Workflow automation in three lines
Workflow automation means an online service repeats a chain of routine steps between your apps for you: when this event happens here, do this there, then that somewhere else. With a no-code tool, you assemble the chain in a visual editor and write no program. “AI automation” is the same thing with one step in the middle handed to an AI model: classify a message, summarize a document, draft a reply.
Three words to know: trigger, action, workflow
| Term | What it means | Example |
|---|---|---|
| Trigger | The event that starts the chain | A form is submitted, an order is paid, an email arrives |
| Action | Each step that runs afterward | Add a spreadsheet row, send an email, create a customer record |
| Workflow | The whole chain, from the trigger to the last action | “For every form entry, add a row, then notify the team” |
The workflow goes by a different name at each vendor: “Zap” at Zapier, “scenario” at Make and Latenode, “flow” at Activepieces, “workflow” at n8n and Pabbly Connect.
Two details matter for what follows.
- A trigger works in one of two ways. Either the source app notifies the tool the moment the event happens (that is a webhook, covered in What is a webhook), or the tool checks on a schedule whether anything is new (that is polling). The difference affects the delay and, at some vendors, the bill.
- Logic sits between the actions. A filter stops the workflow when a condition is not met, a branch sends it down one of two paths, a delay puts it on hold.
A worked example: the new lead
This workflow is described from the rules the vendors publish: automascope has not built or run it.
Trigger: a visitor submits the contact form on your website.
- Add a row to your spreadsheet or CRM.
- Send the visitor a confirmation email.
- Notify the team in your chat app.
At 500 leads a month, here is what each tool counts under its own billing rule.
| Tool | What is counted | Math | Units per month |
|---|---|---|---|
| Make | Every module that runs, trigger included | 500 × (1 + 3) | 2,000 credits |
| Zapier | Every successful action, not the trigger | 500 × 3 | 1,500 tasks |
| Pabbly Connect | Every action, not the trigger | 500 × 3 | 1,500 tasks |
| n8n | The whole run, whatever the number of steps | 500 | 500 executions |
| Activepieces | The whole run, whatever the number of steps | 500 | 500 credits |
| Latenode | Compute time (assumption: 5 seconds per run) | 500 × 5 | 2,500 CPU seconds |
The same automation “costs” anywhere from 500 to 2,500 units. Those units share neither a name nor a price, so the count alone says nothing about the bill. The cost calculator applies these rules and each vendor’s recorded price grid; enter 500 runs, 3 actions and 5 seconds to get this table back.
Three things to read from it:
- 500 credits fit in the Activepieces free plan (1,000 credits a month). At Make, 2,000 credits exceed the free plan (1,000 credits a month).
- Zapier’s free plan only allows two-step Zaps, one trigger and one action: this three-action workflow does not fit.
- At Make, a trigger uses one credit per check, even when it finds no new data. With polling every 15 minutes, that is 2,880 checks a month, 500 of which find a lead: add 2,380 credits, for a total of 4,380.
The rules are detailed in Operations, credits, tasks, executions.
What you can automate
A task is a good candidate when it is repetitive, started by an identifiable event, governed by a stable rule, and when a mistake can be fixed without damage. The most common cases for a solo founder or a small business:
- Copy data from one app to another: form to spreadsheet, order to customer list, booking to calendar.
- Notify: a message to the team for every order, every new rating, every failed payment.
- Follow up on a schedule: a reminder before an appointment, a nudge after a quote that got no answer.
- Sort and prepare with an AI step: classify incoming messages, prepare a draft reply. Automate email replies with AI describes that workflow and what it uses.
- Chain the invoicing steps: create the invoice from an order, send it, chase late payers. See Automate invoicing.
What is better left manual
- Decisions that commit you. Granting a discount, turning down a request, answering an unhappy customer: automate the preparation (gather the facts, propose a draft), keep the decision.
- Sending AI-written text nobody has read. A model can make up a price, a deadline or a promise. The draft can write itself; the send button stays yours.
- Rare tasks. Something you do four times a year pays back neither the build time nor the upkeep.
- A process that is still changing. Automating freezes a way of working. Settle it by hand first.
- Sensitive data, without knowing where it goes. Every piece of data handled passes through the tool’s servers. The table below shows where each vendor says it hosts data.
A workflow does not watch itself: one app renaming a field is enough to make it fail quietly. Set up an error alert, and make sure someone reads it.
Automation, AI step, agent: three levels
A workflow follows steps decided in advance. An AI step replaces one of them with a call to a model. An AI agent goes further: it is given a goal and chooses which tools to call (see What is an AI agent).
AI steps cost extra. At Zapier, an “AI by Zapier” step counts as 1, 3 or 5 tasks depending on the model tier. At Make, the AI the vendor provides adds credits based on the tokens used, in other words on the length of the text processed.
How to pick a tool
Start with two checks nobody can do for you: are your apps on the vendor’s integration list, and how many steps will your workflows have? Each vendor publishes its list, for example n8n, “Integrations” page (accessed ) or Activepieces, “Pieces” page (accessed ). Then read this table, generated from the site’s database.
| Criterion | MakeAutomation tool | ZapierAutomation tool | n8nAutomation tool | Pabbly ConnectAutomation tool | ActivepiecesAutomation tool | LatenodeAutomation tool |
|---|---|---|---|---|---|---|
| Billing unit | creditssource | taskssource | executionssource | taskssource | creditssource | CPU secondssource |
| Free plan | 1,000 credits per monthsource | 100 tasks per monthsource | Nosource | 100 tasks per monthsource | 1,000 credits per monthsource | 10,000 CPU seconds per monthsource |
| Self-hosting | No: no self-hosted plan on the pricing page; AWS hosting (EU, North America) for every plansource | No: no self-hosted plan; the vendor hosts data on AWS servers in the United Statessource | Yes: free Community edition, no license key; paid self-hosted plans Business and Enterprisesource | not recorded | Yes: free Community Edition with no credits; every plan on the pricing page is also available self-hostedsource | not recorded |
| Data residency | Data center in the US or the EU, chosen when the organization is created and not changeable afterwardssource | AWS servers in the United States; no EU-only storage option ("Zapier does not support this option")source | Hosted plans: EU, servers in Frankfurt (Germany); self-hosted: wherever n8n is installedsource | Amazon Web Services (United States) and Hetzner Online GmbH (Finland), according to the privacy policysource | Cloud hosted in the EU, in Frankfurt by default; another region on request; self-hosting in the region of your choicesource | Hosting on Microsoft Azure and Amazon Web Services, both listed as "US" in the list of subprocessorssource |
| Vendor | Visit Make (opens in a new tab) | Visit Zapier (opens in a new tab) | Visit n8n (opens in a new tab) | Visit Pabbly Connect (opens in a new tab) | Visit Activepieces (opens in a new tab) | Visit Latenode (opens in a new tab) |
Data recorded from the official pages: Make (), Zapier (), n8n (), Pabbly Connect (), Activepieces (), Latenode ().
- If your workflows are long (five steps or more), a unit counted per run does not grow with the number of steps: that is the rule at n8n and Activepieces.
- If your workflows are short (one or two actions), per-action counting at Zapier and Pabbly Connect does not penalize you, and their triggers are not counted.
- If your data has to stay in a given region or on your own server, read the “Self-hosting” and “Data residency” rows: n8n and Activepieces can be installed on your own machine, Make has you choose a zone when the organization is created, Zapier states that it hosts in the United States.
- If you want to start without paying, compare the free plans, limits included. n8n publishes none for its hosted version: it offers a 14-day free trial, plus a free edition you install yourself.
- If nobody on the team is technical, go with a version hosted by the vendor: n8n itself reserves self-hosting for experienced users.
The workflow automation tools comparison walks through these six tools row by row.
Data recorded from the official pages: n8n (), Make (), Zapier (), Latenode (), Pabbly Connect (), Activepieces ().
CalculatorAutomation cost calculator: Make, Zapier, n8n and moreFrequently asked questions
What is workflow automation, in one sentence?
It is software that runs a fixed chain of steps for you whenever a given event happens: a trigger, then one or more actions across your apps. No-code tools let you assemble that chain in a visual editor instead of writing a program.
Is a workflow the same thing as a Zap or a scenario?
Yes. Zapier calls it a Zap, Make and Latenode a scenario, Activepieces a flow, n8n and Pabbly Connect a workflow. The name changes, the object is the same: one trigger and the actions that follow it.
Do I need to know how to code?
Not to assemble a common workflow: the visual editor is enough. Some tools offer an optional code step for cases their connectors do not cover. Self-hosting a tool on your own server does take technical skills.
Is workflow automation free?
Five of the six tools compared here publish a permanent free plan, with a monthly quota and limits on steps, active workflows or trigger frequency. Quotas and prices are in the table on this page and in the cost calculator, with the date they were recorded.
What is the difference between workflow automation and an AI agent?
A workflow follows steps decided in advance. An AI agent is given a goal and decides for itself which tools to call and in what order. In between, a workflow can include one AI step, for example to classify or summarize a text.