{}Route To AI
Courses
← AI Automation Builder
18 min

The Anatomy of an Automation

Triggers, steps and actions, when AI belongs in a workflow, essential vocabulary and a design canvas.

Lesson 1 of 7
0 done
In plain English

An automation is a small machine: something happens (the trigger), steps run, and an action is taken. Use plain rules where rules work, AI where text needs understanding and a person where mistakes are costly. Map the real process before you build.

Trigger
The event that starts the automation.
Webhook
A web address that receives an event the moment it happens.
JSON
A common way to write data as names and values.
Picture it
The anatomy of an automation

Use rules where rules work, AI where text needs understanding, and a person where mistakes cost a lot.

1
Trigger
Something happens
2
Steps run
Rules, AI or a person
3
Action
The result you want

Most small businesses lose hours every week to copying information between systems. A form entry goes into a spreadsheet. An email attachment goes into the accounting software. A new customer goes into the CRM, and a welcome message goes into their inbox. Automation removes that copying. AI adds something that rule-based automation never could. It can read messy human text and turn it into structured data or a sensible draft. This path teaches you to design, build, run and sell automations that you can rely on.

The anatomy of an automation

  • Trigger: the event that starts it. A new form entry, an incoming email, a row added to a sheet, a schedule or a webhook call.
  • Steps: changes and decisions. Filter, look up, calculate, branch, loop, wait or call an AI model.
  • Actions: what changes in the world. Create a record, send a message, update a sheet or post a notification.
  • Data: the fields passed from step to step, usually as JSON. Most automation bugs are data-mapping bugs.
  • Error handling: what happens when a step fails, a field is missing or an API is down.

When AI belongs in an automation

Use plain rules where rules work. If the form says Sales, send it to the sales inbox. Reach for AI when the input is unstructured or varied, and a rule would be brittle. For example, work out what a free-text email is asking for. Pull an invoice number and total out of a PDF. Summarize a long thread. Draft a reply. Or judge whether a lead matches your ideal customer. The best automations use AI for the one step that needs understanding, and ordinary logic for everything else.

  • AI is a good fit for: sorting into groups, pulling out facts, summarizing, drafting, translating and scoring against criteria.
  • Rules are better for: exact lookups, maths, date logic, permissions, and anything with a single right answer that you can compute.
  • Add a person when money moves, when messages go to customers, when data is deleted, or when a wrong action would cost a lot.

Words you will hear

  • Webhook: an address that receives an event the moment it happens. It is pushed to you. The alternative is polling, where the tool checks again and again.
  • API: how software talks to software. Most connections call APIs with your permission.
  • JSON: the structured text format that data travels in, with names, values and lists.
  • Iterator and aggregator: one turns a list into single items to process. The other combines the results back together.
  • Idempotent: safe to run twice without doubled effects. This is important for retries.
  • Operation or task: the unit that many platforms charge for. Loops multiply them.

The automation design canvas

automation-canvas.md
NAME:            {{what_it_is_called}}
OWNER:           {{who_is_responsible_when_it_breaks}}
GOAL:            {{outcome, e.g. every new enquiry gets a personal reply within 5 minutes}}
TRIGGER:         {{event + source}}
INPUT DATA:      {{fields available, example payload}}
STEPS:           1) ...  2) ...  3) ... (mark which are AI steps)
OUTPUT / ACTION: {{what changes, where}}
HUMAN CHECKPOINT:{{where a person approves or reviews, and how they are notified}}
FAILURE PLAN:    {{what happens on missing data / API error / AI uncertainty}}
VOLUME + COST:   {{runs per month x operations per run; AI cost per run}}
SUCCESS METRIC:  {{time saved, response time, error rate}}
TEST CASES:      {{5 normal, 3 messy, 2 hostile or broken inputs}}
→Before you build anything, map the current manual process step by step with a real person who does it. Automating a broken process just breaks things faster. And the real process always has exceptions that nobody mentioned.
Key takeaways
  • ✓An automation is a trigger, then steps, then actions, working over structured data, with a clear plan for failure.
  • ✓Use rules for exact logic, AI for understanding messy text, and people for high-stakes decisions.
  • ✓Learn the words: webhooks, APIs, JSON, iterators, idempotency and operations.
  • ✓Design on a canvas first, starting from the real manual process.
Hands-on exercise

Map one process and design its automation

  1. Pick a repetitive process, yours or a friend's business. Write each manual step in order, including the exceptions.
  2. Mark which steps are rules, which need AI understanding, and which need a person.
  3. Fill in the automation design canvas for the version you would build first.
  4. Estimate the monthly runs and the minutes saved per run, to get a rough value.
Check yourself
Q1What kind of step is a good use of AI in an automation?show answer

One that needs understanding of unstructured input, such as sorting, pulling out facts, summarizing, drafting or scoring. Exact logic stays in rules.

Q2Why does idempotency matter?show answer

Automations retry, and sometimes run twice. Idempotent steps avoid duplicate emails, charges or records.

Q3Why map the manual process before you build?show answer

To find the exceptions and the real flow. Automating a process you only assumed, or a broken one, creates failures at scale.

Choosing Platforms & Understanding Real Costs →
0 / 7 finished0%