Always Split the Difference
In the Early 1980s, when computers were crude plastic boxes and the Internet didn't even exist, Intel had a problem.
They were suffering from terrible quality issues.
Suddenly instead of a trickle of defects, half of the millions of microprocessors they were building were failing.
The problems were caused by dust.
When you are printing microscopic metallic traces onto silicone wafers, a dust particle ten times wider than a trace will break everything.
Semiconductors are manufactured in huge clean rooms, with filtered air and where technicians wear surgical masks, rubber gloves and special suits.
Intel spent weeks investigating every room and every process, but they just couldn’t see how dust was entering the process.
So they shifted the blame.
It must be the wafers themselves and they were built by Monsanto so they handed off the problem.
Monsanto then spent months and millions of dollars improving their manufacturing process, making huge improvements.
But the improvements didn’t stop the problems at Intel.
The interesting thing was that Monsanto’s other customers like IBM were not having the same problem.
If both Intel’s and Monsanto's processes were clean there had to be a problem in the middle.
So Monsanto hired a private investigator who followed the next shipment of wafers on their route to Intel.
As the fully sealed and dust proof boxes of fresh wafers reached Intel, the PI watched as a diligent shipping clerk opened every single box in his dusty office, and counted out each wafer by hand on his very dusty desk, to ensure Intel had received every single wafer they had ordered, before he boxed them back up and moved them into the Intel facility.
Every system has pinch points where just one small process or a single person can distrust everything. Even a $1 billion company can be taken down by a single small link in the chain.
It took months of work and cost millions to find this flaw in the system.
Half-Split the Difference
If you are facing an issue in a system of any kind one of the most efficient troubleshooting techniques is the Half-Split Method.
You map out the full system that is failing and find the half-way point.
Run a test there to see if you still have the issue and if you do (or don’t) you’ve now isolated the problem in just half the system. Then you split the isolated issue in half and test again.
Repeat this until you find the problem.
It’s the fastest way by far.
The Science Bit: If you have a 100-step system, you don't check all 100 steps. Every time you split the process, you divide the remaining possibilities by 2, following a base-2 logarithm, which equals approximately 6.64. Rounded up, this means you will never need more than 7 tests to find the broken step.
The Intel/Monsanto dust issue was roughly half way through the process, but time was wasted on either side looking at all parts of the system. Intel should have checked the dust on the wafers as they received them.
Monsanto could have checked dust levels of the wafers they were shipping, avoiding months of wasted effort.
And if anyone had looked at the whole system, not just their part in it, they would have realised that they were missing the vital step of the dusty-desked shipping clerk.
The half-split method is ironically a well used concept in computer science.
It’s called a binary search algorithm and is one of the fundamental tools used to search massive datasets at high speeds.
Whether you are debugging software code, managing a supply chain issue, or trying to figure out why your car won’t start, the principle remains the same.
Stop searching step-by-step.
Start splitting the difference.
