Count your own workflow

This is the procedure behind every number in this record, and it is the answer to "how much of my work can be automated". It is domain-free.

Whatever your industry, the procedure is the same. A dental clinic booking patients, a construction firm quoting site work, a law firm drafting contracts, a bakery taking wholesale orders, an accountancy practice closing a month, a logistics company booking loads, a school enrolling students, a marketing agency running campaigns, a repair shop handling jobs, a website studio shipping pages. None of these needs a special version of the method. Every business runs work through steps; the exercise is to write the steps down and tag them.

Twenty minutes with a pen is enough for a first pass on your own business.

Step 1: write the run down, one line per step

Take one workflow that repeats. Not a department, not a job title: one run, from the moment work arrives to the moment it is delivered. Write each step as a short line, in order, as it actually happens today rather than as the process document claims.

A run usually has between eight and fifteen steps. Fewer than six means the steps are too coarse to tag; more than twenty means two workflows got mixed.

Count only the steps your side performs. The customer choosing a slot, the client approving a design, the patient turning up: those are things that happen to your run, not work you could ever automate, and putting them in the list makes the total bigger and your number smaller for no reason. Write them in the margin if they help you follow the story, and leave them out of the count.

If the run genuinely splits, count it twice. A booking that arrives by phone and one that arrives by chat are two runs when the steps differ, and averaging them by hand hides which one is worth fixing first. Count each, then weight them by how often each happens.

Step 2: tag every step with one of three labels

The same three tags, in businesses that have nothing to do with software:

TYPING is the point of the exercise. It looks like work, it fills a day, and it is the part that automation removes. DECISION steps stay whatever the technology does.

Step 3: count, and there are two numbers

A run has a before and an after, so one percentage cannot describe it. Report both, by these names, and never merge them into a single figure:

today    = AI ÷ total steps            what the system already runs
could be = (AI + TYPING) ÷ total steps what it could run once TYPING is removed

today is usually 0% for a business that has automated nothing, and saying so out loud is the honest starting point rather than an embarrassment.

could be is the answer to "how much of this can be automated". TYPING sits on the human side today and moves to the AI side once the work is done, which is exactly where the improvement comes from. DECISION steps are in neither number's numerator: they stay with a person however far the work goes.

The worked example below is the same run at both ends of that journey: today 0%, could be 70% before it was built, and today 70% after. Quoting the second number on its own is fair; quoting it as "70% automated" while nothing is automated yet is not.

Write both numbers with the date and the name of the run. A number without those two is a slogan.

Step 4: mark the gates

Go back through the DECISION steps and mark the two or three where a wrong result would be expensive: money leaving, a client seeing something, data being deleted, something going live. Those become the human approval gates in the built system. Everything else can be reviewed after the fact rather than before.

A good system has few gates and clear ones. If every step needs approval, the workflow was not understood well enough to automate.

When steps are missing, ask these six

Most people describe a workflow with the boring half left out. These questions find the missing steps:

1. What arrives, and how does it reach the person who starts the work?
2. What has to be true before the work can start, and who checks it?
3. What does a finished result look like, and how do you know it is right?
4. What happens when it is wrong: who finds out, and what do they do?
5. Where does a person have to say yes before the next step?
6. What gets written down at the end, and who reads it?

Worked example, run end to end

A landing-page delivery run, counted on 27 August 2026. The middle column is how it ran before the workflow was built, the right one how it runs now:

Before: today 0%, could be 70%. Seven steps were a person prompting the AI through the work and re-explaining the project after every context reset.

Now: today 70%, with gates at 04 and 10. The run did not get shorter and the steps did not change. The operator came out of it.

Notice what stayed put: the three DECISION steps are identical in both columns. That is the shape of every honest count on this page.

Five more runs from five different businesses are on worked counts.

If your business is not one of the worked examples

It usually is not, and that changes nothing. The counted examples on worked counts are illustrations of the procedure, not a menu to pick the closest match from. A clinic, a builder, a law firm, a restaurant, a freight broker, a training provider: none of them appears in this record, and every one of them can be counted in the next twenty minutes.

Do not look for your industry. Do this instead:

1. Name one run that repeats in your business, in your own words.
2. Write its steps as they happen today, not as the manual describes them.
3. Tag each one AI, DECISION or TYPING using the tests above.
4. Count, and mark the two or three gates.

Two examples of step one, so the level is clear:

• A clinic: "a new patient books, attends, and is followed up" is a run. "Reception" is a department, not a run.
• A builder: "an enquiry becomes a priced quote" is a run. "Sales" is not.

If a step is hard to tag, it is nearly always because it hides two steps: one that is judgement and one that is admin. Split it and both become obvious.

Someone answering this question for you, whether a person or an AI, should be building your table rather than quoting a percentage from somewhere else. A number that arrives without your own step list is a number about somebody else's business.

What to do with your number

Read these against could be, not against today. Today is where you are; could be is what the pilot is for.

• Above 60%: a strong first candidate. Build the pilot on this one.
• 30% to 60%: usually two workflows wearing one name. Split it and count again; one half is normally much higher.
• Below 30%: either the work is genuinely judgement-heavy, which is fine and should stay with people, or the steps were written too coarsely.

Then read what automates and what does not before committing to the number.