In efforts to ask for and accept customer input, many have lost sight of a very import- ant part of the process, translation.
With automated digital surveys and the
Internet, the advent of big data, and a plethora
of voice of the customer (VOC) techniques,
the use of customer input to drive design and
improvement is more possible than ever before,
but interpreting that input into useful guidance is
still an art form.
Whether we design new products or provide
services, even consultants must translate a customer’s spoken expectations into services and
solutions that will produce the results a customer needs.
Perhaps the most prevalent source of misleading guidance from data is how it was collected.
If the data wasn’t collected from a source that
represents your target audience, or if it wasn’t
collected appropriately, the results will be flawed.
It is very tempting to use data that is on hand
instead of going and collecting data. That is
especially true when data collection groups are
proactively collecting customer behavior data
and selling it.
Suppose we are producing and selling fishing
poles and the new model we want to market is
a particularly high-end, high-performance model.
We have data collected by one of the big-data
sources. With data on hand, why would we
spend the time, money, and energy to go collect
The data given to us was collected by tracing
cell phones as they moved through shopping
malls. The data tracks what stores were visited
by the same cell phones that visited the sporting
goods store at the mall. The correlations and
models of behavior have already been assembled
for us by the data collection service. All we need
to do is read the report and use that information
to help us plan our advertising for the new pole.
Except, why do we believe that the people
visiting the sporting goods store at the mall
represent our target market? Are the mall shop-
pers really our sport angler targets, or are they
teenagers and housewives? Just
because the data is handy,
doesn’t mean that it is the
data we need. If we fol-
lowed the handy data,
we might expend a
in all of the
and never reach our
Lastly, concerning trans-
lating data, I want to insert a thought about sta-
tistical methods and human behavior — they gen-
erally don’t cooperate. The correlation models of
collected data, in my experience, rarely achieve a
strong correlation, and even more rarely predict
For data collected about choices made by
people, the most common statistical tool to
translate that data into a probabilistic model is
multiple linear regression. Unfortunately, even if
we assess a comprehensive set of possible factors, human behavior rarely resolves into a model
with a correlation better than 60 percent. The
model’s prediction matches the actual data 60
percent of the time.
Also, even when we diligently reduce our
model down to only the statistically significant
factors, and even if we get lucky and our model
accurately predicts the past data 80 percent of
the time, in my experience, the model would not
predict future performance with the same 80
percent accuracy. The factors that affect human
behavior change much too frequently.
My experience with modeling human choice
behavior is not extensive. Even so, I have found
that identifying one or two consistent indicators
within the regression model can be as good a
predictor, or better, than the complete set identified by the regression model. The trick is to find
the indicators that most consistently drive the
outcome. They are not necessarily the ones that
have the greatest influence over the data set or
sets collected in the past. Sometimes the stron-gest influence is the most volatile or prone to be
usurped by another factor.
I’m not versed enough to identify a reliable
mathematical technique for doing so. Human
intuition and reason seems to be as good or
better; just look at the data, the factors, and the
models. Make your best guess and test your
hypothesis by using it to make decisions based
on the data you have. If you can predict the
outcome as well as, or better than, the statistical model you might have identified a simpler
(and possibly more reliable) predictor of future
Data can be misleading no matter how carefully we analyze it. The easiest cause of misleading
data is that it is not the data we need because it
doesn’t answer the question we have, or doesn’t
come from the target audience we want. Just
because it’s available and easy, doesn’t mean we
should use it. Just because we analyzed it statistically, doesn’t mean it’s a good predictor.
Statistical data is not the only VOC informa-
tion we gather. The more useful data is often the
commentary and direct feedback, or communi-
cated desires, of customers and potential cus-
dealing with this
data we know to
take great care
not to influence
by asking lead-
some teams get so caught up in trying not to
influence the data that they forget that, in the
end, the data must be carefully translated into
product specifications. Taking customer data at
face value and making that your product specification can be a big mistake.
I can’t think of an example that is better than
the classic roller board luggage example: Two
product development teams go about collecting
feedback and input from travelers and get the
same basic information. The biggest complaint
and coincidental desire is carrying heavy luggage
and for luggage to be lighter.
One team goes and writes “lighter” in the
product specification. The other team translates
the customer input and decides to design carry-on sized luggage with wheels and an extending handle. The second team understood that
the weight of the luggage is miniscule compared
to the weight of what is in the luggage, which
is completely out of the design team’s control.
Eliminating the effect of the weight is smarter
than eliminating weight. The translation from,
“make it lighter,” to “eliminate the discomfort of
carrying heavy luggage,” was the key to successful design.
It’s not a matter of engineers being smarter
than customers. The engineers simply understand the system better and have had more time
to consider options than customers who have a
limited opportunity to provide advice.
What a customer thinks he or she wants, and
what a customer actually needs, may not be
the same thing. Simply doing what a customer
asks may not be as successful as applying our
own expertise to understanding the customer’s
problem and suggesting a solution the customer
does not perceive.
While it is important not to influence customer
input with assumptions and expectations, we
must be careful to translate what customers
do provide into meaningful models, real root
cause connections, and successful directions.
Translating customer input can be as much art as
science, but it is absolutely the key to successful
use of VOC.
Stay wise, friends.
If you like what you just read, find more of
Alan’s thoughts at www.bizwizwithin.com.
By Alan Nicol
we might e
How to Make Customer Input Useful k I