and
FA L L 2 017 | C I O . C O M 13
cover story | ANALYTICS
has efficient built-in data readers that have APIs for other languages. Scikit- April 2017 after more than a year of struggling
also support distributed learning. learn doesn’t support PyPy, the fast through buggy release candidates. And yet,
There are currently about 60 sam- just-in-time compiling Python imple- as of July 2017, it has over 500 open issues.
ples in the Model Gallery, including mentation, nor does it support GPU An outsider might get the impression that the
most of the contest-winning models acceleration, which aside from neural project stalled while the deep learning com-
of the last decade. The Cognitive Tool- networks, Scikit-learn has little need for. munity moved on to TensorFlow, CNTK and
kit is the underlying technology for Scikit-learn earns high marks for ease MXNet.
Microsoft Cortana, Skype live transla- of development. The algorithms work as
tion, Bing and some Xbox features. advertised and documented, the APIs are H E AT I N G U P:
T
H E AT I N G U P: “impedance mismatches” between data struc-
Scikit-learn tures. he Jupyter Notebook, originally
Who: Data scientists
S
called IPython Notebook, is an
cikits are Python-based scientific CO O L I N G D OW N: open-source web application that
toolboxes built around SciPy, the Caffe allows data scientists to create and
Who: Data scientists
T
Python library for scientific comput- share documents that contain live
ing. Scikit-learn is an open source he once-promising Caffe deep code, equations, visualizations and explanatory
project focused on machine learn- learning project, originally a text. Uses include data cleaning and transforma-
ing that is careful about avoiding scope creep strong framework for image clas- tion, numerical simulation, statistical modeling,
and jumping on unproven algorithms. On the sification, seems to be stalling. machine learning and much more.
other hand, it has quite a nice selection of solid While the framework has strong Jupyter Notebooks have become the preferred
algorithms, and it uses Cython (the Python to C convolutional networks for image recognition, development environment of many data scien-
compiler) for functions that need to be fast, such good support for CUDA GPUs, and decent por- tists and ML researchers. They are standard
as inner loops. tability, its models often need excessively large components on Azure, Databricks and other
Among the areas Scikit-learn does not cover amounts of GPU memory, the software has online services that include machine learning
are deep learning, reinforcement learning, year-old bugs that haven’t been fixed, and its and big data, and you can also run them locally.
graphical models, and sequence prediction. It is documentation is problematic at best. “Jupyter” is a loose acronym meaning Julia,
defined as being in and for Python, so it doesn’t Caffe finally reached its 1.0 release mark in Python, and R, three of the popular languages
T H I N K S TO C K FA L L 2 017 | C I O . C O M 14
cover story | ANALYTICS
O
CapEx to the OpEx world. Sometimes the action would be to define a
ne of the mantras of efficient anal- new report to answer a question brought up by
ysis is “do the computing where CO O L I N G D OW N: an existing report. The whole cycle would start
the data resides.” If you don’t or Monthly BI reports over, and a month or two later the new report
Who: BI/BA pros, data scientists
B
can’t follow this rule, your analy- would be added to the monthly printout.
sis is likely to have large delays efore self-service business intel- Alas, businesses that want to be agile can’t
if the data moves across the local network, and ligence became popular, BI was the respond to environmental and market changes
even larger delays if it moves over the internet. province of IT. Managers described in months: the time between asking a question
That’s why, for example, Microsoft recently what they thought they wanted and getting an answer should be measured in
added R support to SQL Server. to see, business analysts turned seconds or minutes, not weeks or months. ♦
As the amount of data generated by your com- that into specifications, and BI specialists created
pany grows exponentially, the capacity of your reports to meet the specifications — eventually, Martin Heller is a regular contributor to CIO.com.
FA L L 2 017 | C I O . C O M 15
leading TECH
Incoming CIOs are often ener- CIO transitions can be broken key themes presented here are in vide continuity and minimize dis-
gized by the prospect of putting down into four typical scenarios: the report “Taking charge: The ruption. Such CIOs usually have
their stamp on an IT organization. internal hire, hybrid insider, external essential guide to CIO transitions.” a successful track record and have
1
Each brings his or her unique hire, and mergers and acquisitions already built credibility through-
vision for overhauling technical (M&A) or divestitures. Each presents Internal hire out the company, having forged
operations to provide a broader unique opportunities, challenges and The most common transi- strong relationships with a num-
impact on businesses. But leader- business contexts that dictate how tion scenario is the CIO who ber of key stakeholders and peers.
ship transitions for new CIOs can the transition should be approached. has been promoted from within Internal promotes tend to not
be fraught with challenges. Data and insights around the IT. Hiring from within can pro- want to be viewed as disruptive,
T H I N K S TO C K FA L L 2 017 | C I O . C O M 16
leading TECH
and many are slow to make tal- Hybrid insiders can bring often impress with a success- financial and talent targets, and
ent decisions. Some are first-time instant credibility and existing ful track record in a particular the complex task of either inte-
CIOs; they may find it difficult to relationships. They can be quick domain. For example, CIOs from grating or separating multiple
delegate work and manage reports to identify opportunities for IT to financial services or retail can be business-critical applications and
who were formerly colleagues. contribute to business outcomes. appealing to industries seeking platforms. Often, the CIO may be
Many say they underestimated the They are unlikely to be limited by expertise in building digital plat- working with unfamiliar IT staff.
importance of relationships in get- existing IT culture and can rapidly forms. When massive change needs
ting their work done. make significant cultural and tal- External hires often find it easier to occur in a short timeframe,
Unless they fill their former roles ent shifts. to make key talent changes and constant communication with
quickly, internal hires may end up Coming into this role, the hybrid are able to do so at a faster pace stakeholders and IT staff is likely
straddling both roles for months, insider may need to learn about than many internally promoted essential. CIOs must often main-
impacting performance and focus. relevant technologies, and must CIOs. Because they often inherit IT tain morale as their teams struggle
2
work hard to gain the respect of IT teams with significant challenges, with long workdays and frus-
Hybrid insider staff who may be skeptical of the any achievement is celebrated. trating migrations. Transition-
Sometimes, companies hire hybrid insider’s authority. A com- But outside hires often are given ing CIOs in M&A or divestitures
CIOs with little or no IT mon mistake can be to overcommit a short timeline to end or redirect typically do not have to develop a
experience from within their enter- to business partners, focusing on bad initiatives. They also typically strategic plan; deadlines and road
prise leadership ranks. This hybrid short-term wins at the expense of must simultaneously gain under- maps are usually driven by the
insider is often hired when there is long-term total cost of ownership standing of a new industry, culture business, typically with little input
a need for greater business align- or broader strategy. Many hybrid and company, and they may take from IT. ♦
ment or stakeholder trust. Since insiders report being surprised by longer to establish key stakeholder
they typically don’t have a strong the complexity of the IT environment. relationships than internal hires or Khalid Kark is a managing director with
3
technology background, hybrid hybrid insiders. US CIO Program at Deloitte LLP.
External hire
4
insiders often focus on untangling This publication contains general information
alignment and relationship issues CIOs are typically brought M&A or divestitures only and Deloitte is not, by means of this publica-
while relying on technologists and in from outside the com- In general, M&A and tion, rendering accounting, business, financial,
operational lieutenants to keep the pany when significant change is divestitures can present investment, legal, tax, or other professional
lights on. needed within IT. External hires CIOs with tight deadlines, strict advice or services.
FA L L 2 017 | C I O . C O M 17
appyling TECH
Analytics
success stories:
An inside look
CIOs are embracing data analytics to
fuel efficiency and growth, but not
every effort produces results. Here
are some secrets of success.
BY CLINT BOULTON
CIOs are playing with predictive for big data and business analyt- won advice for peers undertaking and hit the road to woo retail cli-
analytics tools, crafting machine ics expected to reach $150.8 billion similar efforts. ents. Today, sales staff are armed
learning algorithms and battle- this year, according to IDC. with iPads that tell them what
testing other solutions in pursuit But there’s a dark side to this Providing contex- stores to visit, what offers to make,
of business efficiencies and new delirious spending: Most data ana- tual relevance and other crucial metrics. “They
ways to serve customers. And lytics projects fail to yield measur- For years, Dr. Pepper Snapple were glorified order takers,” said
they are spending more than ever able value. Group’s sales route staff grabbed Tom Farrah, CIO of Dr. Pepper
on technologies that support data Here, IT leaders share data ana- a binder with customer data and Snapple Group. “Now they are
science, with worldwide revenues lytics success stories and hard- notes on sales and promotions, becoming intelligent sales people
T H I N K S TO C K FA L L 2 017 | C I O . C O M 18
applying TECH
FA L L 2 017 | C I O . C O M 19