Workflow Automation With Error Handling and Retries: What 10 Platforms Publish
Affiliate disclosure: some links here are affiliate links and we may earn a commission at no cost to you. The order below follows the stated criterion and nothing else; no vendor paid for a position or saw this page before it went up.
A workflow fails at 2am. The only question that matters is what the platform does next: retry the step, branch to a recovery path, or stop and wait for a human to notice. Across the 13 no-code automation platforms whose published documentation we compared on 2026-09-17, exactly one answers that in numbers. Make documents automatic retries via exponential backoff for connection and module-timeout errors, escalating across up to 8 attempts — 1 min, 2 min, 5 min, 10 min, 1 hour, 3 hours, 12 hours, 24 hours — before scheduling is disabled. Everyone else either names retries without publishing the count and interval, or publishes that they do not retry at all.
So that is what this page ranks: what each vendor publishes about failure — retries, error branches, version history, rollback and execution logs — read from their own pricing pages and documentation, not from hands-on use. Nobody here installed or subscribed to anything. Three platforms we checked are not ranked at all: Bardeen publishes nothing on its public site about what happens when a step fails, Gumloop publishes an auto-disable circuit breaker instead of a retry policy, and Flowise’s published End of Life date of August 31, 2026 has already passed while its paid tiers are still displayed.
What this ranking measures
This ranking measures one thing: what each vendor publishes about failure. Five items, in order of weight — whether retries exist and whether the count and interval are stated; whether there is an error branch, error handler or error workflow you can build a recovery path with; whether version history exists and its retention is published per plan; whether a previous version can be restored, and what restoring destroys; and how long execution logs or run history are kept. A vendor that documents all five outranks one that documents three, and “not published” costs a place. It deliberately does not measure execution speed, ease of use, integration counts, AI output quality, support or design. Price is recorded verbatim because buyers need it, but nothing here is ranked up for being cheap or down for being expensive — only for being silent about what happens when a step fails.
The short answer
It is the only platform on this sheet that turns “automatic retries” into numbers: up to 8 attempts at 1 min, 2 min, 5 min, 10 min, 1 hour, 3 hours, 12 hours and 24 hours before scheduling is disabled. Add five named error-handler directives, a statement that error handler modules are not billed against your credit allowance, and execution log storage published at 7 days Free, 30 days Core/Pro/Teams and 60 days Enterprise, and it documents more of what happens on failure than anything else here.
It covers more of the criterion than Make does — automatic retries on every paid tier as a per-node Retry On Fail setting, a workflow-level error workflow, workflow history published at 1 day on Starter, 5 days on Pro and 365+ days on Enterprise, and Git version control from Business up. It takes second only because it never publishes the retry defaults. If rollback and version history matter to you more than a published retry schedule, reverse these two.
The free plan gives 1,000 credits a month, 2 active scenarios, a 15-minute minimum interval between runs and 7 days of execution log storage — enough to read the failure behaviour against your own scenarios. Crucially, Make publishes no plan gate on its retry behaviour the way Zapier explicitly gates Autoreplay to Professional and above and documents Free as having no version history.
All 10 compared
| # | Product | Best for | Price from | What decides its place | |
|---|---|---|---|---|---|
| 1 | Teams that want the retry schedule in writing | Free $0 /mo (up to 1,000 credits/mo); Core $9 /mo paid annually or $10.59 /mo paid monthly, price for 10k credits/mo | Retries up to 8 times: 1, 2, 5, 10 min, then 1, 3, 12, 24 hours | Visit | |
| 2 | Version history plus Git on self-hosted workflows | Starter 20€/mo, billed annually; Pro 50€/mo, billed annually; Business 667€/mo, billed annually; Community Edition free (self-hosted) | Workflow history: 1 day Starter, 5 days Pro, 365+ Enterprise | Visit | |
| 3 | Enterprises that can wait for a sales quote | not published — Pro and Team are “Get a demo”, Enterprise is “Talk to sales”; 30-day free trial with 1M tasks included | Failed-run replay: 7 days Pro and Team, 30 days Enterprise | Visit | |
| 4 | Microsoft 365 shops needing per-action retry rules | Premium $15.00 user/month, paid yearly; Process $150.00 bot/month, paid yearly; Hosted Process $215.00 bot/month, paid yearly | Versions purge at 6 months draft, 12 months published | Visit | |
| 5 | Agent builders who need unlimited version history | Pro $19 / month annual, $29 / month monthly; Team $234 / month annual, $349 / month monthly | Versions kept indefinitely; restore lands in draft, not live | Visit | |
| 6 | Wide app coverage with retries switched on once | Free $0 /month; Professional Starting from $19.99 /month on yearly billing, $29.99 /month billed monthly; Team from $69 /month yearly, $103.50 /month monthly | Version history: none on Free, 1 month on Professional | Visit | |
| 7 | Per-block retry control with pass-through model cost | Free $0 /month + usage; Individual $20 /month + usage billed monthly, or $16 /month + usage billed yearly | Retries set per block; no retention or limit numbers published | Visit | |
| 8 | LLM workflows that need a visible fail branch | Sandbox Free; Professional $59 Per workspace/month, or $590 per workspace / year; Team $159 Per workspace/month, or $1590 per workspace / year | Three node error modes published; no retry page in the docs | Visit | |
| 9 | Auditing what failed, not making it retry | Free; Team $20 per seat /month billed annually ($24 per collaborator/ per month billed monthly); Business $45 per seat /month billed annually ($54 per collaborator/ per month billed monthly) | No retry: a failed run notifies one person and still burns a run | Visit | |
| 10 | Document workflows, not jobs that must not stop | Free $0 per member / month; Plus $10 per member / month yearly, $12 per member / month monthly; Business $20 per member / month yearly, $24 per member / month monthly | A failed automation is disabled until a human re-enables it | Visit |
1. Make
Why it sits at 1 Retries up to 8 times: 1, 2, 5, 10 min, then 1, 3, 12, 24 hours
Make is a visual scenario builder metered in credits, where each module action consumes one credit. It leads this ranking for a single reason: it is the only vendor here that publishes the retry ladder as numbers. Connection and module-timeout errors are retried with exponential backoff across up to 8 attempts, at 1 min, 2 min, 5 min, 10 min, 1 hour, 3 hours, 12 hours and 24 hours, after which scheduling is disabled.
Around that sit five named error-handler directives, and Make states that error handler modules are not billed against your credit allowance — so building the recovery path costs no usage. Execution log storage is published per tier at 7 days on Free, 30 days on Core, Pro and Teams and 60 days on Enterprise, with full-text log search from Pro upward. The hole is version history: no versioning, rollback or restore topic appears in the Help Centre’s Scenarios index, and no version row appears in the pricing comparison table.
- The only platform here that publishes both the number of retry attempts and the interval between them, plus error handlers that cost no credits.
- No scenario version history is published anywhere; the “Rollback” on the pricing page is a transactional error-handler directive, not a way back to a previous version of your scenario.
2. n8n
Why it sits at 2 Workflow history: 1 day Starter, 5 days Pro, 365+ Enterprise
n8n meters whole workflow executions rather than steps — 2.5k on Starter, 10k on Pro, 40k on Business, all with unlimited steps and unlimited users. Failure handling is published at two levels: automatic retries are listed on every paid tier and documented as a per-node Retry On Fail setting that reruns the failing node until it succeeds, and a designated error workflow can be triggered whenever an execution fails.
It covers more of the criterion than anything else here. Workflow history retention is published per tier, Git-based version control arrives on Business and above, execution log retention runs 7 days on Starter and 30 days on Pro, and saved executions are capped separately at 2.5k/25k/50k with 2.5 GB/25 GB/50 GB of storage. What it withholds is the detail Make publishes: no default retry count and no retry interval on either the pricing page or the node docs.
- Failure handling at two levels — a per-node retry and a workflow-wide error workflow — with history retention published in days for every named plan.
- The pricing page rendered only annual per-month prices in EUR, the monthly-billed figure and free-trial length are not published, and the FAQ states n8n does not offer pro-rata refunds.
3. Tray.ai
Why it sits at 3 Failed-run replay: 7 days Pro and Team, 30 days Enterprise
Tray documents error handling per step in four modes, with automatic retries and backoff applied per error type, and states plainly that automatic retries are not billed as tasks. It also publishes the trade-off most vendors leave implicit: choosing Manual, Continue, Break loop or Continue loop disables automatic retries and backoff for that step. Error alerting is a workflow you build yourself from an Alerting Trigger.
Project version history behaves like code commits — titled, described snapshots that list what changed and who changed it, restorable or exportable as JSON, with rollback archiving removed tables rather than destroying them. Failed-run replay is time-limited by plan and unrecoverable afterwards, and restoring a single workflow is capped at 30 days. It ranks below n8n because the retry table’s numbers are referenced but never printed, concurrency limits are withheld, and no dollar figure is published for any tier, so none of those retention windows can be priced.
- Four documented error modes, replay of failed runs, and commit-style project versioning with a per-version record of who changed what.
- No price is published for Pro, Team, Enterprise, any add-on or either support plan — the entire pricing page is demo-gated, and the FAQ contradicts the trial page on whether the trial needs a sales call.
4. Microsoft Power Automate
Why it sits at 4 Versions purge at 6 months draft, 12 months published
Retries are configured per action with a choice of fixed or exponential intervals, the maker setting the initial interval and the maximum retry count. Failure paths are branch-based through Run after, which fires a different path on failure, timeout, skip or success of the preceding action. There is no native try/catch — Microsoft documents it as a pattern assembled from Scope actions and run-after conditions — and automatic failure alerts go to flow owners.
The version history is where it loses ground. It exists only for solution-aware cloud flows stored in Dataverse, so ordinary My flows get none at all. Where it does exist, versions are identified only by timestamp with no diff, notes or filtering, drafts are purged after six months and published versions after twelve, and exporting a solution carries only the last published version. Failed actions and retries bill against the same Power Platform request allowance as successful ones.
- Per-action retry policy plus four Run after conditions gives finer failure-path control than any other platform here that publishes a list price.
- Flows built outside a solution get no version history whatsoever, and Microsoft publishes no default retry count or interval to fall back on.
5. Relevance AI
Why it sits at 5 Versions kept indefinitely; restore lands in draft, not live
Version history here is the strongest published on the list. It covers Agents, Tools and Workforces, every publish is captured automatically, versions are retained indefinitely by default, and restoring an older version is non-destructive — it lands in draft and must be published deliberately. Rollback is marketed as a first-class feature rather than buried in a help article.
Retries, by contrast, are only implied. Error handling is surfaced through configurable alerts on failure statuses, including a retry-exhaustion state that confirms the platform retries at all, with alerts routed to email, Slack or Microsoft Teams. But the vendor never publishes how many attempts precede exhaustion, nor whether the policy can be configured, and per-tier concurrency limits are withheld too. The Free plan is retired and closed to new signups, with no trial published, so a buyer cannot check any of it without paying.
- Indefinite version retention with a genuinely non-destructive restore, and failure alerts you can route to email, Slack or Teams.
- The retry count behind “Exhausted retries” is never published, and with the Free plan retired for new signups and no trial offered, there is no free way to look.
6. Zapier
Why it sits at 6 Version history: none on Free, 1 month on Professional
Zapier’s retry mechanism is Autoreplay, and the plan gate is published explicitly: unavailable on Free, available on Professional, Team and Enterprise. The catch is that it is an account-wide setting rather than a per-step one. Per-Zap overrides — always replay, never replay, and keeping a Zap running despite a high error rate — are restricted to the top Company/Enterprise plan, so granular control is the most expensive thing on the menu.
Version history is the better half of the story: a version is created on every publish and retention is published per plan at none on Free, 1 month on Professional, 6 months on Team and 1 year on Enterprise, with rollback a paid-plan feature and custom error notifications listed across all paid plans. The failure economics are favourable too, since a task is only counted on success. What is missing is the number that matters: Zapier does not publish how many times Autoreplay retries, or over what interval.
- A task is counted only on success, so a flapping integration replays without consuming the allowance.
- Retry control is account-wide unless you are on the top Company/Enterprise plan, and the Free tier has neither Autoreplay nor any version history.
7. MindStudio
Why it sits at 7 Retries set per block; no retention or limit numbers published
Error handling is configured per workflow block, choosing between failing the whole run on the first error or skipping failed items and continuing. Retries are user-configurable per block, and the vendor is upfront that retries consume paid model usage — every headline price on the page carries a literal “+ usage” suffix, the Free plan included, because model tokens are billed on top at provider cost on every tier.
Version history is built in and unusually detailed for a tool at this price: a Versions tab tracks drafts and published releases with version name, publisher, timestamp and an AI-generated change log, and you can roll back to an earlier published version — though restoring discards the current draft. It ranks here because almost nothing is quantified: no version-history retention, no per-tier rate limits, no concurrency caps, no execution timeouts and no execution log retention are published anywhere.
- Retry and failure behaviour are set per block rather than per account, and each published version carries an author, a timestamp and a change log.
- Retries bill model tokens at provider cost on every tier including Free, and no retention, concurrency or timeout figure is published to size the risk against.
8. Dify
Why it sits at 8 Three node error modes published; no retry page in the docs
Dify publishes three node-level error-handling behaviours on LLM, HTTP, Code and Tool nodes, exposes two error variables to workflow authors for building recovery and alerting logic, and documents three failure modes for iteration nodes (loop nodes abort on any child failure). Log history is 30 days on the free Sandbox and unlimited on both paid tiers, which is generous for the money.
Retries are simply absent. The current Handle Errors documentation covers None, Default Value and Fail Branch only, and no retry page exists in the docs navigation — so the fail branch is the whole answer. Version history is limited to Chatflow and Workflow apps, restoring a previous version is destructive to the working draft, restore and version pinning are paid-plan features while Sandbox always runs the latest publish, and the pricing page separately flags app-level version control as not yet shipped on either paid cloud tier.
- A dedicated fail branch and two error variables let you build the recovery path on the canvas, with unlimited log history on both paid tiers.
- No retry setting of any kind is documented, and restoring a previous version overwrites the working draft you were editing.
9. Airtable (Airtable AI / Omni)
Why it sits at 9 No retry: a failed run notifies one person and still burns a run
Airtable does publish what happens when an automation fails, and the answer is a notification to one person. No automatic retry or backoff behaviour is published anywhere. Worse for the budget, a run is counted at trigger time, so failed automations still consume the workspace allowance — 100 runs a month on Free, 25,000 on Team, 100,000 on Business.
What it does publish well is history. Automations carry their own version history showing previous versions of an automation and who changed it. Automation run history retention is tiered at 2 weeks on Free, 6 months on Team, 1 year on Business and 3 years on Enterprise Scale, and base-level revision and snapshot history runs 2 weeks on Free and 1 year on both Team and Business. That is a strong record for the post-mortem and no help at all at 2am.
- Run history retention is published for every tier, up to 3 years on Enterprise Scale, alongside per-automation version history with an author.
- No retry or backoff is published at all, and because a run is counted at trigger time, every failure still spends part of the monthly allowance.
10. Notion AI
Why it sits at 10 A failed automation is disabled until a human re-enables it
Notion’s published error handling is fail-and-pause: a failing automation is disabled and needs manual reactivation. There is no published retry policy, no backoff, and no run or error log described anywhere — so the failure is invisible until someone notices the work stopped. Automations also cannot be chained, which caps how much of a process can live here in the first place, and Free-plan automations are limited to Slack notifications.
The only rollback mechanism published for workflow-driven changes is page history, tiered by plan at 7 days on Free, 30 days on Plus, 90 days on Business and unlimited on Enterprise. To its credit, Notion publishes the clearest commercial terms on this list: a full, non-pro-rated refund if you downgrade within three days of signing up for monthly billing or within 30 days for annual, and 14 days for monthly subscriptions in the EU and UK. That does not move it up a ranking about failure, but it is worth knowing.
- Refund terms are published in full, and page history gives a plan-tiered way back to what the automation changed.
- A paused automation stays paused until a person spots it, and no run or error log is published to spot it with.
Which one fits your situation
Make. The backoff ladder, the five named error-handler directives and the 7/30/60-day log retention are all published. Start on Free at $0 /mo with 1,000 credits, 2 active scenarios and 7 days of execution log storage — for reading the behaviour, the free tier is enough.
Power Automate Premium at $15.00 user/month, paid yearly — but move the flows into a solution before you build anything important. Microsoft publishes version history only for solution-aware cloud flows stored in Dataverse; ordinary My flows get none, and no upgrade fixes that.
n8n Business at 667€/mo, billed annually, which is listed as self-hosted and includes version control using Git. If that figure is out of reach, the Community Edition is free and self-hosted, but n8n publishes no execution allowance or feature ceiling for it, so size it yourself.
Relevance AI Pro at $19 / month annual, $29 / month monthly. Versions of Agents, Tools and Workforces are retained indefinitely by default and restoring is non-destructive: the old version lands in draft and must be published deliberately.
Do not buy anything yet. Both publish that they do not retry — Airtable sends a notification to one person, Notion disables the failing automation until someone reactivates it. No tier upgrade changes that behaviour, so the fix is moving the job to a platform that documents retries, not paying more for this one.
What this market gets people for
- Budgeting as though a failed run is free. Power Automate bills failed actions and their retries against the same Power Platform request allowance, so a flaky integration burns quota twice. Airtable counts an automation run at trigger time, so failures eat the 100 runs a month on Free or the 25,000 on Team. The opposite is also published, and worth using: Zapier counts a task only on success, Tray states automatic retries are not billed as tasks, and Make states error handler modules are not billed against your credit allowance.
- Reading “automatic retries” on a feature list as a retry policy. Five of the ten platforms here list retries as a feature and none of those five publishes the numbers. Zapier does not publish how many times Autoreplay replays or over what interval. n8n does not publish the default max-tries or wait-between-tries. Power Automate says the maker sets the initial interval and maximum retry count but publishes no defaults. Tray’s docs reference a table of maximum attempts and backoff logic without printing a single value. Only Make publishes the ladder.
- Starting on the entry tier that quietly removes your rollback. Zapier’s Free plan is documented as having no version history and no Autoreplay, and rolling a Zap back to an earlier version is a paid-plan feature only. Dify’s Sandbox always runs the latest publish, with restore and version pinning paid-only. And where restore does exist, read how it behaves before you rely on it: restoring a previous version in Dify is destructive to the working draft, and MindStudio’s rollback discards the current draft.
- Comparing an annual per-month figure against a genuine monthly price. Zapier’s Professional is “Starting from $19.99 /month” with the yearly toggle on, and $29.99 /month billed monthly. Airtable’s Team is $20 per seat /month billed annually but $24 per collaborator/ per month billed monthly. Notion’s Plus is $10 per member / month yearly and $12 per member / month monthly. n8n publishes only the annual per-month figure, in EUR, and does not publish a monthly-billed price at all — so there is nothing to compare it against.
Questions buyers ask
Which platform actually publishes how many times it retries?
Make, and only Make. It documents automatic retries via exponential backoff for connection and module-timeout errors, escalating across up to 8 attempts at 1 min, 2 min, 5 min, 10 min, 1 hour, 3 hours, 12 hours and 24 hours before scheduling is disabled. Zapier’s Autoreplay, n8n’s Retry On Fail, Tray’s per-error-type backoff, Power Automate’s retry policy and Relevance AI’s retry-exhaustion state are all named without a published count or interval.
Do failed runs cost me money?
It depends who you ask, and all of this is published. Zapier counts a task only on success, so failed actions do not consume the allowance. Tray states automatic retries are not billed as tasks. Make states error handler modules are not billed against your credit allowance. Against that, Power Automate bills failed actions and retries against the same request allowance, and Airtable counts a run at trigger time, so a failed automation still burns one.
Which platforms publish that they do not retry at all?
Airtable, whose published error handling for automations is a notification to one person rather than an automatic retry; Notion, where a failing automation is disabled and needs manual reactivation; and Dify, whose current Handle Errors documentation covers only None, Default Value and Fail Branch, with no retry page in the docs navigation. Gumloop, which is not ranked here, publishes an auto-disable circuit breaker for triggers with manual re-enable instead of a retry policy.
Can I roll a workflow back to a previous version?
Relevance AI retains versions indefinitely by default and restore lands in draft rather than going live. Tray treats project versions like commits, restorable or exportable as JSON, though restoring a single workflow is capped at 30 days. n8n publishes 1 day of workflow history on Starter, 5 days on Pro and 365+ days on Enterprise, with Git version control from Business. Zapier publishes none on Free, 1 month on Professional, 6 months on Team, 1 year on Enterprise. Power Automate has it only for solution-aware flows in Dataverse. Make publishes no scenario version history at all.
How long are execution logs kept?
Make publishes 7 days on Free, 30 days on Core, Pro and Teams and 60 days on Enterprise, with full-text log search from Pro up. n8n publishes 7-day log retention on Starter and 30-day on Pro. Tray’s failed-run replay expires after 7 days on Pro and Team (30 as a Team add-on) and 30 days on Enterprise, after which the data is unrecoverable. Dify gives 30 days on Sandbox and unlimited on both paid tiers. Airtable retains automation run history 2 weeks on Free, 6 months on Team, 1 year on Business and 3 years on Enterprise Scale.
If the error handling turns out to be wrong for me, can I get a refund?
Mostly no, and several vendors publish the opposite. Zapier states “All purchases are considered non-refundable”. n8n states “n8n does not offer pro-rata refunds.” Bardeen’s terms say “All sales are final.” Notion is the exception with a real window: a full, non-pro-rated refund if you downgrade within three days of signing up for monthly billing or 30 days for annual, plus 14 days in the EU and UK. Airtable publishes courtesy refunds only, within 7 days of a monthly invoice or 14 days of an annual one. Make, Dify, MindStudio, Relevance AI, Tray and Power Automate publish no refund window at all.
What none of them publish
These are the questions a buyer would want settled that no vendor on this page answers in public. If one of them decides your purchase, ask the vendor and hold them to the answer.
- The retry count and interval for every platform except Make. Zapier does not publish how many times Autoreplay replays or over what schedule; n8n publishes no default max-tries or wait-between-tries; Power Automate publishes no default retry count or interval; Tray’s docs reference a table of maximum attempts and backoff logic but print no numbers; Relevance AI names an “Exhausted retries” state without saying how many attempts precede it.
- Whether a failed execution counts against n8n’s monthly workflow-execution allowance. The billing unit is the whole execution, but the pricing page does not say whether failures are metered.
- How many versions any platform keeps. Tray, Dify, MindStudio and Power Automate publish expiry windows or behaviour but no version-count cap; Power Automate publishes only 6 months for drafts and 12 for published versions.
- Tray.ai’s price. No dollar figure is published for Pro, Team, Enterprise, the Agent Development add-on, any individual add-on or either support plan, and no task allowance per tier is published either — entitlement is contract-specific.
- n8n’s monthly-billed prices, and the length of its free cloud trial in days. The page renders annual per-month figures in EUR only.
- What overage costs when you exceed the allowance. Zapier publishes only a multiplier of the base rate, not a per-task figure; Make publishes no per-credit overage rate; Airtable says extra AI credits can be bought but publishes no price; Tray publishes nothing beyond alerts at 80% and 100%.
- Concurrency limits. Relevance AI says concurrency is determined by subscription tier without publishing a number for any tier; Tray directs customers to support instead of publishing a default.
- Refund terms for most of the list. Make, Dify, MindStudio, Relevance AI, Tray and Power Automate publish no refund or money-back window at all — Power Automate’s pricing page has no FAQ, terms link or cancellation language on it.
- What happens to Flowise’s existing paid cloud subscriptions after its published End of Life date of August 31, 2026. The sunset notice covers the open-source repository only and is silent on the cloud service and billing, while the paid tiers are still displayed.
- Any published uptime or SLA figure. Zapier’s pricing page states none, and MindStudio describes SLAs only as “Custom SLAs & MSAs” on the contact-priced Business tier.
Reading this from the other direction? Zapier Alternatives, Ranked by What One Unit of Work Actually Costs covers the same market against a different rule.