low the computer to take actions that are
• Automated actions should be modeled
on how your most competent employees
would solve the problem. Expert systems
resolve many problems attributable to
training, inexperience, complacency, and
• Accurate real-time action can only be
taken against accurate real-time data.
So how do we move beyond real-time exception management and into the flawless
execution of the real-time supply chain?
By teaching the supply chain to execute itself and report the exceptions to us, rather
than inserting ourselves into each step of
the supply chain execution process.
We have done this at Sonic by building a
system that doesn’t allow mistakes to occur
a second time, and creates and processes
purchase orders to completion faster than
you could print the paper planning reports.
Our system also fixes wrong, missing, and
outdated data using the same solutions as
our best employees would. And it updates
thousands of records instantaneously
without budget or fatigue.
The HAL Project
We have over 100,000 parts, 250,000 bill-of-material re- cords, 500,000 attributes
within the master records for parts, cus-
tomers and suppliers, and 3,000,000 trans-
actions to-date. Budgets simply can’t sup-
ply enough staff for enough fingers for
enough keyboards to update the changes,
fix the errors or fill in the omissions in
anything close to real time— if ever. And
if more than a fraction of 1% of these re-
cords is inaccurate, the downstream chaos
and liability issues will prevent further sup-
ply chain automation.
But data maintenance is not sexy. Nobody wants to do it. But even if it were a
priority, keyboards cannot type fast
enough to maintain data quality: there is
simply too much new and changed data to
ever be current. The keyboard is simply the
wrong tool for the job.
But there is a tool that’s right for the job,
and IT already has it. Database languages
such as T-SQL can change thousands of
records in the blink of an eye. If an expert
from supply chain can explain the prob-lem/solution to a programmer as ‘if/then’,
the solution can be programmed as a simple script, often in less than an hour. Taking what an expert would do and converting it to actionable computer language is
called an expert system. We built one and
named it HAL, after the computer in the
movie 200l: A Space Odyssey.
With every error we had to touch, we
converted it into a HAL statement running in the background. The script corrects
“API procurement takes care of the repetitive
ordering so that people take care of strategy, ex-
ceptions, problem solving, and escalations.”