Statistica: The Great Analytics Migration, Part 2: Process

SAS to Statistica:
The Great Dell Migration
Part One: People
Introduction
What do monarch butterlies, sockeye salmon, wildebeest,
sperm whales, red crabs and Dell have in common? They
have all undertaken a great migration fraught with challenge
and peril, but one which ensures survival.
Within weeks of acquiring the advanced analytics product Statistica, Dell set
out on one of the most ambitious migration projects since the company was
founded: a window ending December 31 (approximately six months) during
which all users would move of of SAS and adopt Statistica as the core analytics
platform companywide.
Whether your organization is migrating 10 users or 10,000 users in a 6-week or 6-year
project, we think you’ll end up asking questions like the ones we asked ourselves
during the project. This e-book describes the approaches to people, process and
technology that contribute to the success of a migration from SAS to Statistica.
Dell set out on one of the most ambitious
migration projects since the company
was founded.
To give you insight into the scope and objectives of our migration project, here are
a few high-level results:
• Hundreds of users migrated worldwide
• Substantial, bottom-line impact due to saved fees
• 300+ projects across multiple business units migrated from SAS to Statistica
• Migration project team consisting of 12 points of contact for users
• Overall project duration of less than nine months, with user migration in less than
six months
2
Of course, every organization is diferent. We at Dell achieved these results in so
little time because of our commitment and the extraordinary eforts of teams all
around the company, as you'll see in Part 2. We can't guarantee similar savings
from every migration to Statistica, but we have seen a bottom-line impact that
similar companies making a similar transition could expect to see. In any event, we
invite you to learn from our lessons.
| © 2015 Dell, Inc. All rights reserved | Share:
First, why migrate to
Statistica?
organization. But they didn’t know what a migration project would entail and like
most companies, they were concerned about the downside of an unsuccessful one.
Without a good answer to this question, the others don’t matter.
Your mileage may vary, but Dell was spending a great deal of money every year to
keep SAS in place. Of course, everybody wanted to save money, but everybody
also knew the migration was going to involve a lot of risk, work and change, so we
had to justify the efort.
You want to enable analytics in the enterprise. We’ve all read about it in books,
magazines and news articles — we need to do something about analytics and big
data. Organizations that embed analytics within all parts of their business to make
faster decisions and improve decision making, planning and forecasting have a
distinct competitive advantage. Unfortunately, there is a skills shortage, so we need
a software package that plays well with existing IT investments and is suiciently
easy to use. The goal is to enable all users — experts and line-of-business users
alike — to make the most of their data.
You want to lower the cost of software licensing. Even before Dell acquired Statistica
(formerly of StatSoft), customers and prospects had been telling us that they wanted
analytics software that was less expensive. That’s no surprise, but more important,
they wanted software that made analytics accessible to more people in their
3
You can ind better value for money. A company the size of Dell derives signiicant
value and competitive advantage from applying analytics in areas like marketing,
price optimization, forecasting, technical support, supply chain optimization,
preventive maintenance and inancial credit risk analysis. We believed we could get
better value for less money with the full range of analytics muscle and ease of use
we saw in Statistica.
You want your analysts to analyze data, not write code. SAS is powerful, but you
have to hire people to write and maintain code and administer the complex system to
get the most out of it. We, along with many of our customers, were having increasing
trouble inding good replacements for the SAS-savvy people who were leaving or
retiring, so the cost of keeping SAS was rising beyond our annual licensing fees.
It’s the best way to improve the product. You know that the path to a better
product leads from your front door to your loading dock. Given that Dell is making
signiicant investments in Statistica, it made all the sense in the world to use it
ourselves and see what our customers were experiencing.
It’s part of why Dell acquired Statistica in the irst place. Dell’s acquisition strategy
is driven in part by the opportunity to roll out and use acquired products internally,
improve them and save money with them, then take that message out to our
customers. When prospects ask whether Dell uses the product they are promoting,
our sales teams need to be able to answer, “Yes, we drink our own champagne.”
| © 2015 Dell, Inc. All rights reserved | Share:
How many people are you
going to afect? Where are
they?
As with any software product, most of your users are casual,
working with a subset of the product functions to accomplish their daily tasks. Then you have a group of power users
who eat, sleep and breathe the product.
When we started the project, we quickly identiied hundreds of users whom we
needed to move to Statistica, including 170 users in Dell Global Analytics (DGA).
DGA provides analytics expertise and support to diferent functional organizations
throughout the company — for example, inance, the customer service center,
dell.com, marketing and sales, supply chain, operations, pricing and product
management — that do not have their own internal analytics expertise.
When we started the project, we quickly
identiied hundreds of users whom we
needed to move to Statistica.
Our marketing operations teams all over the world rely heavily on BI and analytics
tools, as do Technical Support, Supply Chain and Dell Financial Services. In
short, analytics is pervasive throughout Dell and is a big part of our competitive
advantage. The migration from SAS to Statistica afected our data analysts and
users in all of those groups. But more important, since analytics are ubiquitous at
Dell and embedded in our systems and processes, there are many more people
who consume and rely on the output. Any adverse change to this output could
drastically impact the business.
4
| © 2015 Dell, Inc. All rights reserved | Share:
How do your people use
analytics?
In almost every company, analytics are inding their way
into marketing to help make sense of what customers and
prospects are saying, tweeting, looking for and buying. We
use analytics to personalize ofers, attract prospects and
keep existing customers. We study things like customer
churn, cross-sell/upsell opportunities, customer sentiment
and satisfaction.
Analytics use case: Customer churn
Suppose that a logistic regression model tells us we’re in danger of losing ive
customers to churn because they are likely to buy another company’s products
instead of Dell’s. We decide to ofer each of them a $50 discount on a Dell product
and they take it. So it costs us $250 to keep them as customers. That sounds like a
good use of $250. But suppose it wasn’t the right model to use and the customers
weren’t really in danger of churn. Then we’ve wasted $250.
With the right model and analytics, it would
have cost us only $50 to save $500.
In our daily operations, we apply analysis to improve the quality of our
manufacturing processes. In Professional Services, our services teams embed it in
customer solutions as part of service engagements. For Customer Support, we use
analytics for predictive/prescriptive/preventive maintenance on hardware products
we sell.
Now look at the lip side. Suppose our Random Forest model tells us that a
particular customer is happy and not in danger of churning, so we don’t ofer her
the discount. If it turns out she buys another company’s product, we might lose
$500 of revenue. With the right model and analytics, it would have cost us only $50
to save $500.
We also use analytics to assess and control
the potential cost and risk of our decisions.
So, you might say that the Random Forest is a better model because it results in
just one error, but it’s a more costly error than the ive combined from the logistic
regression model. Our models need to take that into account, and these tradeofs
of model accuracy and costs associated with incorrect predictions are easy to
illustrate with Statistica.
Dell Financial Services relies heavily on analytics for modeling, assessing credit risk
and detecting fraud. Their models are closely tied to forecasts and bank rates, so
statistical analysis is part of what they do day in and day out.
We use analytics to evaluate models based on not only the overall error rate but
also the type of error that the model is making. It allows us to make more-precise,
better-informed decisions about the models we apply to business factors like
customer churn, product upsell, service level agreements, competitive pricing and
delivery dates.
Like you, we use analytics to make decisions. But since not all of our decisions are
perfect, we also use analytics to assess and control the potential cost and risk of
our decisions.
5
| © 2015 Dell, Inc. All rights reserved | Share:
Analytics use case: Direct mail campaign
Assumptions
Data
• $2 to mail each prospect
Mailing list
1M prospects
• 1 out of 100 will buy
Results
• Proit - Revenue - Cost
Mail cost $2
• -($220 × 10,000) - (1M × $2)
• $200,000
• $220 proit for each response
The value of a prediction
Assumptions
Data
Analytics model output:
Mailing list
250K prospects
• 25% of the entire list are 3× more
likely to respond
Results
• Proit - Revenue - Cost
Mail cost $2
• -($220 × 7,500) - (250K × $2)
• $1,150,000
• 5.75× improvement by mailing
fewer people
Adapted from Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie or Die by Eric Siegel
6
| © 2015 Dell, Inc. All rights reserved | Share:
How will people in your
organization react?
You’ll know the answer to this question in almost no time,
because word travels fast. At Dell, most people’s reactions
landed in one of three buckets:
1.
“But we’ve never used Statistica.”
Since our users had been using SAS for many years, they weren’t familiar with
how robust an analytics platform Statistica is, so naturally they were skeptical.
We pay them to be skeptical.
They knew that their work consisted of mission-critical analytics in SAS and
assumed (incorrectly, as it turned out) that Statistica wasn’t up to it.
2. “We’ve spent years writing thousands of lines of SAS code. We don’t want to
just throw that away.”
Our users anticipated a great deal of work in trying to replicate in Statistica the
programs they had built in SAS, so naturally they balked. Who wouldn’t feel
that way?
7
3.
“We consider ourselves SAS professionals and analysts irst, and employees of
Dell second. For career longevity and our ability to do our job, we believe that
it's really important to continue using SAS.”
That’s a tough one. We found a number of heavy SAS users who had been
working with the product for over 20 years. They were comfortable using it
and they had grown, evolved and become pretty good with it over much of
their career. Asking them to switch to something they didn't know was a huge
disruption for them.
Most users had never heard of Statistica — let alone used it — so it was the
devil they knew versus the devil they didn’t know. Plus, many felt an emotional
attachment to the tried and true product.
We addressed their reactions by having our migration leads from Statistica sit
down and show them that their long years of work would not be simply discarded.
The leads examined the techniques and functions our users had worked with in
SAS, such as K-means clustering, polynomial regression, GLM, ARIMA and neural
networks, then demonstrated how to replicate and enhance them in Statistica.
Nearly all the techniques they had used in SAS were easier to implement in
Statistica without the need to write thousands of lines of code. Users simply
dragged and dropped icons onto the Statistica workspace in its easy-to-use
graphical user interface.
You’ll see that making the migration a success requires consistent communication,
executive buy-in, technical support, training and encouragement, mostly
from peers.
| © 2015 Dell, Inc. All rights reserved | Share:
How will you get them
on board?
Explaining your migration project is one thing; getting all
your users on board with it is another.
First, all of our users and business units said they would need more than a year
to migrate successfully to Statistica. Since the mandate was to migrate before
December 31, we had to come up with several diferent approaches to make the
project palatable to our users:
• Making sure we had identiied all current SAS users and properly
communicating the change to them.
Nobody wants to ind out about something like this at the water cooler. Our
migration project team compiled a list of active SAS users and identiied the
afected executives and managers who would be responsible for keeping the
users apprised of project status. We were mindful that many users would initially
regard the change as a diicult one, so we kept the communications consistent,
positive and frequent so they could digest them completely. We held several
workshops early in the project to ield and address users’ concerns.
• Giving people early access to Statistica.
It’s easy to overlook this as a migration project starts up, but as the message
went out and we talked to group managers about the timeline, they naturally
replied, “We’ll need access to the tool if you expect us to hit those dates.” We
had to show them and their users what they were in for and let them start
working with Statistica. That kicked of the next set of conversations about
setting up infrastructure, rolling out the product internally and paying for it.
• Allaying the concerns of the dyed-in-the-wool SAS professionals.
Using a new tool does not strip those people of their SAS expertise or
marketability, but they consider themselves part of a community that extends
beyond Dell. We knew we couldn’t address that on the emotional level, so we
addressed it on the technical level by deeply discussing their analytics tasks and
demonstrating how Statistica could satisfy them.
cases where it wasn’t the best it and where it was overkill. They applied it to
functions like data movement (ETL), data aggregation and preparation, which
don’t require advanced analytics technologies.
• Guiding users away from ineiciencies and shadow IT.
By centralizing data movement tasks in oicially supported environments and
processes, we lowered internal risk and improved performance within Dell. In
cases of ad hoc data movement, the migration leads showed users how to
accomplish the same task with Statistica or with tools like Toad Data Point to get
them on board with the migration.
We knew that carrot works better than stick at
Dell, so that’s how we approached migration.
• Launching a contest.
To inspire teams to take Statistica with both hands and apply it, we ran an
internal contest. Teams of one to ive people sprang up around the company,
built analytic models to address real-world questions in Dell and competed
against one another. The contest not only attracted a lot of attention to
Statistica but also helped us solve business problems.
Every organization is a diferent ship, steered diferently. You can try a mindset
like “Between now and the end of the year we’re switching to a tool you don’t
know and which you can’t yet access, and in the meantime you have to do your
day job,” but we knew that carrot works better than stick at Dell, so that’s how we
approached migration.
Since this project was mandated by Michael Dell, it enjoyed the support of our
executive leadership. Whenever the migration team encountered a substantial
hurdle, its members got the support they needed. In general, we overcame
most of our obstacles with additional training, support and one-on-one sessions
customized to the teams that needed them.
In short: communication, communication, communication. Add executive support,
cultivate the employees who are not afraid of the challenge, and persist.
• Putting SAS in its proper context.
Why use a Ferrari to deliver a load of dirt? We found that, for lack of a better
(and lower-priced) tool, many people had grown accustomed to using SAS in
8
| © 2015 Dell, Inc. All rights reserved | Share:
Who gets to be an exception,
and why? Lesson learned.
When you look back, you’ll see where you had to change
your original plan and how you had to change it for the
people factor.
We learned plenty of lessons about how people deal with an extensive tool change
in an organization. The bigger lesson, though, was about allowing for exceptions
on a migration project.
By the end of the project, all but 16 of our users had moved completely of of
SAS before the December 31 deadline. Most of Dell Financial Services migrated
successfully, but the modeling team made the case that they needed more time to
migrate their most sensitive functions and received an extension to continue using
SAS for six more months. More important, they demonstrated that any interruption
or luctuation in their credit scoring and fraud models would introduce greater risk
than it would be worth to meet the initial deadline.
That made sense. Some groups use analytics to predict, say, staing levels for IT
call centers. If their predictions are wrong, customers may have to wait longer,
but the immediate impact is not as prominent as a hiccup in inancial modeling. It
was unfortunate that not every user was on Statistica by January 1; the downside,
however, might have been even more unfortunate, so the project team worked
with the modelers on a speciic plan that extended the migration timeline for that
small group.
Other than that, it speaks well to the quality of our employees (and of the product)
that the transition to Statistica was smooth for more than 95 percent of them,
especially considering that we had until December 31 to achieve what everyone
thought would take more than a year to accomplish. We had to help some of
them over their emotional obstacles to using a diferent tool in a diferent way, but
what matters is arriving at the same result. Several teams performed their own
comparisons and convinced themselves of that.
Learn more
Part 2 in this series covers process and the many things we put in place to make
the migration project a success. Part 3 describes the technology component of the
migration project, from architecture to onboarding users.
Visit software.dell.com/products/statistica/
9
| © 2015 Dell, Inc. All rights reserved | Share:
Statistica: The Great
Analytics Migration
Part 2: Process
How do you eat an
elephant?
One bite at a time, of course.
That may not be the exact question you ask yourself as you move from thinking
about people (as we described in Part 1) to thinking about process, but the
questions in this phase boil down to that.
How are you going to take all of the people in the organization who depend on
your analytics platform — in Dell’s case, hundreds of users worldwide — and move
them to a new and diferent platform?
Our high-level plan was to stand up a parallel
Statistica environment and provide our users
what they needed to move onto it.
We avoided the shock treatment of turning of our legacy analytics product one
night and then turning on the Statistica environment the next morning. In general,
to help ease the transition and ensure that our analytics processes kept running
without disrupting the business, our high-level plan was to stand up a parallel
Statistica environment and provide our users what they needed to move onto it.
We ran the two platforms in parallel for several months so users could see that,
with the training and support we provided, they would have the time, opportunity
and motivation to move their work into Statistica. In fact, we made it easy for
them to compare the two products, satisfy their own curiosity and develop their
analytical muscles. We clearly communicated a goal date for them to stop using
the legacy product altogether and use Statistica exclusively.
2
© 2015 Dell, Inc. All rights reserved | Share:
Where do you start?
“Hmm,” you think. “I wonder if that means we’re going to stop using Product A and
start using Product B.”
Imagine that your company has been using a product from
Tool Supplier A for many years. It could be a chemical, an
electronic device or a gas-powered turbine — anything you
and your co-workers use to get your work done. You check
your email one morning in mid-March and discover that
your company has acquired Tool Supplier B, which sells a
competing product.
Around April 1, you stop wondering because executive staf has decided that the
entire company is going to migrate to Product B by the end of the year.
For Dell, the project started that way. We learned in April that our companywide
goal was to stand up a new enterprise-wide instance of Statistica, move of our
legacy analytics environment and migrate all of our existing users to Statistica by
December 31.
Nobody looked forward to the perceived
learning curve our users faced in moving from
a tool of long-standing use to a new one that
few of our co-workers had ever used.
Obviously, this task looked formidable because of the short timeline to get
software, hardware and network infrastructure into place and working properly.
And certainly nobody looked forward to the perceived learning curve our users
faced in moving from a tool of long-standing use to a new one that few of our coworkers had ever used.
But the thing that made it most diicult to set our compass and see where to start
was that this initiative came out of the blue for most of us. In Dell, as in most large
companies, a lot of forethought goes into goals of this scale. We usually plan
them out months in advance and announce them at the beginning of the iscal
year. We submit business requirements documents, deliver systems architecture
requirements, provide concepts back to the business teams, identify sources of
funding, agree on start dates and estimated timelines and form a project team.
Everyone knows where to start. This time we had to put all of that in place during
the irst weeks after the acquisition and the goals had been announced.
You may not have to switch that quickly, but you will have to igure out the best
place for your organization to start. In most cases, that will be to develop timelines
for rolling out the software, hardware and infrastructure you’ll need, and for
putting in place the education and training, centers of excellence, user community
and help desk resources you’ll need for success.
3
© 2015 Dell, Inc. All rights reserved | Share:
How do you schedule IT
delivery?
Your mileage may certainly vary, but we followed the
following timeline for our migration:
Laying the groundwork
April — Where are all the pieces? First, we engaged every available resource from
the Statistica team to help Dell’s IT staf understand the software coniguration,
infrastructure and security protocols for the hardware on which we were going to
run Statistica. Then we worked out which IT skills and resources we would need to
put all of those in place.
Internally, the hardware was relatively easy to address. Our technical staf and
IT staf spent a fair amount of time with the Statistica development team to
understand how Statistica was going to interact with all of the other software that
Dell runs. But that didn’t mean the hardware would magically appear; we spent a
lot of April developing a delivery schedule for that.
Finally, we began identifying the analytics user base within Dell — the “customers.”
Who are they? Where do they work? Who are the contacts we need to work with
so we can start putting together a customer plan for the migration?
Dell acquires Statistica
(from StatSoft)
March 24
April–May
May — How long will this take? Once we’d identiied the users, we asked how long
they thought it would take them to migrate. Most of them answered, “Let me start
using Statistica, and I’ll let you know. But if I had to guess, I’d say it will take at least
a year.”
That kicked of a couple of weeks of dealing with purely logistical questions from
the user community: Who’s going to pay for this? How do we get the hardware?
Which data center will host them? Do we need multiple data centers? How will
we accommodate of-network users? From those answers, we came up with an
estimate of the hardware needed for the migration, then secured approval to
proceed based on the estimate.
By the end of May, we had a solid plan to present to the IT and business leaders
and say, “Here’s what we are going to do.”
June — How many users really need to migrate? Our IT team spent June ironing
out technical details around security and coniguration.
Meanwhile, on the business side, we performed a valuable exercise in
rationalization with the Centers of Excellence (CoEs) for each of our business
functions — inance, professional services, customer support, marketing
operations and sales. Originally, we had asked the CoEs to identify their analytics
users. But many users were running the analytics product for its non-analytic
features like data management and extract, transform, load (ETL), and others
weren’t actively using it anymore.
We asked the CoEs to examine more precisely which users really needed analytical
functions, and we rationalized down to a few hundred validated candidates.
How many users?
Migration
Security, coniguration, user
inventory, ready by August 1
Training continues, CoEs established,
users onboarded
June–July
Executive decision made to migrate
to Statistica company-wide
July–August
Installation and training
December 31
Deadline
Migration complete
(extended to June 30
for 16 users in Dell
Financial Services)
Initial IT planning and
discussions begin
4
August–December
© 2015 Dell, Inc. All rights reserved | Share:
Knowing exactly how many people needed to migrate and who they were, we were
able to estimate the number of servers we would need, how to conigure them and
which groups we would have to set up for a successful transition.
we started the procurement process with our eye on the August 1 deadline. Late
August probably would have been a more realistic target date, but we let everyone
in the supply chain know the urgency of the project.
In late June, our chief data oicer made the decision to move forward. That was
the good news. Also, he wanted Statistica implemented and ready for migration by
August 1. That would have been good news, too, except that everyone suspected
it would take much longer than that — six to 12 weeks — just to get the hardware.
Run, Forrest, run!
To our surprise and relief, our colleagues amazed us with what they are able to do
when they really put their collective mind to it. We secured the physical servers,
installed them in a data center, got them online, had the software conigured
and had Statistica available for delivery to users on August 1. In other words, our
co-workers helped us accomplish in ive weeks what would normally take up to
three months. This was possible due to transparent communication of the plan,
the timelines and the urgency of meeting speciic deadlines. When you empower
people with information, they are more willing and able to do really good things.
July — How fast can we get the hardware up and running? With approval to
purchase and install the hardware and conigure the software for delivery to users,
5
© 2015 Dell, Inc. All rights reserved | Share:
How do you train and
support the users?
Having hardware and software in place means you can put
Statistica in the hands of your users, but don’t forget that you
need comprehensive training and technical support for them
to successfully migrate to a new technology, especially when
you’re on a tight schedule.
Training
Some teams don’t wait to be invited for training; they get a jump on it.
When the Dell Global Analytics (DGA) group, many of them based in Bangalore,
India, learned of the Statistica acquisition, they assumed that Statistica would
replace the legacy product eventually. Immediately, they began thinking about
ways to use Statistica in DGA. They contacted the Dell oice in Delhi, obtained
a trial license of Statistica and invited regional Statistica experts to their oice to
provide an introduction to the product’s capabilities. Thus, they had an informal
orientation long before the formal migration project was announced.
Elsewhere, beginning in mid-July, our Statistica subject matter experts (SMEs)
facilitated three rounds of training:
• Train-the-trainer sessions — We asked each of the teams that would be using
Statistica to send two people for a two-week training session in July, while IT
was still installing hardware and coniguring software. Statistica SMEs traveled to
Dell HQ in Austin, Texas, to instruct 40 users from all over the world in train-thetrainer sessions. Our goal was to create champions and advocates in diferent
groups around the company and around the world, particularly within DGA.
• In-person, deep-dive training — Later, to meet the particular needs of our
users concentrated in DGA in India, we conducted a formal session for the
DGA professionals in Bangalore.
• Web conference — We also recorded virtual sessions and made them available
as knowledge transfer resources for all migration candidates. The recordings
remain available to all appropriate employees within Dell.
The train-the-trainer and in-person sessions gave attendees their irst opportunity
to learn about and use Statistica, even before it was available at their desks.
6
Technical support
As more users began adopting Statistica, we created a separate, speciic SharePoint
site where migrating users could submit support issues or questions. We built
up a knowledge repository with frequently asked questions and how-to articles,
and the Statistica team assigned several sales engineers to provide answers both
electronically and in person.
We decided to keep the information separate from the standard tech support
queue for three reasons:
• We could apply a service-level agreement (SLA) and expectation of response
that were diferent from those applied to the standard queue.
• We monitored the queue, escalated urgent items and kept an eye out in case a
signiicant need for widespread training should suddenly come to light.
• If a group had questions or concerns about a particular area of Statistica, we
could switch from technical support to a model more like a professional
services engagement and a workshop.
With IT assets, training and support in place, we embarked on the actual
migration work.
© 2015 Dell, Inc. All rights reserved | Share:
What goes on during the
actual migration?
The important thing to keep in mind is that the process of
migration happens behind the scenes — almost invisibly —
as users steadily adopt the new tool and cease using the
old one.
In Dell’s case, although IT managed the project and kept executive staf apprised
of overall status, it was the business leads in the CoEs who knew how close or far
each user was from completely migrating to Statistica. And for several months,
even that was diicult to discern because of all the moving parts inside the
actual migration.
What goes on during the actual migration is a
lot more than merely replacing one software
tool with another.
You bump into resource constraints. During the migration, users spent time
converting their daily tasks — processes, models, algorithms and so on — to the
Statistica environment while continuing to perform their existing functions with
the existing product. These employees were already allocated for 75 to 90 percent
of their time, so when the company asked them to take on the additional work of
migration, they began to wonder (aloud, mostly) if they could meet the December
31 deadline.
You spend time double-checking. Users ran tasks on both products to ensure they
were getting the same result from both tools. (That’s the healthy skepticism we hire
them for.) They knew how their models and outputs should look in the old product,
and they igured out how to get them to look that way in Statistica as well. That
was the result of all the resources Dell dedicated to the project, such as training,
technical support and informal knowledge transfer.
You herd cats. The individual business groups run their own analytics business
processes, so they formulated their own migration plans to meet the deadlines and
started executing on them at their own pace. Of course, many of the timelines and
migration plans were developed with input from the Statistica SME team.
You stop to correct ineiciencies. The last thing you want to think about in the
middle of migration is process improvement, but ineicient processes come out
of the woodwork, and you have to deal with them. We discovered data extraction
tasks that took hours with existing tools because they had never been properly
conigured or had not been updated for changing requirements. Correcting
ineicient processes wasn’t initially part of the migration plan, but we knew we
couldn’t ignore them, particularly when the analytics innovation team saw the
chance to whittle a four-hour data extraction process down to a few minutes in
Statistica, for example.
You eliminate over-dependence on the incumbent software platform. As
described above, the CoEs discovered that many users had been using the
analytics product for non-analytic functions like data management and ETL. Users
soon found that Statistica did not yet support some of the data management
functions as well or in the same manner as the legacy programming language.
Besides showing the product development team features to add or enhance
in future versions of Statistica, the discovery opened up another migration
opportunity: to steer users toward Dell Software products designed for data
management and ETL, such as Toad Data Point and Toad Intelligence Central.
The Statistica-Toad combination allowed those teams to meet their deadline for
migrating of the legacy product at a fraction of the cost. Also, because these
products are all part of the Dell Software portfolio, we were able to begin work on
integrating Statistica and our Toad products more closely.
1
You run a contest. To encourage our power users to showcase their talents, we
hosted a contest in which our analytics teams submitted business cases with
real-world examples of how it was easier, more appropriate or more robust to
deploy certain analytics processes using Statistica. More than 20 teams developed
advanced analytics processes and worklows that highlighted their migration
story through production applications including customer lifetime value, data
manipulation, pricing optimization and — the winner — predicting rates of hard
drive failure.
In short, what goes on during the actual migration is a lot more than merely
replacing one software tool with another.
1
7
Annual maintenance cost for Statistica licensing was approximately 70 percent lower.
© 2015 Dell, Inc. All rights reserved | Share:
How do you stick to the
deadline?
The middle of the timeline is often the smoothest part of the
process. It’s usually at the beginning and at the end of the
project where you encounter the most resistance.
With a December 31 deadline, our biggest obstacles arose between the
U.S. holidays of Thanksgiving (late November) and Christmas. Over the long
Thanksgiving weekend, we saw people planning holiday absences and determined
that one-week tasks and approval cycles were going to start dragging out to twoand three-week delays. Users still had access to both products, and we detected
that some were reluctant to relinquish their licenses to the legacy product until
they absolutely had to do so. To avoid blowing our deadline, we appealed to the
CoE leads during early December.
Pressure from the business leads
“Your people are doing great,” we told them, “but the number of users who have
not yet migrated and relinquished access voluntarily is awfully high considering
how little time remains on the project plan.”
We had them identify a lead on each functional team to tell their users pointedly, “If
you're using Statistica and you haven't touched the old product in a while, then I
need you to give up your access.”
We found that most of those users simply said, “OK,” and relinquished their
licenses. During the two weeks after Thanksgiving, we saw a dramatic increase in
the number of users switching to exclusive use of Statistica. By mid-December, all
but a few dozen of our validated migration candidates had relinquished access.
If nobody is telling you that you have to
stop using the incumbent analytics software
platform right now, then naturally you’ll keep
it in your back pocket.
Pressure from IT
IT project management made phone calls to the inal three dozen users to obtain
speciic reasons why they wanted to retain access until the December 31 deadline.
Among the holdouts were regional oices that planned to remain open during
the last week of December (when much of Dell is closed for business) and
continue working on year-end analysis. They had sound business reasons for not
relinquishing until necessary, so IT allowed them to retain access until the deadline.
The other signiicant holdout was Dell Financial Services (DFS). As described in
Part 1, the 16 users in DFS obtained permission from the company’s executive staf
to continue using the legacy product through the following June.
Our main lesson about sticking to the deadline: If nobody is telling you that you
have to stop using the incumbent analytics software platform right now, and if
retaining access to it doesn’t appear to cost anything, then naturally you’ll keep
it in your back pocket until the very end, just in case. This eleventh-hour cleanup
operation is an integral part of sticking to the deadline.
8
© 2015 Dell, Inc. All rights reserved | Share:
What else do you learn
during a migration project?
You learn many of your lessons while the project is still under
way. Others don’t occur to you until the dust settles and
people begin talking about the project in the past tense.
Unexpected dividends
As described in “What goes on during the actual migration?” above, we realized a
great deal more eiciency and process improvement than we had anticipated. The
migration project was an unexpected opportunity to turn over a lot of rocks and
scrutinize how we’d been doing things for a long time. We didn’t speciically set
out to make a better Dell, but when the chance arose, we couldn’t pass it up.
If you’re migrating more than a handful of users, you need an organizational
structure that is segmented into manageable sizes of user groups. CoEs can do
that because they understand what their users do with the incumbent product,
what they wish they could do with it and what they need in the new one. While
IT deploys the application and puts infrastructure in place, the CoE handles the
migration at the user level, a task for which IT is not set up.
Users with three jobs
As described above, many users performed double duty by not only building new
models and processes in Statistica, but also maintaining existing models and
processes in the legacy product. During the train-the-trainer phase of the formal
education process, some users performed triple duty.
For train-the-trainer sessions, each business unit identiied a small set of power
users who underwent in-depth training. These trained users interacted regularly
with Statistica SMEs and, in theory, would be the irst line of reference for general
questions within their business or functional areas.
In practice, however, we found that the teams still needed support from the
Statistica SMEs for three reasons:
The migration project was an unexpected
opportunity to turn over a lot of rocks and
scrutinize how we’d been doing things for a
long time.
The importance of CoEs
IT managed our migration project because software and hardware were
involved. IT was responsible for knowing where each team was in the process of
relinquishing the old product, but not for knowing status at the user level.
Dell has analytics and business intelligence CoEs within inance, professional
services, customer support, marketing operations and sales. The IT project
managers asked each CoE to make a single migration lead accountable for its
population of users. The CoE enabled IT to break the user population down
into smaller units and have the organizations that depended on analytics drive
the migration.
9
• Some teams were not physically co-located, so the trained users had to help
co-workers in other oices or time zones.
• When applied to an intricate analytics modeling problem, even a software
solution as robust and easy to use as Statistica does not always ofer a
simple solution.
• The trained users also had to perform their day-to-day tasks while migrating
to Statistica.
Ultimately, Dell conducted formal classroom training for all users (see the section
“How do you train and support the users?” above). Train-the-trainer sessions
were a good idea, but we underestimated the amount of efort required for these
individuals to learn the new product, act as the primary Statistica point of contact
for their team and perform normal analytics tasks while migrating.
Learn more
Part 1 in this series covers the people component of our Statistica migration. Part 3
describes the technology component of the migration project, from architecture
to tooling.
Learn more at dellsoftware.com/statistica
© 2015 Dell, Inc. All rights reserved | Share:
For More Information
© 2015 Dell, Inc. ALL RIGHTS RESERVED. This document contains proprietary
information protected by copyright. No part of this document may be reproduced
or transmitted in any form or by any means, electronic or mechanical, including
photocopying and recording for any purpose without the written permission of
Dell, Inc. (“Dell”).
Dell, Dell Software, the Dell Software logo and products — as identiied in this
document — are registered trademarks of Dell, Inc. in the U.S.A. and/or other
countries. All other trademarks and registered trademarks are property of their
respective owners.
PURPOSE, OR NON-INFRINGEMENT. IN NO EVENT SHALL DELL BE LIABLE FOR
ANY DIRECT, INDIRECT, CONSEQUENTIAL, PUNITIVE, SPECIAL OR INCIDENTAL
DAMAGES (INCLUDING, WITHOUT LIMITATION, DAMAGES FOR LOSS OF PROFITS,
BUSINESS INTERRUPTION OR LOSS OF INFORMATION) ARISING OUT OF THE USE
OR INABILITY TO USE THIS DOCUMENT, EVEN IF DELL HAS BEEN ADVISED OF
THE POSSIBILITY OF SUCH DAMAGES. Dell makes no representations or warranties
with respect to the accuracy or completeness of the contents of this document
and reserves the right to make changes to speciications and product descriptions
at any time without notice. Dell does not make any commitment to update the
information contained in this document.
The information in this document is provided in connection with Dell products.
No license, express or implied, by estoppel or otherwise, to any intellectual
property right is granted by this document or in connection with the sale of
Dell products. EXCEPT AS SET FORTH IN DELL’S TERMS AND CONDITIONS AS
SPECIFIED IN THE LICENSE AGREEMENT FOR THIS PRODUCT, DELL ASSUMES NO
LIABILITY WHATSOEVER AND DISCLAIMS ANY EXPRESS, IMPLIED OR STATUTORY
WARRANTY RELATING TO ITS PRODUCTS INCLUDING, BUT NOT LIMITED TO,
THE IMPLIED WARRANTY OF MERCHANTABILITY, FITNESS FOR A PARTICULAR
About Dell Software
Dell Software helps customers unlock greater potential through the power of
technology — delivering scalable, afordable and simple-to-use solutions that
simplify IT and mitigate risk. The Dell Software portfolio addresses ive key areas of
customer needs: data center and cloud management, information management,
mobile workforce management, security and data protection. This software,
when combined with Dell hardware and services, drives unmatched eiciency and
productivity to accelerate business results. www.dellsoftware.com.
If you have any questions regarding your potential use of
this material, contact:
Dell Software
5 Polaris Way
Aliso Viejo, CA 92656
www.dellsoftware.com
Refer to our Web site for regional and international
oice information.
Ebook-TheGreatAnalyticsMigration-Part2-US-GM-26622
10
© 2015 Dell, Inc. All rights reserved | Share: