Catalyst Studio · Operations
The AI Constraint Diagnostic
Find what’s holding your business back, then decide what to do about it. This is the full method in one place: the worksheets, the checklist, and the exact prompts, so you can run it on your own business today.
You don’t need to know what AI is, or what Theory of Constraints or Lean are, to use this. You need information about how your business actually works, and a willingness to let the evidence change your mind about what’s really wrong.
What you’re looking for
We call the thing that’s limiting a business its constraint. It might be a person, a decision, a system, a policy, or a recurring mistake. It’s rarely the loudest complaint. The goal isn’t to find something inefficient. It’s to find the one thing that, if it changed, would actually change what the business can do.
Choose one process
Something with a clear start and end. Not “the whole business.”
Evidence checklist
Use whatever you already have. You don’t need all of it.
- Emails and customer messages
- CRM records
- Project or task data
- Invoices and proposals
- Support tickets
- Meeting notes
- Spreadsheets and reports
- Calendars, where approvals or reviews happen
- SOPs, if they exist — and where reality has drifted from them
The friction finder
Waiting
- Ready to move, but stuck on approval, reply, or availability.
Handoffs
- Work changes hands between people or systems.
Rework
- Something gets corrected, re-sent, or done twice.
Exceptions
- The normal process breaks and everyone improvises.
Bottlenecks
- Work piles up at one person, approval, or system.
Dependency
- Only one person can do it, and everything waits on them.
The constraint test
For every candidate, finish this sentence, out loud or in writing:
“If we fixed this, the business would be able to ______.”
Specific — keep it
- Handle more orders without hiring
- Get proposals out in a day, not four
- Stop the owner reviewing every deliverable
Vague — keep digging
- Be more efficient
- Save some time
- Have a cleaner process
List your candidates
Once the friction finder gives you a shortlist, lay them side by side before picking one.
The root-cause questions
Ask why, repeatedly, until the answer is a rule, a decision, an ownership gap, or a real capacity limit — not just a task. Then test whatever AI gives you:
- What evidence actually supports that?
- What would we expect to see if this were true?
- What else could explain the same pattern?
- What’s missing that would help us be sure?
Intervention options
Not ranked by preference. Start wherever the cause points — that often means ruling out the cheaper options first, but not always.
Before committing to a fix, ask: does this remove the constraint, or does it just make the surrounding inefficiency faster? A faster report doesn’t fix a decision that isn’t getting made.
The measurement worksheet
The master prompt
Copy this into any AI assistant, along with the evidence for one process. It walks the whole method in order and won’t jump to a solution before the diagnosis is done.
I want you to help me diagnose an operational constraint in my business. Your job is not to give generic business advice or recommend an AI solution. Your first job is to understand how the work actually happens. I will give you information about one business process: documents, emails, spreadsheets, CRM data, task records, customer messages, meeting notes, or other evidence. Work through these stages, in order, and do not skip ahead: 1. RECONSTRUCT THE PROCESS Describe what actually happens, start to finish, based only on the evidence I give you. Identify the major steps, who and what systems are involved, where a step depends on another person or system, and where work seems to wait, get handed off, get repeated, or get escalated. Clearly separate facts from assumptions, and say plainly where there isn't enough evidence to know something. 2. IDENTIFY FRICTION Find where work waits, gets stuck, gets repeated, gets handed off, requires escalation, or depends heavily on one person. Don't assume every inefficiency is important. 3. IDENTIFY CANDIDATE CONSTRAINTS For each significant issue, say whether it looks like a symptom, a local inefficiency, or a real constraint on the larger business. Evaluate its likely effect on throughput, revenue, capacity, cash flow, and management attention. 4. CHALLENGE THE DIAGNOSIS Don't simply agree with my assumptions. For each candidate: what evidence supports it, what evidence contradicts it, what else could explain it, what's missing, and what we'd expect to see if it really were the constraint. 5. FIND THE UNDERLYING CAUSE Ask "why" repeatedly to move from the visible problem to its cause. Don't call something a root cause unless the evidence supports it. Separate observed facts, likely causes, hypotheses, and unknowns. 6. NAME THE CONSTRAINT Based on the evidence, identify the issue most likely limiting the larger system, and explain your reasoning. Do not recommend a solution yet. 7. GENERATE INTERVENTION OPTIONS Only now, generate options: eliminating the work, simplifying it, changing ownership, changing the decision rule, improving the information available, and technology or AI. Don't assume AI is the answer. 8. DESIGN A TEST Recommend the smallest practical change that would test whether the intervention actually affects the constraint. Define what we'd change, what we expect to happen, what to measure, how long to test it, and what result would tell us the diagnosis was wrong. Rules throughout: Don't make up missing information. Don't confuse activity with results. Don't confuse an annoying problem with a business constraint. Don't recommend technology just because it's available. It's fine to conclude the evidence is insufficient. Your goal is to find what's actually holding the business back, not to make it sound more sophisticated than it is.
You don’t need to become an AI expert to use this. The point is a better way to look at your business, with AI doing the reading and the reconstruction that used to take weeks of someone’s time by hand.
Catalyst Studio · Operations Notes