Imagine you’re managing a
company and your staff tells you
that you have 8,800 customers.
Then you find out that only
5,400 of them are “real.” How
would you feel?
• The customer base is smaller
than management thinks it is
• The proportion of revenue
from key accounts is different
than management thinks it is
• The profile of industries that
you service may be different
than management thinks it is
• Etc.
How
many
business
decisions
could
that
affect? If the company is
public, how much trouble
with regulators could that
cause?
Investor lawsuits?
How do you plan if you don’t
know where you really are?
In the case of one retail trade
association, we discovered that
most of the people who drove
the original decision to join had
moved to other positions. Their
replacements had continued the
memberships because it was
grandfathered into the budget,
but had no knowledge of the
association or the services it
offered.
The
association
thought
it
had strong relationships with
its clients, when in fact the
relationship was almost nonexistent.
The
association
thought it needed to refresh
its service offerings to address
flagging
sales,
when
the
problem was entirely different.
Statistical modeling on bad
data is a waste of time and
money. If you are building a
customer loyalty model and
one-quarter or more of the
data is wrong, the model isn’t
going to be particularly useful.
You can calculate the model
and run significance tests, of
course. Most statistical routines
require no information about the
quality of the input; they expect
the person running the model
to know the rules. Statistical
tests based on least squares
calculations (e.g., regression,
commonly used in key driver
analysis)
are
particularly
vulnerable to data outliers. The
bottom line is that the calculated
model is likely to be wrong.
Significance tests run on bad
data are meaningless.
The
solution
is
on-going
maintenance.
Maintenance
isn’t sexy, but it has to be done.
Like anything else humans build
(homes, for example), databases
require
garbage
removal
at regular intervals.
If that
doesn’t happen, the structure
can begin to smell. Corporate
environments are complicated
in that there are multiple
databases.
Errors in one
database can be propagated in
others and then fed back later.
That means corrections can be
overwritten by bad data. That
actually happens with unnerving
frequency.
Fixing
problems
typically
requires
a
commitment
of
resources
• Personnel, including a leader
appointed to oversee data
management
• A system for
reporting errors
identifying/
• A mechanism to ensure that
corrections
are
replicated
immediately
in
all
interconnected databases
Ultimately, you can’t change
human nature.
People
will enter bad information
through carelessness or for
some other reason. What
you need to do is to make
sure that when corrections
are made, the data stays
corrected.
You also can rely on your Vo
C vendor to provide annual
corrections.
However, that
means actually reaching out to
customers, not just sending them
an email.
About the author:
Victor Crain is a market research
veteran and currently a Senior
Partner at Crain Associates
Research LLC which is build
on a guild
business model.
His particular strength is in
combining data from multiple
sources
to
assess
market
direction. Some of the recent
projects that he has done
includes:
*
Understanding
customer
purchasing plans and channel
preferences during the holiday
shopping season
* Women’s attitudes
retirement planning
toward
*
Energy
management
corporate data centers
in
* Growing membership in a retail
energy cooperative
* Satisfaction among Federal
program managers with logistics
services
* Assessing/forecasting growth
in demand for coffee in fifteen
countries
You can follow his insightful tweets at
twitter.com/VicCrain