Join AI Insiders
Back to Resources

How Do I Find the Real Bottleneck in My Business When Everything Feels Like a Problem?

How Do I Find the Real Bottleneck in My Business When Everything Feels Like a Problem?

You do it by asking what is actually scarce right now instead of what used to be scarce, and this article walks through how I found out I had been fixing the wrong thing for months.

I Spent Months Making Something Faster That Was Already Fast Enough

I was proud of the work. That is the part that stings.

For most of the spring I poured my mornings into making our content and output better and faster. Sharper prompts. Cleaner systems. Tighter drafts. I told myself this was the highest leverage thing I could touch, because two years ago it absolutely was. Then I sat down one evening, looked at what had actually shipped, and realized the drafts had never been the problem. They were piling up. Waiting. The constraint had quietly moved to something much less fun to talk about, which is the number of hours a human being on my team had available to review, approve, and actually put things into the world.

I had been sharpening a knife that was already sharp while the cutting board sat empty.

Here is the direct answer, because you came here for one. The real bottleneck in your business is the single step where work waits the longest before moving forward. Not the step that feels hardest. Not the step you enjoy improving. The step with a queue in front of it. Find where things sit and stack up, and you have found your constraint. Everything upstream of it can only run as fast as that step allows, which means every improvement you make anywhere else is invisible.

And the harder truth underneath that: the constraint moves. It is not a permanent feature of your business. What was scarce in your operation eighteen months ago is very likely not scarce today, especially if you have been using AI in any serious way. Most of us are still optimizing against a map of a constraint we already solved.

The biggest companies on earth are learning this same lesson right now, in public, with numbers so large they are hard to read. I will get to those. But I want to be honest first that I did not learn this from a case study. I learned it from wasting a season.

Key Takeaways

  • Your real bottleneck is the step where work waits the longest, not the step that feels the most difficult or the most interesting to improve.
  • Constraints move over time, so a system you optimized last year may be pointed at a problem that no longer exists.
  • In 2026 the frontier AI labs are leasing compute from direct competitors because raw capacity, not model quality, became the scarce thing.
  • For most small businesses the scarce resource is now implementation capacity, meaning the human hours to review, decide, and ship.
  • Re-checking your constraint on a schedule is more valuable than any single optimization you will make this quarter.

Why We Keep Optimizing the Thing That Already Got Cheap

We optimize what we know how to optimize. That is the whole trap, and it is not a character flaw.

When you spend a year getting good at something, that skill becomes the tool you reach for. You built the prompt library. You learned the tool. You developed real judgment about what good output looks like. Then the ground shifts underneath you, and instead of asking a fresh question you keep applying the skill you already have, because applying it feels like progress. It produces artifacts. It produces a feeling of momentum. And momentum in the wrong direction is very hard to detect from the inside.

There is a second reason, and this one is less flattering. The real constraint is usually boring. In my case the answer was not a better model or a smarter workflow. It was that a human had to read things and make decisions, and that human had a finite number of hours and a family and a life. There is no clever hack there. There is only capacity, delegation, and hard prioritization. So I avoided it. Not consciously. I just kept finding more interesting work to do a few steps upstream.

I think a lot of us do this. We reach for the fixable problem instead of the important one. My wife would tell you I do it at home too, that I will happily reorganize a garage shelf while the actual project waits. She would be right.

Here is the reframe that changed how I look at my business.

Stop asking what could be better. Almost everything could be better, so that question gives you no signal at all. Start asking what is scarce. Scarcity is specific. Scarcity has a queue in front of it. Scarcity is the thing you would buy more of instantly if you could, and the thing everything else is waiting on.

Once I framed it that way, my answer changed within about ten minutes of honest thinking. Ideas were not scarce. Drafts were not scarce. AI made both of those abundant, and abundant things do not deserve your best hours. What was scarce was the decision-making and shipping capacity at the end of the line. Everything I had been building was pushing more volume into a funnel that was already backed up.

That is not a strategy failure. It is a measurement failure. I was measuring the wrong end of the pipe.

Even the Biggest Companies Are Chasing Moving Constraints

This is not just a small business problem. Watch what the largest technology companies in the world are doing right now, because it is the same pattern at a scale that makes it impossible to miss.

Competitors are renting from competitors, which only happens under real scarcity. In July 2026, CNBC and other outlets reported that Anthropic is in early talks to lease computing power from Meta in a deal potentially worth around $10 billion over two years. Anthropic reportedly initiated the proposal. That is a leading AI lab going to a direct rival for capacity. It is not the only one. Anthropic reached a separate arrangement with SpaceX earlier in the year reported at roughly $45 billion over three years. Companies do not rent from their competition when the thing they need is easy to get. They do it when the thing they need is the only thing that matters.

The buildout numbers have stopped making intuitive sense. Meta expanded its Hyperion data center project in Richland Parish, Louisiana to a five gigawatt target with investment pushing past $50 billion, inside a broader infrastructure commitment the company announced in November 2025. Meta is forecast to spend up to roughly $145 billion in capital expenditures in 2026, and in March 2026 signed a five-year agreement with Nebius valued at approximately $27 billion. Across Google, Microsoft, Meta, and Amazon, analysts have put 2026 capital expenditure at roughly $725 billion combined.

The bottleneck moved from chips to the components nobody was watching. In April 2026, Microsoft guided 2026 capital spending to roughly $190 billion. CFO Amy Hood attributed about $25 billion of that figure to higher component pricing, driven largely by memory and storage. TrendForce data shows why. Conventional DRAM contract prices were revised upward to increases of 90 to 95 percent for the first quarter of 2026, the largest quarterly jump on record, and when combined with the prior quarter the cumulative increase approached triple the earlier price. Hood also warned Microsoft expects to stay capacity-constrained on GPUs, CPUs, and storage. Think about that. The bottleneck was model quality, then it was GPUs, and now a meaningful slice of it is memory chips. It moved twice in about two years.

Outside of tech, the same discipline produces the same results. Hospitals are a good example because the stakes are human. Operations researcher Eugene Litvak has spent decades arguing that hospital overcrowding is not primarily caused by unpredictable emergency patients but by artificial variability in scheduled elective admissions, which hospitals control. Hospitals that smoothed elective admissions across the week rather than clustering them have reported safer nurse-to-patient ratios, fewer errors, faster access to surgery, and millions in additional annual revenue, according to reporting in STAT in March 2026. The constraint was never the number of beds. It was the scheduling pattern in front of the beds.

Chick-fil-A did something similar with a much simpler problem. Their drive-thrus regularly serve over one hundred cars an hour, and the natural assumption is that the kitchen is the limit. Their answer was to send team members out into the lane with tablets to take orders face to face, so the kitchen receives orders earlier and starts cooking while cars are still moving. They did not first make the kitchen faster. They moved the order intake upstream of the queue.

And for the rest of us, the constraint is implementation. MIT’s Project NANDA report, “The GenAI Divide: State of AI in Business 2025,” found that roughly 95 percent of enterprise generative AI pilots delivered no measurable financial return. The researchers pointed at what they called the learning gap, which is the failure to integrate AI into actual workflows, structures, and daily operations. Their conclusion was blunt. The divide was not driven by model quality. It was driven by approach. Meanwhile, survey work from Time Etc found the average entrepreneur spends more than a third of the work week on small administrative tasks like invoicing, data entry, and scheduling.

Put those two findings side by side and you get the picture of most small businesses in 2026. The models are good. The tools are cheap. The owner has no hours left.

How I Now Find My Constraint, and Why I Re-Check It

I run a much simpler process now, and I run it more often than I used to.

The first thing I do is look for the pile. Not the pain, the pile. Pain is emotional and it lies. Piles are physical and they do not. When I looked honestly at my own operation, I found approved-but-unpublished content, half-finished offers, and messages sitting unanswered. That pile was pointing directly at the constraint, and it had been visible for weeks. I simply had not been looking at it as data.

The second thing I do is ask what I would buy more of if money were not a factor. Not what I would fix. What I would buy. That question cuts through a lot of noise, because it forces you to name a resource rather than a complaint. My honest answer was not more AI capability or more traffic. It was more skilled human review time. That is a very different business decision than the one I had been making all spring.

The third thing I do, and this is the one that took real humility, is I ask what already got cheap. This is the part almost nobody does. AI has made a specific list of things nearly free in the last two years. Drafting. Research summarization. First-pass analysis. Variation and volume. If you built systems and habits when those things were expensive, and you have not revisited them since, you are almost certainly still spending your best attention on the wrong end. I was. I had a whole morning routine built around producing more of the thing I already had too much of.

The fourth thing, and the one that actually changed my calendar, is that I put a recurring reminder on my schedule. Every quarter I sit down and ask those three questions again from scratch, without looking at last quarter’s answer first. The constraint moves. Microsoft’s moved from compute to memory in a matter of months. Mine will move again, and when it does, my current systems will quietly become the wrong systems. The only defense is scheduled re-checking, because you will not notice the shift in real time. You will only notice that you are working hard and things are not moving.

What I did after that evening was less dramatic than you might expect. I did not tear anything down. I stopped adding volume, I moved review capacity to the front of the week instead of the end, and I made peace with publishing less while shipping more. Ideas are not the scarce thing anymore. Finished work is.

I will tell you the honest emotional part. It felt like admitting the last few months were wasted. They were not entirely wasted, because the systems still work and I will use them. But I did burn real time being productive in a direction that did not matter, and I would rather tell you that plainly than pretend I have always been disciplined about this.

Practical Steps to Find What Is Actually Scarce

  1. Map the path of one unit of work from start to finish. Take one customer, one project, or one piece of content. Write down every step it passes through from first contact to done. Do not write down how you wish it worked. Write down how it actually worked last week.

  2. Mark where the work sits and waits. For each step, note roughly how long the work is actively being worked on versus how long it sits in a queue waiting for someone or something. The step with the longest wait in front of it is your constraint. This is usually obvious once you write it down, and usually invisible until you do.

  3. Ask what you would buy more of tomorrow. Name a resource, not a feeling. More review hours. More qualified conversations. More fulfillment capacity. If your answer is vague, you have not found the constraint yet.

  4. List what has gotten cheap in the last eighteen months. Be specific and be honest. For most of us the list includes drafting, research, summarizing, ideation, and first drafts of almost anything. Then look at how much of your best attention is still allocated to those things.

  5. Move one resource, do not add one. The instinct is to add tools, add spend, add volume. Resist it for one cycle. Try shifting existing time or people toward the constrained step first, the way a hospital smooths its schedule rather than building more beds.

  6. Protect the constraint from interruption. Whatever step is your bottleneck should be the one thing that never sits idle and never gets pulled into low-value work. Everything else in your business should be arranged to keep that step fed and moving.

  7. Put a quarterly re-check on your calendar right now. Not a vague intention. An actual recurring appointment. When the constraint moves, and it will, this reminder is the only thing that will catch it.

Frequently Asked Questions

What is the difference between a bottleneck and just being busy?
Being busy means everyone has work to do. A bottleneck means work is waiting in a queue at one specific step while other steps sit idle or produce output nobody can use. Busy is a feeling. A bottleneck leaves physical evidence, usually a backlog you can point at and count.

How often does the real constraint in a business actually change?
More often than most owners expect. Microsoft’s constraint shifted from compute capacity to memory and storage pricing within roughly two years, adding about $25 billion to its 2026 capital plan. For small businesses, any significant change in tools, staffing, or demand can move the constraint within a single quarter.

Can AI tools fix my bottleneck for me?
Sometimes, but only if the bottleneck happens to be a task AI is good at. MIT’s 2025 research found roughly 95 percent of enterprise AI pilots produced no measurable return, largely because the tools were not integrated into real workflows. AI applied away from your constraint produces more output and no more results.

What if my bottleneck is me, the owner?
That is the most common answer and the hardest to accept. Survey data suggests entrepreneurs spend more than a third of their week on small administrative tasks. If decisions route through you, start by removing low-value work from your plate before adding anything new to the system.

Should I fix the bottleneck or work around it?
Try working around it first, because it is faster and cheaper. Chick-fil-A did not rebuild every kitchen. They moved order taking out into the drive-thru lane so the kitchen received orders sooner. Rerouting or resequencing work often relieves a constraint without any new spending at all.

The Thing I Wish I Had Asked Sooner

I keep thinking about that evening, looking at all the work I had produced and realizing almost none of it had reached anybody.

There was nothing wrong with the work. That is what made it so easy to miss. Good work aimed at the wrong constraint still looks like good work, right up until you count what actually shipped. If you are reading this and something in your chest just tightened a little, sit with that. You probably already know where your pile is. Most of us do. We just have not let ourselves look straight at it, because the answer is usually less interesting than the thing we would rather be working on.

So go look. Not at what is broken. At what is waiting. Find the place in your business where work sits still, and give that place your best hours for the next month. Everything you have been optimizing will still be there, and it will finally matter, because the output will actually reach the other side.

I am going to keep writing about this as I work through it, including the parts I get wrong, because I would rather be useful than impressive. If this landed, follow along and tell me what your constraint turned out to be. I read the replies, and honestly, other people’s answers to this question have taught me more than any framework I have read. Connect with me and let me know what you found.

The thing you are best at improving is almost never the thing holding you back.

About the author

Jonathan Mast is the founder of White Beard Strategies. He is an entrepreneur and speaker, a husband and a father, and he writes about AI, business, and the honest parts of building. He is more interested in what actually works than in what sounds good on stage.

Sources

Put This to Work With Us.

AI Insiders gives you practical training, direct coaching, and a community of entrepreneurs who are applying AI to real businesses.

Join AI Insiders