Guide · name the class
When automations fail — a practical taxonomy for small teams
Name the class first. Totals are not published yet. Self-hosted is a separate lane. A score is not a reason to rewrite a flow. This page does not add new scores.
Use this taxonomy, or skip it
Use
Use this taxonomy if…
A job failed, a write did not happen, or a run looks wrong, and you need to name the kind of failure before you rewrite the flow. Auth, bad data, rate limits, timeouts, quota, handled versus unhandled, and expected non-match are different jobs. Name the class. Then do that job.
Skip
Skip it if…
You still need to decide whether the job is worth automating — stop and open What’s worth automating. You want how each product names retries and reconnects — that is Error handling across Make, n8n, and Zapier. You want to count units or pick a tool — that is metering or the picker. This page classifies the miss. It does not rank products.
A classify-or-skip check for one failed job.
- You want to name the kind of failure before you rewrite the flow — not which tool.
- You will treat auth, bad data, rate limits, timeouts, quota, handled vs unhandled, and expected non-match as different jobs.
Read more
What this page tells you. Help a small team classify an automation failure — kinds of failure and how to name the class — before you rewrite the whole flow. Auth, bad data, rate limits, timeouts, quota, handled versus unhandled, and expected non-match are different jobs. For each class: what it looks like, who notices first, and the first check. 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 you name the class.
GuideSome scores not shown yetVendor pages accessed 2026-09-16Live scorecard
This page is not a tool pick
The question here is what kind of failure you have. Whether the process is ready lives on What’s worth automating. What to watch on the meter lives on How to meter automation. How each product names success, retries, and reconnects lives on Error handling across Make, n8n, and Zapier. Which product to use lives on 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. 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 the class of the miss — not which plan to buy. After you name the class, 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.
Name the class first
An error in the history is not one problem. Reconnecting an expired account is not the same work as fixing a mapping, waiting out a 429, splitting a long run, or buying credits. Name the class. Then do that job. Rewrite the flow only when the class says the design is wrong.
This table is the canonical class list, not a 1–5 score. It adds no score cells and does not change rankings. After you name the class, catch silent breakage on When automations look fine — that page does not add an eighth class.
Failure classes A–G
Seven classes. Vendor names below are the labels those products already use. Primary docs accessed 2026-09-16.
Swipe sideways on a narrow screen. Not a scored table.
| Class | What you see | What to do |
|---|---|---|
| A · Auth or connection setup | Make AccountValidationError (also HTTP 401/403). Zapier 401 or On hold from a disconnected account. The job never starts because the account is dead or expired. | Reconnect the app. On Make this error happens during setup and cannot be swallowed by an error handler. Do not rewrite the flow. |
| B · Bad or incomplete data | Make BundleValidationError or DataError. Zapier 400 or 422. A required field is empty, or the type or format is wrong. | Fix the mapping and the payload. Review required fields and formats. This is a data job, not a new scenario. |
| C · Rate limit or the other app is down | Make RateLimitError (HTTP 429) or ConnectionError (502/503/504). Zapier 429 or 500. Often temporary. | Wait and retry. Make pauses about 20 minutes; turn on incomplete executions if you want stored retries. On Zapier, replay the run or enable Autoreplay. Do not rebuild the job. |
| D · Timeout or a long run | Make ModuleTimeoutError when a request sits past about 40–60 seconds. ExecutionInterruptedError when a run lasts more than 45 minutes (10 minutes on Free). | Add a Retry handler and store incomplete executions. If the whole run is too long, split the job or lower search limits. Do not treat a slow API as a mapping bug. |
| E · Quota or a fatal stop | Make OperationsLimitExceededError (credits) or DataSizeLimitExceededError (data transfer). Both are fatal and disable scheduling immediately. | Buy extra credits or change the plan. This is a quota stop, not a broken mapping. Do not keep replaying until the allowance is there. |
| F · Handled versus unhandled | n8n Error Trigger plus Stop And Error. Zapier Handled error versus Errored. An unhandled Zap that errors on about 95% of runs in 7 days turns off. | Design who is alerted. A handled path pages someone and keeps the job on. An unhandled path can turn the job off. Name the owner before go-live. |
| G · Expected non-match | Zapier Safely halted — usually a search found no row. This is a purposeful stop, not an error. | Do not treat a clean non-match as a failure. It will not auto-off the Zap. Later required fields mapped from that search can still error — decide what those steps should do. |
Make stores unfinished runs only when incomplete executions are on — they are off by default. n8n’s Error Trigger runs on automatic failures, not on a manual test. Zapier Autoreplay retries Errored runs; it does not replay a Safely halted search.
Class detail — looks like, who notices, first check
Canonical classes only. Each block cites a primary vendor or community page. Access date: 2026-09-16. We do not invent reply counts or a quote statistic. The reconnect-loop story stays on When automations look fine.
A · Auth or connection setup
- What it looks like. Make AccountValidationError (HTTP 401/403) during initialization. Zapier 401, or On hold from a disconnected account. An n8n credential can still be saved after the token stops refreshing — that is still this class, not a mapping bug.
- Who notices first. The connection owner — or the person who owns the destination write — notices first. A scheduled Make scenario can still look on. A Zap can stay On while the run is On hold.
- First check. Test the connection. Reconnect or reauthorize. On Make, AccountValidationError cannot be swallowed by an error handler. Do not rewrite the flow. If reconnect loops, use the silent-breakage checklist — do not invent an eighth class.
Next read: When automations look fine — reconnect loops.
Sources: Make — fix errors and warnings · Zapier — how to troubleshoot errors in Zap workflows · Zapier — review run statuses in Zap workflows · Zapier — troubleshoot held Zap or step runs · accessed 2026-09-16
B · Bad or incomplete data
- What it looks like. Make BundleValidationError when a required field is empty or a type does not match. DataError when the other app rejects the payload. Zapier 400 or 422, or “Required field is empty.”
- Who notices first. The builder reading history notices first. The destination owner notices next — a blank required field, a wrong type, or a rejected write.
- First check. Open the failed step. Confirm every required field has a value and the right format. Zapier: only map required fields to fields that always send a value; add a default or a filter if they do not. Make: review the mapping. This is a data job, not a new scenario.
Next read: What does not migrate — mappings do not travel.
Sources: Make — fix errors and warnings · Zapier — how to troubleshoot errors in Zap workflows · Zapier — fix required field is empty · accessed 2026-09-16
C · Rate limit or the other app is down
- What it looks like. Make RateLimitError (HTTP 429) or ConnectionError (502/503/504). Zapier 429, “Throttled by Zapier,” or a one-off 500. A Make Community thread (429, wait and try again) is this class — wait, then retry the same module.
- Who notices first. The operator reading history notices first — a pause, a 429, or a 5xx. The destination owner notices a gap, not a bad mapping.
- First check. Wait and retry. Make pauses about 20 minutes; incomplete executions are off by default — turn them on if you want stored retries. On Zapier, replay the run or enable Autoreplay. Do not rebuild the job.
Next read: Parallel-run — compare the failure path.
Sources: Make — fix errors and warnings · Make — incomplete executions · Make Community — 429 rate limit / wait and try again · Zapier — how to troubleshoot errors in Zap workflows · Zapier — replay Zap runs · Zapier — Zap limits (rate limits and task allowance) · accessed 2026-09-16
D · Timeout or a long run
- What it looks like. Make ModuleTimeoutError when a request sits past about 40–60 seconds (Airtable can wait up to 80). ExecutionInterruptedError is a warning when a run lasts more than 45 minutes (10 minutes on Free). Zapier: “The app did not respond in time.”
- Who notices first. The operator notices a timeout or a warning that leftover bundles did not run. The destination owner notices a partial write — early rows landed, later ones did not.
- First check. Add a Retry handler and store incomplete executions. Split the job or lower search limits. Zapier: limit trigger data, do not query a large set without a filter, then replay. Do not treat a slow API as a mapping bug.
Next read: What’s worth automating — split or leave manual.
Sources: Make — fix errors and warnings · Make — error handlers · Make — incomplete executions · Zapier — how to troubleshoot errors in Zap workflows · Zapier — the app did not respond in time · accessed 2026-09-16
E · Quota or a fatal stop
- What it looks like. Make OperationsLimitExceededError (credits) or DataSizeLimitExceededError (data transfer). Both are fatal and disable scheduling immediately. Zapier On hold from a task-limit overage. This is allowance, not auth and not a mapping bug.
- Who notices first. The plan owner or billing inbox notices first. On Make, scheduling stops at once. On Zapier, runs sit On hold until tasks reset, the plan changes, or pay-per-task is on.
- First check. Read the vendor error name. If it is quota, stop replaying. Compare the same job on three meters before you buy allowance. Do not invent a monthly dollar from a /500 score.
Next read: Same job, three meters — units only.
Sources: Make — fix errors and warnings · Make — introduction to errors and warnings · Zapier — review run statuses in Zap workflows · Zapier — Zap limits (rate limits and task allowance) · Zapier — troubleshoot held Zap or step runs · accessed 2026-09-16
F · Handled versus unhandled
- What it looks like. n8n Error Trigger plus Stop And Error. Zapier Handled error versus Errored. A handled path ran an alternative workflow. Unhandled Errored did not complete. Zapier turns a Zap off if it errors 95% of the time it runs and has run more than 20 times in the past 7 days.
- Who notices first. The named owner notices first — if a handler exists. n8n’s Error Trigger runs on automatic failures, not on a manual test. If nobody is named, the destination owner or a Zap that turned off notices later.
- First check. Ask whether a handler ran and who was paged. Name that owner on Who should own your automations. A handled path can keep the job on. An unhandled path can turn it off. Wire the Error Trigger in the workflow’s settings.
Next read: Who should own your automations.
Sources: n8n — handle errors gracefully · n8n — Error Trigger · n8n — Stop And Error · Zapier — how to troubleshoot errors in Zap workflows · Zapier — review run statuses in Zap workflows · Zapier — Zap is not running (95% error ratio) · accessed 2026-09-16
G · Expected non-match
- What it looks like. Zapier Safely halted — usually a search found no row. Autoreplay retries Errored runs; it does not replay a Safely halted search. Later required fields mapped from that search can still error (class B).
- Who notices first. Often nobody — the Zap stays On and the history can look clean. The destination owner notices a missing write. That is why the look-fine checklist exists. This page only names the class.
- First check. Read the vendor status. A clean non-match is not a dead account. Decide what later steps should do when the search is empty. Do not treat this as class A or class F.
Next read: When automations look fine — false-green runs.
Sources: Zapier — how to troubleshoot errors in Zap workflows · Zapier — review run statuses in Zap workflows · Zapier — replay Zap runs · accessed 2026-09-16
Silent-failure rule
If nobody would notice a skip, a duplicate, or a bad write, do not ship the job yet. A quiet miss is a process problem. Name the class so you know whether to reconnect, fix data, retry, split, buy allowance, or design a handler — then decide whether the miss would still be invisible.
For the checklist — reconnect loops, false-green runs, who gets paged, and what to log — see When automations look fine — catch silent breakage. This page stays the class list. That page stays the narrative checklist.
Pass or defer that job on What’s worth automating — and what to leave manual. Automate only when a miss is visible or a person reviews the sensitive step.
After you name the class
You named the kind of failure. Now use the sibling that owns the next job — without treating this page as a ranking.
- Still deciding if the job is ready: What’s worth automating — and what to leave manual.
- Product-lane retries and reconnects: Error handling across Make, n8n, and Zapier — not a second class list.
- Looks fine, but the write did not happen: When automations look fine — catch silent breakage — that page does not add an eighth class.
- Nobody is named for reconnects or 2am: Who should own your automations.
- The stop is allowance, not a mapping: How to meter automation · Same job, three meters — tasks vs credits vs executions — units only.
- You are switching tools: Parallel-run before you switch · What does not migrate between Zapier, Make, and n8n — auth, history, and proprietary nodes stay on the source.
- Four questions, then a pick: Which automation tool for small teams.
- When a named product is the job: When Make is the right pick (and when it is not) · When Zapier is still the right pick (and when it is not) · When n8n self-host is worth it (and when it is not).
- When to pick each: Make vs n8n vs Zapier — when to pick each · live scorecard (totals are not published yet; self-host is a separate lane).
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.
Product links
Sources
Primary vendor error and replay pages, plus community threads where this page names a concrete failure mode. Access date for every citation on this page: 2026-09-16. This page does not publish list prices. We do not invent a cross-vendor error type, a reply count, or a quote statistic.
- Make — fix errors and warnings · accessed 2026-09-16
- Make — error handlers · accessed 2026-09-16
- Make — incomplete executions · accessed 2026-09-16
- Make — introduction to errors and warnings · accessed 2026-09-16
- Make Community — 429 rate limit / wait and try again · accessed 2026-09-16
- n8n — handle errors gracefully · accessed 2026-09-16
- n8n — Error Trigger · accessed 2026-09-16
- n8n — Stop And Error · accessed 2026-09-16
- Zapier — how to troubleshoot errors in Zap workflows · accessed 2026-09-16
- Zapier — review run statuses in Zap workflows · accessed 2026-09-16
- Zapier — replay Zap runs · accessed 2026-09-16
- Zapier — fix required field is empty · accessed 2026-09-16
- Zapier — Zap limits (rate limits and task allowance) · accessed 2026-09-16
- Zapier — Zap is not running (95% error ratio) · accessed 2026-09-16
- Zapier — the app did not respond in time · accessed 2026-09-16
- Zapier — troubleshoot held Zap or step runs · accessed 2026-09-16
FAQ
Why name the failure class before I rewrite the flow?
Auth, bad data, rate limits, timeouts, quota, handled versus unhandled, and expected non-match are different jobs. Reconnecting an expired account is not the same work as splitting a long run or buying credits. Name the class first so you fix the right thing.
How is this page different from When automations look fine?
This page is the canonical class list. When automations look fine is the operator story — reconnect loops, false-green runs, who gets paged, and what to log. It does not add an eighth class. Name the class here, then use that checklist.
How is this page different from Error handling across Make, n8n, and Zapier?
This page names the kind of failure. Error handling is the product-lane compare: how each tool names success, retries, and reconnects. Use both. Do not treat this table as a second scorecard or a tool pick.
Who notices first, and what is the first check?
Each class names who notices first and the first check. Auth is a reconnect. Bad data is a mapping. Rate limits and timeouts are wait, retry, or split. Quota is allowance. Handled versus unhandled is who is paged. Expected non-match is a purposeful stop, not an error.
What is the difference between a handled error and an unhandled one?
A handled error ran an alternative path you designed — n8n’s Error Trigger, or Zapier’s Handled error status. The job can stay on and someone can be alerted. An unhandled Zapier Errored run did not complete. Zapier turns a Zap off if it errors 95% of the time it runs and has run more than 20 times in the past 7 days.
When is a halt not a failure?
Zapier’s Safely halted status is a purposeful stop, usually because a search found no results. It does not turn the Zap off. Autoreplay does not replay a Safely halted search. Later steps still run; if they require a value from that search, they can error. Decide that path on purpose.
Is a quota stop the same as metering?
No. Class E is a fatal stop — Make disables scheduling at once; Zapier holds runs when the task allowance is gone. Metering is how tasks, credits, and executions count the same job. Compare units on How to meter automation and Same job, three meters. This page does not convert those units into a monthly dollar.
What if nobody would notice a miss?
Do not ship the job yet. A quiet skip, duplicate, or bad write is a process problem. Pass or defer that job on What’s worth automating. Name who owns reconnects and 2am on Who should own your automations. If you are switching tools, parallel-run one critical job and compare the failure path before you cut over.
Does this page pick Make, n8n, or Zapier?
No. Name the class first. Then open Which automation tool for small teams if you still need a product. 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.
Does this guide change the Make vs n8n vs Zapier scores?
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.