Guide · pass or defer

What’s worth automating — and what to leave manual

Pass or defer the job first. Totals are not published yet. Self-hosted is a separate lane. A score is not a reason to automate a messy job. This page does not add new scores.

Automate this job, or leave it manual

Automate

Automate if…

The path is already lean and written down. The same trigger, inputs, and outputs repeat often enough that you can count the runs. Two people would follow the same rule. A miss is visible, or a person reviews the sensitive step. Someone owns alerts and a dead connection.

Skip

Leave it manual if…

The process is still changing, two people still do it differently, nobody owns the outcome, or a quiet miss is too costly without a review step. If the current row lives in more than one place, stop and name one hub first. That is a process problem, not a tool problem.

A pass-or-defer check for one job.

  • You want to know if this process is worth automating — not which tool.
  • You will leave changing, high-judgment, or unowned work with a person.

Live scorecard

Read more

What this page tells you. Help a small team decide what is worth automating — a pass-or-defer checklist — before picking Make, n8n, or Zapier. Automate a stable, frequent, written job that already has an owner and a failure path. Leave unstable, high-judgment, or high-silent-failure work manual until those are true. This page does not rank tools, does not reprint the four-question flowchart, and it adds no score cells. Dated list prices live on Which automation tool for small teams after a pass.

GuideSome scores not shown yetVendor pages accessed 2026-09-17Live scorecard

This page is not a tool pick

The question here is whether the process is ready. Which product to use comes after a pass: Which automation tool for small teams. When a named product is already the job, use When Make is the right pick, When Zapier is still the right pick, or When n8n self-host is worth it. When to pick each tool is also on Make vs n8n vs Zapier — when to pick each. This page does not reprint those reads. It does not rank tools. It does not add new scores.

This page also skips a list-price table. The question is whether the job is ready — not which plan to buy. After a pass, dated official list prices live on the picker. This page does not claim you will save money, and it does not invent a monthly dollar from a /500 score.

Stop if there is no one hub

If status, owner, or the current row live in more than one place — a Sheet, a Notion page, and Slack pins — do not encode the mess. Name one hub first. Automations read that hub, write that hub, and notify. They are not a second tracker.

  • Pass the hub. One named place for the current row. People edit that row tomorrow. In-hub edits can stay there.
  • Defer the job. Two hubs for the same job, or an automation writing to a copy while the team still edits the old page.

Stop here and open Name one system of record for small-team ops. Come back to this check after the hub is named.

Eight-step audit

Work the list in order. Each step is pass or defer — not a 1–5 score. If a step fails, leave the job manual and come back. Examples below do not invent a monthly dollar.

1. Lean the process first

Map trigger → inputs → outputs → owner. Cut waste before you encode it. NIST MEP’s Automation 101 says map the work and cut waste first: automation can amplify a bad process.

  • Pass. The path is already lean. You would not be automating extra steps or a second copy of the row.
  • Defer. You would encode a workaround, a duplicate tracker, or a step nobody can name.
  • Example. Pass: a form already writes the fields finance needs, then a person files the PDF. Defer: the same request is typed into email, a Sheet, and a Notion page.

Source: NIST MEP — Automation 101 · accessed 2026-09-17 · if the hub is messy, name one hub and stop.

2. Frequency and volume math

How often does it run — weekly, daily, hourly? Estimate steps × runs. Zapier’s plan guide asks you to list the processes and how often they run; a task is a successful action, so tasks ≈ actions × volume. Triggers do not count. Compare that estimate in each vendor’s own unit — do not treat one task as one credit as one execution.

  • Pass. You can estimate volume in the vendor’s unit and say why the job repeats.
  • Defer. You cannot count runs, or you are guessing a monthly bill from a /500 score.
  • Example. Pass: “about 20 paid-invoice writes a week, three actions after the trigger.” Defer: “this should save a lot of time” with no run count.

Source: Zapier — how to select your Zapier plan · accessed 2026-09-17 · Tasks vs credits vs executions · Same job, three meters

3. Rules you can write down

If two people still do the job differently, write the rule first. An automation will pick one version and run it at volume.

  • Pass. The happy path and the exception path are written. Two people would follow the same rule.
  • Defer. People still disagree on the edge cases, or nobody can state what success looks like.
  • Example. Pass: “if status is Paid, copy the invoice id; if status is Void, skip.” Defer: “qualify this inbound lead” when two people still use different rules.

4. Cost of a quiet miss

Ask what happens if nobody notices a skip, a duplicate, or a bad write. For sensitive or high-consequence work, keep a person on the review step. Zapier’s AI Guardrails help page recommends a person review for sensitive workflows and says the product is not a standalone compliance tool.

  • Pass. A miss is visible, or a person reviews the sensitive step.
  • Defer. A silent failure is too costly — refund, contract, payroll, or a write nobody will check.
  • Example. Pass: Slack the owner when the write fails; the empty hub row is also visible. Defer: a refund or contract change with no review step.

Source: Zapier — how to get started with AI Guardrails · accessed 2026-09-17 · after go-live, catch jobs that look fine.

5. One scoped job

Prefer a single critical job, or 3–5 related actions — not one flow that updates everything. Zapier’s Agents best practices: keep a limited scope, then test with real data and edge cases. On Make, split large work into subscenarios and apply Grid filters early so you process less data.

  • Pass. The first job is one owned outcome.
  • Defer. You are automating the whole shop in one run, or “keep the CRM in sync” across three trackers.
  • Example. Pass: paid invoice → one hub row → notify finance. Defer: one scenario that updates CRM, billing, Slack, and a personal Sheet.

Source: Zapier — best practices for working with Zapier Agents · accessed 2026-09-17

6. Failure path before go-live

Decide who is alerted, and what happens when a connection dies. In Make, a run of consecutive errors can disable scheduling; AccountValidationError means the connection needs a fix and cannot be swallowed by an error handler. In n8n, set an Error Trigger workflow in the job’s settings so a failure can page someone. Name the class first — auth, bad data, rate limit, timeout, quota, handled versus unhandled, or expected non-match.

  • Pass. Alerts and a dead-connection path exist. Someone would notice.
  • Defer. Nobody would notice. The job can look on while the write never happens.
  • Example. Pass: Error Trigger or consecutive-error disable pages the named owner. Defer: “we will check history if someone complains.”

Sources: Make — fix errors and warnings · n8n — handle errors gracefully · accessed 2026-09-17 · When automations fail — name the class first · When automations look fine

7. Owner and rollback

Name the person who gets the alerts. Operator, builder, or engineering is a different job — see Who should own your automations. Before you switch tools, keep the source job on, rebuild one critical job on the destination, compare outcomes, then cut over.

  • Pass. An owner and a rollback path exist.
  • Defer. The job is orphaned, or the only connection is under a contractor’s email.
  • Example. Pass: named owner, documented reconnect, and a parallel-run before cutover. Defer: “whoever built it” left last quarter.

Parallel-run one job before you switch · What does not migrate

Pass vs defer examples

Operator patterns, not an industry share. We do not invent a monthly dollar, a review count, or a converted list.

Swipe sideways on a narrow screen. Not a scored table.

Pass-or-defer examples for small-team jobs. Not a 1–5 score and not a monthly dollar.
JobVerdictWhy
Form submit writes one row in the named hub, then Slack the owner.PassSame trigger, same fields, same owner. A miss is an empty row or a quiet Slack channel.
When invoice status becomes paid, copy the invoice id into the hub and notify finance.PassOne scoped write. Two people would follow the same rule. Volume is countable in the vendor’s unit.
Nightly copy of yesterday’s closed tickets into the one table the team already edits.PassStable path. The hub is named. Failure can page the owner.
Qualify this inbound lead.DeferTwo people still use different rules. An automation will pick one version and run it at volume.
Friday exec narrative that still changes every week.DeferThe process is still changing. Encode it later, after the outline is lean and written.
Status lives in a Sheet, a Notion page, and Slack pins.DeferThere is no one hub. Name one place for the current row first. Do not wire Make, n8n, or Zapier on top of three copies.
Refund or contract change with no written exception path.DeferA quiet miss is too costly. Keep a person in the loop until the exception is written.
Keep the CRM in sync across three trackers.DeferNo named hub. Syncing three trackers is not one scoped job.

Leave these manual

Defer automation — keep the work manual, or have a person review it — when any of these are true:

  • The process is still changing.
  • Exceptions are unknown or already frequent.
  • Nobody can state what success looks like.
  • Nobody owns the alerts or the rollback.
  • A silent failure is too costly without a review step.
  • There is no one hub — two places, or a write to a copy.

That is a process problem, not a tool problem. Lean the job, write the rule, name the owner, name the hub. Then come back to this list.

After every step is a pass

The job is worth automating. Now pick how it runs — without treating this page as a ranking.

Then pick a tool

Primary CTA stays the live scorecard. This page does not reprint the picker flowchart.

How this sits on the scorecard

How this sits on the scorecard

The live card is Make vs n8n vs Zapier. Totals are not published yet.

When-to-pick copy lives on Make vs n8n vs Zapier — when to pick each. Longer compares: Make vs n8n · Make vs Zapier · n8n vs Zapier. Scoring rules: How scoring works.

Live scorecard When to pick each

Product links

Official

n8n

Official site: n8n · no affiliate ID · not a partner mark

Official

Zapier

Official site: Zapier · no affiliate ID · not a partner mark

Sources

Primary vendor pages and NIST MEP Automation 101. Access date for every citation on this page: 2026-09-17. This page does not publish list prices. We do not invent search volumes, testimonials, or a converted list.

FAQ

What’s worth automating first?

A job that is already lean, runs often, follows a rule you can write down, and has a named owner plus a designed failure path. Start with one critical job — or a small set of related actions — not the whole shop.

When should I leave a job manual?

Leave it manual, or add a person on the review step, when the process is still changing, exceptions are unknown or frequent, nobody owns the outcome, success is undefined, or a quiet miss is too costly. Write the rule first. Automate second.

What if the current row lives in more than one place?

Stop. If status, owner, or the current row live in more than one hub — a Sheet, a Notion page, and Slack pins — name one hub first. Automating on top of two hubs writes to a copy. See Name one system of record for small-team ops. Then come back to this check.

Does this page pick Make, n8n, or Zapier?

No. After every step is a pass, open Which automation tool for small teams. That page has the four-question flowchart and dated list prices. This page does not reprint it and does not rank tools. n8n Self-host is a separate lane.

Who should own the job before I turn it on?

Name the person who gets reconnects, alerts, and rollback. Operator, builder, or engineering is a different page: Who should own your automations. If nobody is named, defer.

How do I estimate volume before I pick a plan?

List the processes you want to automate and how often they run — weekly, daily, or hourly. Zapier counts a successful action as a task; triggers do not count. Tasks ≈ actions × volume. Compare that estimate in each vendor’s own unit. See Tasks vs credits vs executions and Same job, three meters. Do not treat one task as one credit as one execution. Do not invent a monthly dollar from a /500 score.

What if a connection dies after go-live?

Design the failure path before you turn the job on. In Make, a run of consecutive errors can disable scheduling; AccountValidationError means the connection needs a fix and cannot be swallowed by an error handler. In n8n, set an Error Trigger workflow in the job’s settings so someone is alerted. Name the owner who gets those alerts. Name the class on When automations fail, then catch jobs that still look fine.

Does this guide rank Make, n8n, or Zapier?

No. Totals are not published yet.

Affiliate disclosure

Make Partner (pc=stackscorecard) is live elsewhere on the site and is not the primary call here. n8n and Zapier stay official-site URLs with UTM only — no invented IDs. Zapier has no Stack affiliate. Rankings are not sold. Policy: Affiliate disclosure · Disclosures.