A Free Diagnostic

The Reframe Ladder

Five states of a business operation. Scored on one question: how much of this runs without you?

No email required. Nothing to download. Read it, score one process, and do whatever you want with the answer.

Score one process, not the whole business

Do not ask "where is my company on this ladder." You’ll get a useless average.

Ask it about one process at a time. Your invoicing might be at state 3 while your customer follow-up is still at state 0. That’s normal and it’s the whole point. The gap between your best process and your worst is where the money is.

Pick a process. Score it. Then read the row below it to see what the next step actually costs.

0

State 0

By Hand

Your role

You are the process.

Touches per run

Every step, every time.

What it looks like

The process lives in someone’s head. Nothing is written down. If that person is out sick, it doesn’t happen. Paper, sticky notes, texts, a spreadsheet somebody re-types from another spreadsheet. The knowledge walks out the door at 5pm every day, and permanently when they quit.

Why this is the dangerous one

Every other state on this ladder can have a bad day. This one can have a bad year. When the person who knows leaves ... quits, retires, gets sick, gets hired away by somebody who noticed how much they carry ... the process leaves with them. You don’t rebuild it from documentation, because there isn’t any. You rebuild it from memory and guesswork while the work still has to get done and customers still expect it done right.

That’s the real cost of state 0, and it has nothing to do with AI. Documenting a state-0 process is worth doing even if you never automate a single thing. It’s insurance on knowledge you’re currently storing in one person.

What’s actually blocking you

There’s no process here to improve. There’s only a person to ask. You can’t hand off, automate, or fix what has never been written down. Most "we tried AI and it didn’t work for us" stories start right here, at state 0, with someone buying software to fix a documentation problem.

What moves you up

Get it out of your head. An SOP, not software. The real steps in order, who does what, what "done" looks like. Written is fine. A screen recording of you doing the job while you narrate it is often better and always faster, which matters because the person who knows is usually the person with no time to write. Either way, it now exists outside one person’s head. This is the cheapest rung on the entire ladder and it’s the one everybody skips.

Guardrails

Security

Know who holds which keys, and what walks out with them when someone leaves.

Ethics

Nothing to check yet. Nothing is deciding anything.

Privacy

Inventory what customer data exists and where it lives. The surprise is always the same: more places than you thought.

0 → 1

Next step

Write it down before you buy anything

One page per process. If you can’t write it, you don’t understand it well enough to hand it to anyone, human or machine.

1

State 1

Assisted

Your role

You’re still the process. You just have a faster tool.

Touches per run

Every step, every time. Unchanged from state 0.

What it looks like

Somebody on the team drafts the customer email in ChatGPT, then pastes it into the mail platform by hand. The renewal notice takes 4 minutes instead of 30. Real time saved on each task. Zero tasks removed from the calendar.

The unlock

A thing that took 30 minutes takes 5. That’s real. It also caps out fast.

What’s actually blocking you

This is the trap, and it’s where nearly everyone stops. Your touch count didn’t move. You bought speed, not leverage. The work still needs a person present start to finish, so it still competes for the same hours in the same day. There’s a second problem underneath that one. Adoption is uneven, everyone has their own prompts, nobody’s results match, and none of it is repeatable next month.

If you’ve spent money on AI and can’t point to an hour you got back, you’re at state 1. That’s the diagnosis, and it’s fixable.

What moves you up

Pick one process that repeats on a schedule (daily, weekly, per order, per job) and make it start without a human. A trigger, not a prompt.

Guardrails

Security

Nobody pastes customer lists, card data, or employee records into a public chatbot. Write that rule down this week.

Ethics

Disclose AI-assisted communication where a customer would reasonably want to know.

Privacy

One written rule about what data is allowed to leave the building, and in what form.

1 → 2

Next step

Give one process a trigger

Something other than a person remembering has to start it. That single change is the whole jump.

2

State 2

Delegated

Your role

You start it and you check it.

Touches per run

Two. Kick off, review the output.

What it looks like

The Monday numbers report builds itself off your point-of-sale or accounting system and lands in your inbox at 7am. You read it. You don’t build it. Follow-ups draft themselves per completed job, and you approve twenty in one sitting instead of writing twenty emails.

The unlock

An entire category of work stops competing for your calendar. You review output instead of producing it.

What’s actually blocking you

Reviewing. You’re now the QA department for something that produces faster than you can check. If you don’t build real trust in the output, you’ll quietly re-do the work yourself, and you’ll have gained nothing but a subscription.

What moves you up

A verification step you actually trust. Something that checks the work before you ever see it: a rule, a second pass, a threshold that flags outliers and stays quiet otherwise. At this rung, trust is the product. The automation is already done.

Guardrails

Security

Least privilege. The system gets access to exactly what it needs and nothing more.

Ethics

A human approves anything that reaches a customer or moves money. No exceptions at this rung.

Privacy

Log what the system touched, and set a retention rule on those logs.

2 → 3

Next step

Build the check, then stop reviewing the good runs

Define what "wrong" looks like for this specific process, and make something detect it before you do. Then give yourself permission to stop reading the runs that went fine.

3

State 3

Standing

Your role

You get told when it breaks.

Touches per run

Zero on a good day.

What it looks like

It runs on its own schedule. It checks its own work. It reaches you only by exception, and the exception is specific: "this job is tracking 40% over the quoted hours", not "here’s today’s report." Several processes run at once and you’re not the bottleneck on any of them.

The unlock

Work that used to wait for somebody to find the time now runs continuously in the background. Maintenance stops piling up.

Where the human stays

Zero touches on a good day is not zero touches ever, and it shouldn’t be. Nobody looks at the runs that go fine. But on anything that carries real consequence ... money moving, a commitment made to a customer, something you’d have to unwind ... a person still decides, or reviews what the system decided. That’s a design choice, not an unfinished climb. Volume decides how often somebody sits there. Stakes decide whether they sit there at all.

What’s actually blocking you

Trust and sprawl. You have five or six standing processes now and no single view of what’s running, what it costs, who owns it, or what changed last week. When one drifts, you find out late, usually from a customer.

What moves you up

Monitoring and a written standard. What’s running, what it costs, who owns it, what good looks like. It’s the same discipline as an org chart, applied to processes instead of people.

Guardrails

Security

Nothing standing has more access than you’d give a new hire on day one.

Ethics

A written escalation path, and a named human for every decision that touches a customer, money, or an employee.

Privacy

Retention and deletion actually execute on schedule, and you can answer a "what do you have on me" request inside a day.

3 → 4

Next step

Write the standard down and measure against it

What is each process supposed to produce, at what cost, at what quality bar? You can’t steer what you don’t measure.

4

State 4

Self-Correcting

Your role

You set direction.

Touches per run

Zero. You set the target, not the steps.

What it looks like

The system notices the follow-up sequence’s booking rate slipped, and tests a fix. New processes get built because a gap was detected, not because you scheduled a project for it. You manage outcomes and exceptions across the whole operation rather than any single workflow.

The unlock

The quarter-long project becomes something you kick off and check on.

Where this actually fits, and it’s narrower than it sounds

State 4 belongs to work where being wrong is cheap and recoverable. Which of three drafts goes out. What time it sends. How a queue gets ordered. When a bad call costs you a little money and an afternoon, letting the system test its own fixes beats scheduling a project to do it.

It does not belong to work where being wrong is expensive. Anything that moves real money, commits you to a customer, touches a contract, or would take weeks to unwind stays at 2 or 3 permanently. Not because the technology can’t attempt it. Because you would still have to check it, and a check you have to do anyway isn’t autonomy.

Most operations have some state-4 work in them. Very few have much. If an entire business looks like it belongs up here, that usually means nobody has priced what being wrong costs.

What moves you up

Nothing. This is the top, and the top is a smaller place than it looks.

Guardrails

Security, ethics, privacy

They stop being build decisions and become standing reviews on a calendar. Hard cost controls. And a kill switch you have actually tested, not one you assume works.

The part nobody selling AI will tell you

Most businesses should stop at 2 or 3

State 4 isn’t a goal for a 12-person business. It’s barely a goal for a 200-person one.


The right rung for any given process is the one where climbing to the next costs less than what it gives back. For a lot of operations that’s state 2, and staying there is the correct answer, not a failure.

Some processes should never leave state 2, at any size. Not because the technology can’t do it, but because a person has to own the judgment. Anything an attorney signs. Anything a licensed professional puts their name on. Anything where being wrong costs somebody their money, their health, or their case. In work like that the review step is the work, and automating it away doesn’t make you faster. It makes you liable. There are dozens of businesses in this category and they are not edge cases.

Anyone selling you a climb to state 4 across your whole business is selling implementation, not diagnosis. The diagnosis is the part that tells you which processes deserve the climb, and which ones are fine exactly where they are.

What This Costs You To Move

Rough shape, per process

So the ladder isn’t abstract.


MoveWhat it actually isWhere the effort goes
0 → 1Write the SOPNothing technical. Sitting down and writing what you already know.
1 → 2Give it a trigger, connect two systemsDeciding what starts it, and what should happen when it fires and something’s missing.
2 → 3Build the verification and the alertThe real work. Where most projects die.
3 → 4Measurement, standards, governanceOngoing, not a project. A governance job, not a technical one.

No timelines on this page, on purpose

Anyone who quotes you a schedule before scoping your actual process is guessing. How long one move takes depends on what shape that process is in, what systems it touches, how clean the data is, and who’s available to answer questions. None of that is knowable from a page like this. Scope first, then a real timeline. That order protects you a lot more than it protects whoever’s quoting.

Note where the difficulty sits. It’s not in the AI. It’s in step 2 → 3, building something that checks its own work well enough that you’ll stop looking. That’s a judgment problem, not a software problem, and it’s why tool purchases alone strand people at state 1.

Shape adapted from "Steps of AI Adoption" by Boris Cherny (Anthropic), Jul 16, 2026. That version scores engineering teams by how many coding agents they run. This one scores an operation by how many times a human has to touch it. Different unit, different audience, same idea: you can’t skip rungs.

Scored one and want a second opinion?

Tell me which process and what state you landed on. I'll tell you whether the climb is worth it, and I'll tell you when it isn't.