Predictive Modeling
Wal-Mart had a clear model for how this product would perform on
a daily, if not hourly, basis. The model was aligned with the assump-
tions behind the decision to offer this computer at its “special”
price. The manager and even the stores could have their models to
check both assumptions and implementation quickly. It also
allowed headquarters to look at patterns between stores to unveil
any anomalies or innovations in implementation that might reveal
real-time learning.
Predictive models give us a way to “bet” on a probable outcome
and to alert us to the need to check our assumptions and even
implementation at every point along the way. In widely dispersed
organizations, these predictive models are even more critical, as
they can serve as early-warning systems for other stores. In the case
at hand, the stores on the West Coast were alerted earlier in the
day about this problem and were able to make on-the-fly changes.
The key is to not get overly invested in the predictions. Build a
model that can not only point out flaws in the plan but also vali-
date how good the planning was.
Real-Time Information for Real-Time Learning
Corporations must create new learning at the speed of change to
stay competitive. In the case of Wal-Mart, if headquarters had to
wait until the end of the week or the end of month to make
changes, corrective action would have been delayed beyond salva-
tion. Any enterprise has to assume that a percentage of its plans are
flawed; how rapidly flaws are discovered is directly related to how
rapidly we can learn, correct, test, and disseminate. Real-time
information can also lead to multiple attempts at correction. One
food store chain tries multiple approaches when a problem in stores
is noted. Several simultaneous corrections are made by different
stores, with the ability to rapidly track which approach has the
greatest impact on sales.
WHYAREN’TTHOSESPECIALSSELLINGTODAY? 5