Should You Point AI at That?
Every leadership meeting I sit in lately has the same agenda item... AI.
Where can we use it?
What can we automate?
Which vendor?
How fast can we roll it out?
The energy is real, and so is the technology. I'm not here to tell you to slow down on AI. I'm here to ask a different question first.
Where are you pointing it?
Because not everything that can be improved should be improved. And AI makes it easier than ever to improve the wrong thing very quickly.
Necessary, But Not Sufficient
Eli Goldratt saw this coming more than twenty-five years ago. In Necessary but Not Sufficient, written when companies were pouring fortunes into ERP systems and seeing little return, Goldratt made an argument that applies word for word to AI today.
Technology can only bring benefits if it removes a limitation.
And removing the limitation is not enough. You also have to change the rules you created to live with that limitation.
He framed it as six questions:
What is the power of the technology?
What limitation does it remove?
What rules, policies and habits did we build to accommodate that limitation?
What new rules should we use now?
In light of the new rules, what changes are needed in the technology?
How do we cause the change?
Most AI initiatives answer the first question beautifully. Demos are impressive. Pilots look great.
Very few answer the second and third. And without those two, the technology gets layered on top of the old rules, and the business keeps behaving exactly as before. Just with more expensive software.
Tesla and the Robots
Here's a case worth studying, because it involves one of the most technology-forward manufacturers in the world.
In 2017 and 2018, Tesla was ramping production of the Model 3. The plan leaned heavily on automation, including in final assembly, with a complex network of conveyors and robots.
The ramp stalled.
Tesla missed its 2,500-per-week target in the first quarter of 2018, far short of the 5,000 per week it had promised. The company paused production in April to address bottlenecks.
That same month, Elon Musk said publicly that excessive automation at Tesla had been a mistake. His mistake, in his words. And he added three words that stuck:
"Humans are underrated."
Then what did they do? Tesla ripped out much of the conveyor system and, with no room left in the factory, built a second general assembly line inside a giant tent in a matter of weeks.
In the last week of June 2018, Tesla produced 5,031 Model 3s. About 20 percent of them came off the tent line.
Here's the part that's easy to miss. Tesla didn't abandon automation. According to its own filings, it kept a high level of automation in stamping, welding, paint, powertrain and battery assembly.
Automation was never (and it won't be) the problem. Automation applied without asking what limitation it removed, and where the constraint would land as a result, was the problem.
How to Know You're Pointing It at the Wrong Thing
Here's something worth paying attention to…
Take any AI or automation initiative on your list and ask three questions:
Does this touch the constraint? If it speeds up a non-constraint, the best case is that nothing changes. The worst case is that work piles up in front of the constraint faster than before.
What limitation does it remove, and who feels it? If the answer is "it saves us time," ask whose time, and whether that time was holding back throughput.
What rule are we going to change? If you can't name a policy, measure or habit that will be different after the technology goes live, the business will keep working the old way.
If you can answer all three clearly, you may have a high-leverage initiative.
If you can't, you have a pilot looking for a purpose.
Stay with me now! Here's the encouraging bit. AI pointed at the constraint can be one of the most powerful exploit tools you've ever had.
Better scheduling of the constraint.
Faster quoting when sales is the bottleneck.
Making sure the constraint never waits on information.
Same technology, very different results, depending on where it's aimed.
Your October Challenge
This month, pick one AI or automation initiative in your organization, either underway or proposed. Run it through Goldratt's second and third questions.
What limitation does it remove?
What rules did we build to live with that limitation, and are we prepared to change them?
If the answers are clear, push it forward.
If they aren't, pause it long enough to find them.
One last thing Before I go! AME International Conference: Milwaukee
The AME International Conference is coming up this month, October 26-29 in Milwaukee, and I'd love to see you there.
This year's theme is People First, Progress Together.
And the AI conversation fits right in. Technology works best when it's aimed at helping people focus on what moves the business.
I'll be part of three sessions:
Monday, October 26, 1:00 to 5:00 p.m. A half-day TOC Flow Solutions workshop, a condensed version of our certificate program, with hands-on simulations, the Four Pillars and the Five Focusing Steps.
Wednesday, October 28. An Idea Exchange on using the Theory of Constraints in industry. Bring your questions and your toughest scheduling problem.
Thursday, October 29, 10:45 a.m. Scott Crosley, President of ACT Test Panels, and I will share how ACT achieved double-digit growth with the same people and the same equipment.
Learn more and register: https://www.ame.org/conference/ame-milwaukee-2026-international-conference
And if you're in the Milwaukee area, the full TOC Flow Solutions Workshop with AME follows on November 12 and December 10.
Comments