Find the task that makes you sigh.

Not the most interesting task. The one that happens every week, follows the same steps each time, and costs you time or money when it is late.

That is usually the right one to start with.

1. List it

Think about last week. What did you do more than once that followed roughly the same steps?

Write down five tasks. Actual tasks, not big goals.

2. Score it

Give each task a score from 1 to 5 on:

• How often it happens
• How long it takes
• How much money or momentum it affects
• How easy it would be to hand off
• How risky it is if the workflow gets it wrong

Do not pick the most exciting task.

Pick the task with the best mix of frequency, pain, value, and low risk.

3. Map it

Before touching any tool, write out the winning task like this:

What starts it
-> What comes in
-> What the workflow does
-> Where a person checks
-> What goes out
-> What happens next

This is the step most people skip.

It is also where the useful automation hides.

The map shows which parts need judgment and which parts are just repeated clicks.

4. Run it once manually

Follow your map exactly.

Time yourself. Notice where you slow down, hesitate, or make a decision.

Those points are either places that need a person in the loop or places where a workflow could take over.

5. Automate one step

Not the whole thing.

One step.

The step that costs the most time and needs the least judgment.

Test it. Let it run. Then move to the next one.

That is the 60-minute workflow test.

example: the missed follow-up

A local service business gets a lead while the owner is busy.

Maybe it is a missed call. Maybe it is a website form. Maybe it is an Instagram DM.

In one research example, 62% of calls to small local businesses went unanswered.

The exact number will change by niche, but the pattern is easy to spot: owners are busy doing the work, so leads wait.

The manual version usually looks like this:

Lead comes in
-> owner is busy
-> message sits
-> owner replies later
-> follow-up is forgotten
-> lead goes cold

That is not a character flaw.

It is a leaky workflow.

The mapped version:

Lead comes in
-> basic details are captured
-> a short reply is drafted or sent
-> follow-up task is created
-> owner gets a summary
-> owner steps in when needed

The first useful version does not need to do everything.

Start with capture, reminder, and summary.

Add the AI-drafted reply after you trust the basic flow.

Add two-way conversation only after you know the normal questions, edge cases, and handoff points.

choose the job type before the tool

This is the 2026 workflow rule:

Do not choose the tool first. Choose the job type first.

Ask what the workflow needs to do:

• Fixed rules: good for saving form entries, sending alerts, creating tasks, and moving records.
• AI with human review: good for draft replies, lead summaries, routing, scoring, and handoff notes.
• Conversation flow: useful when a lead may reply with questions and the workflow needs to respond.
• Code-first workflow: better when reliability, custom logic, testing, or client data matters.

After that, tools make more sense.

Examples: Zapier or Make for fixed rules, Relay or Vellum for human review, Dify or BuildShip for AI-heavy flows, and n8n, Pipedream, LangGraph, or custom code when reliability matters.

But the tool comes after the map.

Otherwise you end up debugging a workflow you never understood.

how this turns into money

The same test can pay off in a few ways.

Use it inside your own business and save time.

Set it up for a local business that misses leads, forgets follow-ups, or copies the same data between tools.

A simple missed-follow-up workflow can sell for $300 to $500 a month when it is tied to recovered leads.

Five clients is $1,500 to $2,500 a month before tool costs, support time, and churn.

Turn the repeated setup into a template, checklist, or workflow pack for a narrow audience.

The service path is usually easiest first because you learn from real messy workflows.

The product path gets easier once you have built the same thing a few times.

If you sell this as a service, do not start with “AI automation.”

Start with the business math:

How many leads do you miss each month?
What is one booked job worth?
What would two extra booked jobs be worth?
What part of that follow-up can we fix first?

That is a much better sales conversation than “I can build you a bot.”

Reality check:

Do not automate money movement, legal questions, medical decisions, or angry-customer handling on day one.

Start with low-risk admin work:

Lead capture. Reminders. Summaries. Intake routing. Draft replies. Simple handoffs.

use this prompt

Run this in Claude, ChatGPT, or any AI tool you already use before you touch an automation tool.

You are a workflow strategist with experience building AI-assisted automations for small businesses.
I want to find one repeated task that is worth automating or turning into a paid service.
Here is my context:
My business or audience: [describe it in one sentence]
The tasks I repeat every week: [list at least five, be specific]
Tools I already use: [list them]
My skill level: [non-technical / comfortable with no-code / I can write code]
What I want: [save time / build a service / build a product]
Do this in order:
1. Score each task from 1 to 5 on frequency, time cost per week, revenue impact, ease of automation, and risk if it breaks. Show your working.
2. Pick the best starting point and explain why the others lost.
3. Map the winning task as:
What starts it -> What comes in -> What the workflow does -> Where a person reviews -> What goes out -> What happens next
4. Identify which steps need a person and which steps are repeated clicks.
5. Tell me what kind of workflow this needs: fixed rules, AI with human review, conversation flow, or code-first workflow. Explain why.
6. Draft the first message or output this workflow should produce. Write it as if a real person sent it.
7. Tell me the one thing most likely to go wrong in the first week and what to do about it.
8. Give me one manual test I can run in the next 60 minutes without touching any tool. The test should tell me whether this workflow is worth building.

if you write code

Before wiring anything together, write the workflow as a small schema.

Define the trigger, inputs, steps, outputs, failure path, and human review point.

That is the useful part of FlowSpec.

It forces clarity before you start building.

One more rule: set up the workflow where the person already works.

If they live in their inbox, start there.

If they live in texts, start there.

If they live in a team chat, start there.

Friction kills good workflows.