How NCR Corporation Improved Quality and Time to Market for Advanced Analytics
By Audrey Ng, Senior Director, Strategic Alliances, Hortonworks
and Cesar Rojas, Product Marketing Director, Teradata
As NCR says, its customers “turn the gears
of the global economy.” They include thousands
of banks and retailers that depend
upon tens of thousands of NCR self-service
kiosks, point-of-sale (POS) terminals, and
ATMs and barcode scanners, among other
equipment.
“Our customers—such as banks and retailers—can’t have downtime, so it’s our job
to actively support that requirement,” said
Brian Valeyko, NCR’s director of enterprise
data and business intelligence.
That downtime is increasingly rare, thanks
largely to how NCR uses big data. The company
has the capacity to gather and store
real-time performance data not just from
a sampling of devices, but from every one
of millions of deployed devices. NCR can
store data securely in a Hortonworks Data
Platform–powered “data lake.” From that
data, NCR can create rich predictive models
that prevent downtime, using analytics
capabilities built on a Teradata Unified Data
Architecture (UDA) platform.
Expanded Collection of Data and the
Role of Hadoop
“We’re using Hadoop as the landing zone for
our machine data,” said Valeyko, “and as
we acquire new devices into the portfolio of
equipment that we support, we’re going to
continue to expand the amount of machine-learning
information that’s coming off those.
I think Hadoop has a big impact on our
ability to scale; the only way to effectively
store that volume of data is in a commodity
Hadoop environment.”
Before Hadoop, NCR could collect only
snapshot data, which limited the accuracy
of its predictive models. For example, the
company might collect snapshots from a
sampling of ATMs (perhaps once-daily performance
data). With an enterprise-grade
data lake built on Hadoop, NCR can monitor
every ATM it has manufactured and build
predictive models based on data from 100
percent of those ATMs.
But how does the company make sense
of it all? NCR uses the Teradata Enterprise
Data Warehouse for its day-to-day analysis
and the Teradata Aster Discovery Platform
deep analytics capabilities to move that data
around effectively in the architecture. With
Teradata Aster, NCR can query a truly global
data set. All of NCR’s ATMs, POS terminals,
and self-checkout devices collect and transmit
“heartbeat data” about the condition of
the device, whether components are operating
properly, and generally how the network
is performing.
“When we marry that data in Teradata Aster
with help-desk data from service technician
call logs, we can enhance the predictive
capabilities of those systems,” Valeyko
said. “That’s how we use Teradata Aster, for
discovery of algorithms that we use in our
predictive operational systems.”
NCR can sample data for analytics and
insight, and it can also use an entire set of
data, which includes historic data and new
data types, thanks to Hadoop.
Expanded Data Processing and Analysis
Capabilities
“Hadoop is a way for us to efficiently and
effectively store those large amounts of data,
even data we’re not sure yet how we’ll use,”
Valeyko said.
Collecting and querying a vast amount of
data creates new capabilities. “Right now
we’re using Hadoop to store our telephony
data—that’s a new usage we’ve found,” he
said, “and we’re using that data to analyze
our use of telepresence devices worldwide.
When we’re doing communications between
offices, we’re trying to figure out the most
efficient path and determine where to put
telephony equipment in our network.”
That’s a purely operational use of data, one
that NCR customers won’t see. Valeyko’s
group fosters “organizational altruism,” sharing
information across multiple lines of
business.
“An analyst responsible for financial products
might never have looked at retail service
calls,” he said. “Using big data, we’re able
to expand the scope of a query and root
cause analysis to see how a collocated
printer might be shared between a point-of-sale
device, an ATM, and a self-checkout
device,” with implications for retail as well
as financial services. This is the benefit of
being able to leverage all the data in one
place, for analytics across lines of business,
and sharing analytical insights across the
business as well.
The data is easy to find and move. “As we
find pieces of that lake that we want to analyze,
we can quickly move the data using
Unity Data Mover [another Teradata product]
into Teradata Aster for analysis and then use
that to build an algorithm, for example, and
put that into another operational system,”
Valeyko said.
Accelerated Time to Market
NCR continually feeds machine data back
into its processes, which enables continuous
engineering and improvement.
“Before Teradata Aster, it took six months
to create, test, and release an algorithm
that enabled predictive replacement of a
component,” Valeyko said. “Now we have an
effective algorithm in production in less than
three weeks.”
As Valeyko described it, the global scale
of its data set allows NCR to “pick off the
pieces of that large stream for further analysis
within Aster.
“We can quickly update the business rules in
those operational systems to ensure that we
have the right device, the right component,
even the right technician on the right truck,
to get to the customer as quickly as possible,”
he said.
NCR can perform enhanced scheduling of
support for those devices, matching components
and technicians to specific needs.
Finally, leveraging all the data, NCR can
identify a much higher percentage of actual
faults than it had in the past and reengineer
its processes to predict those faults and
preempt them.
The result, said Valeyko, is that “our customers
know that their customers have a great
experience with our devices.”
Data curation and analytics have a direct,
almost immediate impact upon NCR device
performance and customer experience. But
that all begins with the ability to collect,
store, and mine that data cost-effectively.
Lessons
IT can provide tremendous benefit across
an organization leveraging storage and processing
in the Hortonworks data platform for
Apache Hadoop, heavy analytics from Teradata
Aster, and a Teradata Enterprise Data
Warehouse. As NCR discovered, some of
those benefits and uses are obvious—such
as predictive modeling—but others are not
so obvious, such as analysis of its telepresence
devices and their use.
However, said Valeyko, “an organization may
need to be sold on the enterprisewide need.
It can be difficult to figure out how to fund a
discovery platform because it supports the
entire organization.”
NCR funded its initial investment in its
discovery platform with sponsorship from
its financial and services organizations, but
now enjoys enterprisewide support. At first
NCR’s data scientists were the most enthusiastic,
but as business users began to see
the utility of the Teradata UDA—for example,
in process reengineering, freight optimization,
and saving money on telephony—they
became interested as well. NCR formed a
data governance council with executive-level
representatives from across its lines of business,
including engineering and operations.
Customers, too, expect the deep discovery
capabilities that UDA enables through the
integration of the enterprise data warehouse
with Hadoop.
“What it comes down to is being able to look
at your customers individually and collectively
so that you can effectively give them
the best service you can,” Valeyko said.