Nine things break, and they break in order

On a first call, an owner can almost always name his constraint. He’ll say the reports take three weeks and he still doesn’t trust them, or that the team ran an AI pilot and got nothing out of it, or that he knows he should be looking at a weekly scorecard and isn’t. He’s not wrong about any of that. He’s describing something that’s true about his business.

He’s also almost never describing the thing to fix.

There are nine ways the information in a business breaks, they sit in three tiers, and they break in a fixed order. The ones you can name are the ones at the top, because those are the only ones with numbers attached. The ones causing them sit underneath, where nothing is measured and so nothing looks broken.

Here’s the whole map, bottom to top.

The nine constraints

Capture: does the data get born clean?

Everybody wants to talk about connecting AI to their systems. If what’s in those systems is garbage, it doesn’t matter what you connect.

Never recorded is the floor. You can’t measure your win rate on quotes if nobody goes back into the CRM and marks whether the opportunity was won. In a smaller shop it’s often time: are you recording how long people worked, and crucially, what they worked on? That’s the only way anyone ever answers how many hours went into a job.

No shared definition is the one people skip past. You’re recording that a lead came in, but you never record whether it was qualified, and underneath that, nobody ever agreed what qualified means. Order value is the same trap. Sometimes it’s the discounted number, sometimes it’s gross. Sometimes it includes tax and freight, sometimes not. Put two systems side by side where one includes freight and one doesn’t, try to add them up, and you don’t have a total. You have two different questions being answered at once.

Error-prone entry is what happens when your CRM is only as current as whichever salesperson felt like typing up their call notes. Often they don’t. Sometimes they do and miss the standardized field. Dropdowns beat free text for exactly this reason, and anything keyed by hand is going to carry a rate of wrong.

Connect: does the data reach the other data?

Tier 2 is for a business where the back-end systems are individually fine and none of them talk.

Dead handoffs are where a piece of work jumps systems and a human carries it across. You write the quote in a quoting tool, somebody re-keys it into the ERP to run the job. That’s error-prone on its face, and it also just doesn’t happen sometimes.

Shadow spreadsheets are when you’ve bought and implemented the system and your people are still doing the work in Excel next to it. They’re using a spreadsheet to produce something the tool was supposed to produce.

One person is the system is the salesperson with the great close rate whose outreach nobody can describe, because it isn’t in a system and there’s no SOP. Or the one person who knows all the edge cases about which quotes get entered which way. It works, right up until it doesn’t.

Decide: does anybody act on it?

Hand-built and too-late answers is month end taking three weeks. Somebody pulls revenue from one system, project completion from another, and hours from a time tracker, and assembles them by hand. By the time the number reaches you, the moment it described is gone.

No cadence is having the thing stitched together, in real time, in a portal nobody opens. There’s no standing meeting where it gets read and decided on.

Starving tools is your BI, your automation and your AI, bought and paid for and connected to nothing.

Why does the order matter?

Because a fix at the top inherits every problem underneath it, and it inherits them silently.

Take the most common version right now. Somebody wants to run an AI pilot: connect Claude to the CRM and the email system so it can draft replies that respect where the client relationship stands. Good idea. But if nobody fixed error-prone entry, and the CRM is full of stale and half-entered records, you’ll succeed at every visible step. The connection works. The tool reads the system. The drafts come out. They’re just built on records that don’t reflect reality, and now you’re producing wrong answers faster and with more confidence than before.

That’s the pattern. Tier 3 failures are loud and Tier 1 failures are invisible, which is exactly backwards from how urgent they are.

What the cascade looks like in practice

Here is a walk down the map, starting from the thing an owner would say out loud.

He says there’s no cadence. “I know we should be assessing our KPIs on a weekly scorecard, and we’re not. We look them up manually, or once a month, or once a quarter when we get to it.” What he wants is the data in real time. Constraint 8, clearly.

Ask why it doesn’t happen and you get constraint 7. The reports are hand-built, so they only get pulled when somebody has a free afternoon.

Ask why they’re hand-built and you’re at constraint 4. The systems don’t talk. Any KPI that needs numbers from two places has to be assembled by a person.

So say you want the true cost of a job, labor plus materials. Labor lives in the time tracker, materials live in inventory, and there’s no connection between them. Dead handoff, fix the handoff.

Except go look, and the time entries were only ever recorded at the employee level, not against a project. Now you’re at constraint 1, never recorded, and there’s nothing to connect. Or one step sideways at constraint 2, where time does get logged to projects but nobody ever defined what counts as project time versus general labor, so two people are logging two different things into the same field.

The owner came in asking for a real-time scorecard. The work is four layers down, and no amount of dashboard building was ever going to surface it.

The same thing happens with the AI version. “We ran a few pilots and got nothing out of them” is constraint 9. You don’t fix constraint 9 by running a better pilot.

Where does “one person is the system” belong?

Constraint 6 is the one I’d argue about, and I think the placement is open. Here’s both sides.

The case for Tier 1. Think about a salesperson who never enters anything in the CRM and keeps the relationships in his head, or a founder doing the same. Nothing is being collected at all. That’s not a connection problem, that’s constraint 1 wearing a different hat, and it belongs on the floor with the rest of the capture failures.

The case for Tier 2. Now think about somebody who’s in three systems every day and produces good insight from the combination. He sees the pipeline in the CRM and the monthly revenue in QuickBooks, he understands how one feeds the other, and he reports that read verbally at the Monday leadership meeting. He’s not failing to record anything. He’s doing the connecting himself, in his head, because no system does it. That’s Tier 2, and the fix is to make that report run without him.

So the honest answer is that it’s two different failures sharing one name, and which tier it lands in depends on whether the person is the one capturing or the one connecting. That distinction matters, because it changes the fix.

Where to start

At the bottom, which is rarely where the pain is.

That’s an uncomfortable thing to be told when what’s bothering you is that month end takes three weeks. But the reason it takes three weeks sits below the reporting layer, and every hour spent building a better report on top of unrecorded or undefined data is an hour spent making the wrong number arrive faster.

The useful thing about the order is that it turns an overwhelming problem into a sequenced one. You don’t have nine problems. You have one binding constraint, it’s lower than you think, and once it clears, the next one is already identified.