Launching an Analytics Practice: Ten Steps to Success
- By Wayne Eckerson
- February 26, 2010
Everyone wants to move beyond reporting to
deliver value-added insights through analytics. The problem
is that few organizations know where to begin.
Here is a ten-step guide for launching a vibrant
Launching the Practice
Step 1: Find an Analyst. You can't do
analytics without an analyst! Most companies have one or
more analysts burrowed inside a department. Look for
someone who is bright, curious, likes to work with numbers
and has strong Excel and SQL skills, and is comfortable
working with various types of databases. Ideally, the
analyst should understand statistics and have a working
knowledge of OLAP and data mining tools.
Step 2: Find a Business Person. The
quickest way to kill an analytics practice is to talk about
predictive models, optimization, or statistics with a
business person. Instead, find one or more executives who
are receptive to testing key assumptions about how the
business works. For example, a retail executive might want
to know, "Why do customers stop buying our product?" A
social service agency might want to know, "Which spouses
are most likely not to pay alimony?" Ask them to dream up
as many hypotheses to their questions as possible and then
use those as inputs for your analysis.
Step 3: Gain Sponsorship. If step 2
piqued an executive's interest, then you have a sponsor.
Tell the sponsor what resources you need, if any, to
conduct the test. Perhaps you need permission to free up an
analyst for a week or two or hire a consultant to conduct
the analysis. Ideally, you should be able to make do with
people and tools you already have in-house. A good analyst
can work miracles with Excel and SQL, and there are many
open source data mining packages on the market today as
well as low-cost statistical add-ins for Excel and BI
Step 4: Don't Get Uppity. "You never
want to come across smarter than the executive you are
supporting," says Matthew Schwartz, a former director of
business analytics at Corporate Express. Don't ever portray
the model results as "the truth" -- executives don't trust
models unless they make intuitive sense or prove their
value in dollars and cents. For example, Schwartz was able
to get his director of marketing to buy in to the results
of a market-basket analysis because director recognized the
model's logic for cross-selling office products: "Ah!
People are buying office kits for new employees."
Step 5: Make It Actionable. A model is
worthless if people can't act on it. This often means
embedding the model in an operational application, such as
a Web site or customer-facing application, or distributing
the results in reports to salespeople or customer service
representatives. In either case, you need to strip out the
mathematics and dissect the model so it's understandable
and usable. For example, a sales report might say, "These
five customers are likely to stop purchasing office
products from us because they haven't bought toner in four
Step 6: Make It Proactive. The kiss of
death for an analytical model is to tell people something
they already know. Rather than tell salespeople about
customers who are purchasing fewer products and likely to
churn (as in the example in step 5), tell them about
customers who may stop purchasing because they have fallen
below a critical statistical threshold and are vulnerable
to competitive offers. You'll become an overnight hero.
Here's another example: Rather than forecast the number
loans that will go into default, identify the
characteristics of good loans and bake that information
into the loan origination process.
Sustaining the Analytics Practice
Let's assume your initial modeling efforts worked their
magic and garnered strong executive sponsorship. How do you
build and sustain an analytics practice? What
organizational and technical strategies do you employ to
ensure that your analysts are as productive as possible?
The following four steps will solidify your analytics
Step 7: Centralize and Standardize the Data.
The thing that slows down analysts the most is
having to collect data spread across multiple systems and
then clean, harmonize, and integrate it. Only then can
analysts start to study the data. Obviously, this is what a
data warehouse is designed to do, not an analyst, but a
data warehouse only helps if it contains all or most of the
data analysts need in a format they can readily use so they
don't have to hunt and reconcile data on their own.
Typically, analytical modelers need wide, flat tables with
hundreds of attributes to create models.
Step 8: Provide Open Access to Data.
Data warehouse administrators need to give analysts access
to the data warehouse without having to file a request and
wait weeks for an answer. Rather than broker access to the
data warehouse, administrators should create analytical
sandboxes using partitions and workload management that let
analysts upload their own data and comingle it with data in
the warehouse. This creates an analytical playground for
analysts and keeps them from creating renegade data marts
under their desks.
Step 9: Centralize Analysts. Contrary
to current practice, it's best to centralize analysts in an
Analytical Center of Excellence under the supervision of a
director of analytics. This creates a greater sense of
community and camaraderie among analysts and gives them
more opportunities for advancement within the organization.
It also minimizes the chance that they'll be lured away by
recruiters. Although they may be part of a shared services
group, analysts should be physically embedded within the
departments they support and have dotted-line
responsibility to those department heads.
Step 10: Offload Reporting. The
quickest to undermine the productivity of your top analysts
is to force them to field requests for ad hoc reports from
business users. To eliminate the reporting backlog, the BI
team and analysts need to work together to create a self-
service BI architecture that empowers business users to
generate their own reports and views. When designed
properly, these interactive reports and dashboards will
meet 60 to 80 percent of users' information needs, freeing
up business analysts and BI report developers to focus on
more value-added activities.
There you have it -- ten steps to analytical nirvana.
Easy to write, hard to do! Keep me informed about your
analytics journey and the lessons you learn along the way!
I'd love to hear your stories.