Faster This, Faster That, Faster Bottlenecks: Why AI Makes TOC More Important, Not Less

In my previous article, “AI & Cybersecurity in 2026: Should we Worry?”, I explored the mounting challenges businesses face as they rapidly adopt Artificial Intelligence (AI). I specifically examined the dual nature of AI in the security sector, weighing whether its integration serves as a powerful defense or a dangerous new liability.

In this article, I shift from defense to optimization by exploring the Theory of Constraints (TOC)—a management philosophy developed by Eliyahu M. Goldratt. TOC helps managers identify the single “weakest link” holding back their entire system. A well-known adage says,

No chain can ever be stronger than its weakest link.

TOC is built on the idea that every system must have a constraint or bottleneck that limits its output. We know this because no system has infinite output and no company has infinite sales or profit.

It was during my MBA years that I became aware of TOC. Like most graduate students, I spent a significant amount of time absorbing and regurgitating a dense “alphabet soup” of management theories. From Lean and Six Sigma to Activity-Based Costing (ABC), we were taught to seek out efficiency with surgical precision.

Back then, TOC seemed like just another academic requirement, easily overshadowed by newer, flashier frameworks. Fast forward to the AI revolution, and I’m looking at Goldratt’s “weakest link” with fresh eyes. It’s a striking paradox: as our tools become more sophisticated, our success depends less on the power of the technology and more on our ability to pinpoint exactly where the system is failing.

Goldratt was quite clear on his stance concerning technology: technology is "necessary but not sufficient" to create value. During one of his many taped lectures, he used the rise of MRP (Material Requirements Planning) systems as a cautionary tale when implementing any new technology—including AI.

MRP and a Cautionary Tale

Before MRP usage became mainstream in the mid-to-late 70s, factories struggled with slow data processing, making it nearly impossible to manually manage parts and orders. If you were a factory and manufacturing a product, it would be nearly impossible to calculate the exact timing of thousands of parts across hundreds of orders manually. If you were a production manager, your “rule” was simple:

• "Order everything early."

• "Keep large safety stocks of everything."

• "Build in huge lead times just in case."

The unfortunate downside to this “rule” was costly inventory (i.e., high levels of WIP and finished goods).

Now comes the era of MRP, and production thought it had finally found a solution to its slow data limitation problem. The MRP system could calculate, with mathematical precision, exactly when every “nut and bolt” needed to arrive to meet the shipping date. The hype was substantial, yet the actual returns for a majority of the plants proved underwhelming, if not negative. Key reasons for MRP’s underperformance are outlined below in the “Lessons Learned” segment.

Lessons Learned

• Data Integrity trap: The old adage “Garbage In, Garbage Out” is based on the principle that the “quality of an output is strictly determined by the quality of the input.” If the foundational data is flawed (e.g., lead time or inventory levels), even the most sophisticated "engine" (be it an MRP system or a Generative AI) will simply produce high-speed, high-fidelity errors.

• Local Optimization: MRP systems were meant to keep workers 100% busy. This led to the "trap" of making a single department, person, or machine work at maximum speed without considering the overall goal of the factory. The result would be excessive amounts of inventory, choking the plant floor.

• Ignoring Reality: Like any system, MRP had to rely on certain assumptions (e.g., “fixed” lead times, “due dates”); however, “Murphy’s Law” brings us back to reality—machines will break down, workers will get sick, assumptions are almost always wrong, and so on.

• Know the Capacity of the Bottleneck: An MRP system would often release work based on given assumptions, such as "due dates," rather than the capacity of the bottleneck. Goldratt stressed the importance of “releasing work into the system at the speed of the bottleneck.”

The shortcomings of MRP provide a somewhat historical mirror for the challenges businesses face with AI today, one of those being AI’s “faster bottleneck” dilemma.

AI’s Faster Bottleneck dilemma

When Goldratt wrote The Goal, bottlenecks were usually visible: a slow machine on a factory floor or a pile of physical inventory. More often than not, the bottleneck was obvious, and it was just a matter of making “the throughput of the system equivalent to the throughput of the bottleneck or constraint.”

In Goldratt’s The Goal, the protagonist, Alex, is asked a simple question: “What is your company's ultimate Goal?” With time, Alex comes to the realization that the “goal of his company” is to make money. How you make more money is by “identifying your constraints and focusing on improving them.”

Does AI, being the “new tech kid on the block,” alter the way you need to assess your firm’s constraints and bottlenecks? The answer is a resounding NO.

The AI revolution that’s taking place now is about speed—speed of collecting and processing data. It’s not just about the existence of data, but the unprecedented velocity at which data is now captured and utilized. This “speed” is transforming AI from a retrospective tool into a dynamic strategic partner.

While AI’s speed seems like a win-win, where everyone is working faster, your company as a whole is not reaching its goals any more quickly because the “true constraints” (human judgment and systemic alignment) remain unchanged. You’re simply creating a bunch of “non-constraint” bottlenecks that, in the end, only waste resources and cost your firm money.

In the context of Goldratt’s TOC, the “faster bottleneck” dilemma describes the unintended consequence of accelerating non-constraints. In the scenario below, we describe a common problem that arises between marketing and sales—marketing complaining that sales are not following up on leads, and sales complaining that marketing’s leads are of “low quality.”

Goldratt’s “Five Focusing Steps”

This dilemma isn’t limited to marketing and sales. Let’s consider another context: CPA firms confronting similar “faster bottleneck” issues as junior accountants use AI to create work faster than it can be reviewed.

The main challenge is getting “everyone on the same page,” where the overriding goal is to increase throughput by efficiently turning client data into finished billable products.

Goldratt’s “Five Focusing Steps” can help address this.

1.  Identify the Constraint

In most CPA firms, the bottleneck isn’t doing the returns, it’s high-level review and sign-off.

• The Trap: If you give AI to junior accountants to "draft" returns faster, you have not helped the firm. You have simply increased the "Work-in-Process" (WIP) pile sitting on the Senior Manager’s or Partner’s desk.

• Action: Look at your practice management software. Where do files sit the longest? Usually, it’s the "Pending Manager Review" stage. If so, that’s your constraint.

2. Exploit the Constraint

Before buying expensive new AI tools, ensure your existing constraint is working at 100% capacity on high-value tasks.

• AI Application: Use basic AI (like Xero’s AI or HubSpot AI) to handle the "noise" that distracts your bottleneck.

• Action: The Senior Manager or Partner should only see “clean” files where the AI has already verified the basics, allowing the reviewer to focus strictly on high-level advisory and risk.

3.  Subordinate Everything Else

This is the hardest step for the firm. You must limit AI-driven production in non-bottleneck areas to match the pace of the review stage.

• The Rule: If a Senior Manager or Partner can only review 10 returns a day, tell the staff to stop using AI to “finish” 30.

• Action: Use AI to "buffer" the bottleneck. Automate client follow-ups via TaxDome or similar portals so that by the time the reviewer is ready, all necessary documents are already verified and categorized.

4.  Elevate the Constraint

If the bottleneck still limits your firm from achieving its goals, invest in major AI solutions to "break" the constraint.

• AI Application: Instead of hiring another expensive Senior Manager or taking on another Partner, implement advanced generative AI or specialized tax LLMs (like TaxGPT) that can perform a "First Pass" review or draft complex advisory memos.

5.  Repeat

Once you solve the issue of one bottleneck, another one is going to pop up, so always be prepared.

• The AI Era Shift: Senior Managers or Partners might find that they can now review work so fast that they’re suddenly waiting on client documentation (the external constraint).

Image generated by Chat-GPT4

Final Word

Don’t just automate—Synchronize  

When Goldratt conceptualized TOC some 40 years ago, his original audience was production and manufacturing. Back then, when talking about bottlenecks, they usually had to do with production capacity (physical flow). They were the “slow machine” on the factory floor or a pile of physical inventory. You could see the bottleneck, and once you saw it, you could then use TOC to synchronize your production so the “throughput of the system was equivalent to the throughput of the constraint.”

The illustrations above (i.e., marketing vs sales) showed that the problem today is not enough output, but over-activating non-bottlenecks. Goldratt warned against this, which is why “adding AI to a non-bottleneck is like putting a Ferrari engine in a car that has no tires—you’re burning a lot of fuel, making a lot of noise, but you aren’t moving any faster” (this quote was said by David Heinemeier Hansson during a Lex Fridman podcast).

About Carl Burch

Carl Burch holds an MBA, CMA, CIA, FCCA, and is a QuickBooks ProAdvisor. He is also the co-founder of BURCH Business Services (BBS) located in Boston, MA.

You can contact Carl at carl.burch@burchbusinesservices.com, or for more information on BBS, visit www.burchbusinesservices.com