Zapier vs Make vs n8n: When to Switch and How to Migrate (2026)

You already have Zaps, or a Make folder a freelancer left open, or an n8n Cloud workspace someone stood up on a Friday. The question is not which tool is “best” in 2026. The question is who bills you, who can fix a red run, and whether you will still understand the graph in six months.

Stay on Zapier if volume is modest, the owner is not a developer, and you will pay for Autoreplay. Pick Make if you think in routers and you want cheaper credits than Zapier tasks. Pick n8n if one workflow run should cost one execution no matter how many steps it has, and someone on your team can own the host or the Cloud plan.

Do not migrate because a thread told you to. Migrate because the bill, the failure mode, or the person who can maintain it no longer matches the tool. If you inherited a stack and nobody has a map, start with inventory. That map is Takeover. A coming sibling: Inherited automations: the audit.

If you still need the beginner definition, read What Is AI Automation for Small Business? once, then come back.

The short answer

Stay on Zapier if

You are under a few thousand successful action steps a month, the owner can click through a Zap, and you need the 9,000+ app catalog. Stay if you will pay for Autoreplay on Pro, Team, or Enterprise — Free is manual replay only. Stay if 60 days and 10,000 runs of history is enough. Stay if you are not putting an agent on invoices, contracts, or customer secrets.

Leave when extra action steps, MCP tool calls, or Lead Router blow the task math, or when nobody can explain a Zap three contractors have edited.

Pick Make if

You live in branching. Router and error-handler modules cost 0 credits. Most other modules cost 1. That is a different bill from Zapier, where every successful action step is a task and error-handler path steps count too.

Pick Make if a 1-minute schedule and a 40-minute max run on paid plans matter, 30-day logs are enough, and 3,000+ apps (350+ AI) cover your stack. Pick Make if you will bring your own OpenAI or Anthropic key on a paid plan, so an AI module is 1 credit plus the provider invoice.

Do not pick Make because it looks cheaper and then paste Claude through Make’s own provider on every lead. Token credits add up. Make’s agent guide says to avoid agents for sensitive data, high-stakes finance, and strict legal work. Believe that page.

Pick n8n if

You want the bill to follow runs, not steps. One execution is one full workflow run, any number of nodes. Unlimited users, workflows, and integrations sit on every plan.

Pick Cloud if you will stay inside Starter or Pro caps and you accept Frankfurt hosting and that plan’s retention. Pick Business self-hosted if you need 40,000 executions and Community’s missing features. Pick Community only if the Sustainable Use license fits — own internal business or non-commercial — and someone will own upgrades, queue mode, and secrets.

Do not pick n8n because it is “free.” Community has no license fee. It is not free of a person. You lose SSO, Git version control, environments, projects, sharing, log streaming, external secrets, and multi-main. Queue mode is included. Commercial redistribution needs a deal. If nobody can own HTTP and a container, n8n is a new on-call rotation.

How each platform bills you

Read the unit before you read the price. Zapier sells tasks. Make sells credits. n8n sells executions. The same five-step flow is five tasks, five or six credits, or one execution.

Zapier bills tasks

A task is a successful action step. Triggers do not count. Filter, Paths, Formatter, Delay, Looping, Sub-Zap, Digest, Storage, Tables, and Forms do not count. Error-handler path steps do count. If you replay a run, already-successful steps count again.

MCP is 2 tasks per successful tool call. Lead Router is 5 tasks per lead. Code is 1 task at standard runtime. Zapier published that measurement on 21 August 2026: How is task usage measured in Zapier.

Free is 100 tasks, two-step Zaps, 15-minute polling. Pro yearly starts at $19.99/month. Team yearly starts at $69/month for 25 users. List prices on zapier.com/pricing:

Pro monthly / yearly: 750 tasks $29.99 / $19.99; 2,000 $73.50 / $49; 5,000 $133.50 / $89; 10,000 $193.50 / $129; 50,000 $433.50 / $289; 100,000 $733.50 / $489; 300,000 $1,599 / $1,069.

Team monthly / yearly: 2,000 $103.50 / $69; 10,000 $253.50 / $169; 50,000 $598.50 / $399; 100,000 $898.50 / $599.

Overage is pay-per-task at a higher rate, capped at 3× the plan limit, then held. If pay-per-task is off, you are held at the limit.

Autoreplay exists on Pro, Team, and Enterprise only. The owner or super admin turns it on account-wide. Free is manual replay only. See Autoreplay.

History is at most 60 days and up to 10,000 runs. Custom retention is Enterprise-only. See Zap history.

Agents are a separate meter. They bill in activities: Free 400/month; Pro $33.33/month on yearly ($400/year) for 1,500 activities. AI-by-Zapier model-tier task rates were not published in a table we could verify. Do not forecast an AI step as one task.

Make bills credits

Credits replaced operations on 27 August 2025. Most modules cost 1 credit per operation. Router modules and error-handler modules cost 0. The Code app costs 2 credits per second. Make documents this in Credits (updated 23 September 2026).

On paid plans, if you use your own OpenAI or Anthropic key, the AI module is 1 credit per operation and you pay the provider. If you use Make’s AI provider, credits are token-based. Claude Sonnet 4.6 is about 301 input tokens or 60 output tokens per credit. GPT-5 mini is about 3,616 input tokens or 452 output tokens per credit. Make marks those rates as subject to change.

Free is 1,000 credits, 2 active scenarios, a 15-minute schedule, a 5-minute max run, and 7-day logs.

Core, Pro, and Teams start at 10,000 credits, a 1-minute schedule, a 40-minute max run, and 30-day logs. Enterprise is 60-day logs plus overage protection.

We confirmed the live list prices from the 4 September 2026 archive of Make’s official pricing page, because the live page blocks bots. Re-check https://www.make.com/en/pricing before you buy. Archived monthly / yearly:

10,000 credits — Core $10.59 / $9; Pro $18.82 / $16; Teams $34.12 / $29.

40,000 — Core $34.12 / $29; Pro $62.35 / $53; Teams $116.47 / $99.

80,000 — Core $64.71 / $55; Pro $107.06 / $91; Teams $203.41 / $172.90.

150,000 — Core $116.47 / $99; Pro $180.53 / $153.45; Teams $343.01 / $291.56.

300,000 — Core $214.31 / $182.16; Pro $315.93 / $268.54; Teams $600.26 / $510.22.

When you run out of credits, scenarios stop. Webhooks queue until storage is used. Extra credit packs come in 1,000 or 10,000. Auto-purchase is available on Core, Pro, and Teams.

Incomplete executions are off by default. Turn them on if you want Make to keep a failed run so you can retry it. See Incomplete executions. Error handlers are Retry, Resume, Commit, Rollback, and Skip, documented in error handlers. The pricing FAQ still says Break and Ignore. The names do not match. Plan against the error-handler page, not the FAQ wording.

n8n bills executions

An execution is one full workflow run, any number of steps. You pay for runs, not nodes.

All plans include unlimited users, workflows, and integrations. Cloud prices on n8n.io/pricing: Starter €24 / €20 monthly / yearly for 2,500 executions. Pro €60 / €50 for 10,000, or €145/month for 50,000. Business self-hosted is €800 / €667 for 40,000, up to €5,245/month for 500,000. Overage on Business is €4,000 per extra 300,000 executions. Start-up is 50% off Business under 20 employees. Enterprise is custom.

Retention is not the same as the execution cap. Starter keeps 2,500 executions of history, 7 days, and 1-day version history. Pro keeps 25,000 / 30 days / 5-day versions. Enterprise keeps 50,000 / unlimited retention / 365+ versions. Hitting the history limit trims history. It does not stop workflows. Cloud is hosted in Frankfurt.

Community self-host has no license fee. You do not get SSO, Git version control, environments, projects, sharing, log streaming, external secrets, or multi-main. Queue mode is included. The license is Sustainable Use / fair-code: own internal business or non-commercial use. Commercial redistribution needs a deal. Read the fair-code license and LICENSE.md.

Error handling is an error workflow you assign per workflow. It starts with an Error Trigger. See error handling.

The AI Agent node is the Tools Agent since v1.82. The v1 node was removed in n8n 3.0. Read AI Agent. Expect 2,000+ integrations (the page showed 2,274; say 2,000+).

There is no hidden Free 50-execution Cloud plan in the public pricing we used.

Comparison table

If you cannot name who maintains it, the cheapest row is a trap.

Worked monthly cost at 3 volumes

Assumptions, labeled so you can replace them. Sample flow: form trigger → enrich → CRM upsert → Slack → Sheets (trigger + 5 actions).

Zapier: 5 tasks if the actions succeed. Make: 6 credits if Make bills the trigger module, 5 if not — verify in your history; we price the pessimistic 6. n8n: 1 execution.

No Paths, loops, MCP, Lead Router, Code-by-the-second, or Make-hosted LLM in the base case. Yearly prices are the prepaid monthly equivalent each vendor publishes. Add maintenance hours. The tool bill is not the cost of the system.

1,000 runs a month

Zapier: 5,000 tasks. Pro 5,000 is $133.50/month or $89/month yearly.

Make: 6,000 credits. Core 10,000 is $10.59/month or $9/month yearly.

n8n: 1,000 executions. Starter 2,500 is €24/month or €20/month yearly.

Add 2–4 hours of a person’s time to read failures, replay, and explain a field change. The hours can still exceed every tool bill.

10,000 runs a month

Zapier: 50,000 tasks. Pro 50,000 is $433.50/month or $289/month yearly.

Make: 60,000 credits. Core 80,000 is $64.71/month or $55/month yearly.

n8n: 10,000 executions. Pro 10,000 at €60/month or €50/month yearly sits on the cap. Price the next step: Pro 50,000 at €145/month.

Add 4–8 hours of operator time. Autoreplay and incomplete executions only help if a human owns the alert.

50,000 runs a month

Zapier: 250,000 tasks. There is no 250,000 row. You buy Pro 300,000 at $1,599/month or $1,069/month yearly.

Make: 300,000 credits. Core 300,000 is $214.31/month or $182.16/month yearly.

n8n: 50,000 executions. Pro 50,000 at €145/month is at the cap. Or Business at €985/month for 50,000 on this sample (public list starts at €800 / €667 yearly for 40,000, up to €5,245/month for 500,000). Or Community self-host at $0 license plus server plus the hours of the person who patches it.

Add 8–16 hours of operator time if this stack has no Care plan. A silent failure here is a week of bad CRM rows.

Add one LLM step

Keep the same 5-action flow and add a classify-or-draft step.

Make with your own key, paid plan: +1 credit per run, plus the OpenAI or Anthropic invoice. At 1,000 runs that is 1,000 credits and a token bill you control.

Make-hosted Claude Sonnet 4.6: about 10 credits for an illustrative ~1,500 input / 300 output tokens (301 input or 60 output tokens per credit, rates subject to change). At 1,000 runs that is ~10,000 extra credits — a full Core 10,000 plan. At 10,000 runs it is ~100,000 credits. Check Credits.

n8n: no extra executions. You pay the model provider.

Zapier: AI-by-Zapier model-tier task rates are unknown / not published in a table we could verify. Agents are a separate activity meter. Do not forecast a made-up task count.

Add maintenance hours

Always. Twenty live flows with no owner cost more than a $289 Zapier year line. A Community n8n box with no disk alerts costs more than Pro Cloud. If you want a third party on failures, Care is $1,000/month for maintenance, not a rewrite. New flows and migrations are a separate quote.

For what is worth a person’s time, use the AI Automation Audit and the hour math in How to Automate 40 Hours a Week.

Reliability in production

Price is the happy path. Reliability is the Friday the mapping is wrong.

History you can actually use

Zapier: at most 60 days, 10,000 runs. Custom retention is Enterprise-only. A quarterly Zap you debug in week 10 may already be gone. Zap history.

Make: 7 days on Free, 30 on Core/Pro/Teams, 60 on Enterprise. A month-end recon that failed on day 1 is gone on Core before finance asks on day 32.

n8n Cloud: hitting the history limit trims history. It does not stop workflows. Starter is 2,500 executions / 7 days. Pro is 25,000 / 30 days. Community self-host: you own retention and backups. Export the runs you need for audit before you need them.

What happens when a run fails

Zapier without Autoreplay stays failed until a person replays. Autoreplay is Pro/Team/Enterprise, account-wide, owner or super admin. Autoreplay. Replay re-counts already-successful steps.

Make incomplete executions are off by default. A failed scenario does not wait unless you turned that on. Incomplete executions. Handlers: Retry, Resume, Commit, Rollback, Skip. Router and error-handler modules are 0 credits — put them on. Do not skip handlers to “save operations” from 2024.

n8n: assign an error workflow that starts with an Error Trigger. If you do not, a failed run is a row in executions and a hope. Error handling.

None of the three pages you when a freelancer still owns the login.

What happens when you run out of quota

Zapier: pay-per-task at a higher rate up to 3× the plan, then held. Pay-per-task off: held at the limit. Held means the Zap stops. The form still submits. The CRM does not update.

Make: scenarios stop. Webhooks queue until storage is used. Then a 1,000 or 10,000 credit pack, or you wait. Auto-purchase on Core/Pro/Teams avoids a weekend stop and can surprise the card.

n8n Cloud history trim is not a stop. Running past the paid execution allotment is a plan problem. Read n8n pricing for your exact plan. Do not assume Community behavior on Cloud.

Who gets the Slack when any of that happens? If the answer is nobody, the platform does not matter.

Where AI agents and LLM steps fit

An LLM step that classifies or drafts is not an agent. An agent that picks tools at runtime is. Bill them differently. Trust them differently.

Zapier Agents

Agents bill in activities, not in the Zap task table. Free: 400/month. Pro: $33.33/month yearly ($400/year) for 1,500 activities. Second meter.

AI-by-Zapier model-tier task rates are unknown. If a Zap step calls a Zapier-hosted model, do not forecast one task. Run a week and read usage.

Do not put a Zapier Agent on refunds, contracts, or anything you would not let a new hire send unsupervised. 9,000+ apps make a write path easy.

Make AI and agents

Own key, paid plan: 1 credit for the module, tokens on the provider invoice. Use Make’s models only if you have accepted token credits and “subject to change.”

Claude Sonnet 4.6 at ~301 input or 60 output tokens per credit will eat a Core plan on long threads. GPT-5 mini is cheaper on Make’s table (~3,616 input or 452 output per credit) and still not free. Credits.

Make’s agent intro: avoid agents for sensitive data, high-stakes finance, and strict legal. Introduction to Make AI Agent. If counsel would not let a contractor do the action, do not give the agent the credential.

n8n AI Agent

Tools Agent since v1.82. The v1 node is gone in n8n 3.0. AI Agent.

Three tool calls do not add executions. You pay the model provider. That is why n8n looks cheap on agent-heavy graphs and expensive in people-time when the agent loops.

Self-host does not make the agent safer. It makes the logs yours. You still need a human on write actions.

Where an LLM step belongs

Draft, classify, extract, summarize. Put a human or a hard rule in front of send, charge, delete, and legal text.

A fixed path — form, enrich, upsert, notify — does not need an agent. Agents belong when the next tool is not known at design time. That is rare in a 20-person ops stack.

For sequencing work that returns hours, use How to Automate Your Business with AI. Do not start with an agent because the vendor homepage leads with one.

Decision framework: 6 questions + scoring grid

Score 1 (Zapier-shaped), 2 (Make-shaped), or 3 (n8n-shaped). Sum. Then apply the overrides. Do not average away a 3 on data sensitivity.

The six questions

1. Who can maintain this in six months? Ops lead who should not SSH: 1. Can read a Make scenario: 2. Can own Cloud or a box: 3.

2. What is the honest monthly volume in action-steps, not “Zaps”? Under ~5,000 successful Zapier tasks: 1. Heavy branching where routers should be free: 2. Tens of thousands of multi-step runs that should be one execution: 3.

3. How much of the graph is branching, waiting, or long-running? Linear two-to-five steps: 1. Routers and error routes you will actually draw: 2. Long runs, queues, or many nodes: 3. Make’s paid 40-minute max run and 1-minute schedule matter here. Zapier Free is 15-minute polling and two-step Zaps only.

4. What data moves? Marketing lists and public form fields: 1. Internal ops, not regulated: 2. Customer secrets, money, or legal: 3 — then the override. You may still stay on Zapier or Make. You do not give an agent a write token.

5. Do you need to leave later? Fine in a SaaS you cannot export as runnable code: 1. Screenshot and rebuild is enough: 2. You need a repo, queue mode, and a license you have read: 3. Community n8n is not a redistribution license.

6. Who gets paged when a run dies? Nobody, and you will not buy Autoreplay or turn on incomplete executions: stop. Fix ownership before you score. Named human plus Zapier Autoreplay: 1. Plus Make handlers: 2. Plus an n8n error workflow: 3.

Scoring grid

Overrides (they beat the sum):

• No named owner → do not migrate. Inventory and alerts first.
• Agent on finance or legal → do not “pick n8n to be safer.” Remove the write path. Make already says this on Make; apply it everywhere.
• Commercial redistribution of n8n → you need a deal.
• Under 1,000 clean runs/month and Zapier works → stay. Migration hours will not pay back.
• 50,000 runs/month on Zapier tasks → do the cost table again before you call it “just how Zapier is.”

Write the six answers in the runbook.

When the bill explodes: fix before you migrate

Most “we have to leave Zapier” tickets are a billing shape, not a platform ceiling. Fix the shape. Then decide.

Count tasks, not Zaps. Five successful actions is five tasks. Paths that run extra action steps are extra tasks. Filters are free. The actions after a filter that still fire are not.

Stop blind replay. Replay re-counts successful steps. Autoreplay is correct for transient failures. It is expensive if a bad mapping fails and you replay 800 runs.

Price the specialty steps. MCP is 2 tasks per successful tool call. Lead Router is 5 tasks per lead. Code is 1 task at standard runtime.

Turn pay-per-task into a decision. 3× the plan then held is a budget line. Held at the limit is a reliability line. Pick one on purpose.

On Make, kill Code-by-the-second in a loop. 2 credits per second is how a “small script” becomes a pack purchase. Prefer a module. Cap runtime if you need Code.

On Make, do not default to Make-hosted long-context models. Own-key + 1 credit is the predictable pattern. Sonnet 4.6 token credits are a different product.

On Make, turn incomplete executions on for production scenarios you intend to retry. Off is how a failure becomes “it never ran.”

On n8n, do not treat Pro 10,000 as a comfort zone. You are on the cap. The realistic paid step at 10,000+ is Pro 50,000 at €145/month.

Delete the zombie. Duplicate Zaps, old webhooks, a test scenario still on. Inventory first. The AI Automation Audit is the scoring model for whether the work should exist at all.

Add the hours line. If the tool bill dropped $200 and you spent 20 hours rebuilding, you changed vendors. You did not save money.

If the bill is high because the flows are wrong, migrating copies the waste.

How to migrate Zapier → n8n (or Make)

There is no converter you should trust. A Zap export is not an n8n workflow. A Make blueprint is not a Zap. Inventory, rebuild the ones that earn their keep, run both, cut over, write the runbook.

Migrations are a separate quote from Takeover and Care. Do not start from a 12-flow rebuild as the default. Start from the list of flows allowed to stay on.

Inventory (0.5–1 day for 20–40 flows)

Our estimate, for a stack someone still has login to. For every live flow write: tool, name, trigger, destination, last success, last failure, cost last 30 days, who has the login, whether it writes to a customer-facing system, whether an agent or Code step sits on it.

Turn off anything with no success in 30 days and no owner. You are not migrating dead Zaps.

This is the same map Takeover produces: $600–$1,200 for inventory, alerts, and a runbook, then stay / fix / migrate. AI Solutions.

Rebuild times (our estimates)

Simple linear flow (trigger + 3–5 actions, no Code, no agent): 1–2 hours each, including a test record in the destination.

Branching (Paths / Router, filters, two or three destinations): 0.5–1 day each. Most of the time is matching the path Zapier treated as “not a task” to Make routers (0 credits) or n8n IF nodes (still one execution).

AI, Code, or MCP/tool-calling: 1–3 days each. Re-learn prompts, retries, and failure shape. On n8n, use the Tools Agent, not a v1-node tutorial. On Make, decide own-key vs Make-hosted first. On Zapier, do not assume the AI step is one task.

A 30-flow account is not 30 × 2 hours. It is 10 simple, 8 branching, 3 AI, and 9 you will delete. Do the delete first.

Parallel run

Rebuild in the destination. Keep the source live. Send the new side to a shadow Slack, staging sheet, or CRM test pipeline. Compare 50–100 real events. Check field maps, timezone, and “ran twice.” Dedup keys first — duplicate CRM upserts are the usual cutover bug.

Do not parallel-run an agent that sends email. Parallel-run the draft. Cut over the send when the draft matches.

Cutover

1. Confirm Autoreplay / incomplete executions / error workflow on the new side.
2. Point the trigger (webhook, form, mailbox) at the new flow.
3. Turn the old flow off. Do not leave both writing.
4. Watch the first business day with a human on the alert, not “we will check Friday.”
5. Keep the old flow exported or screenshotted for 30 days. You will need a field you forgot.

On Zapier, off is not delete. On Make Free, watch the 2-active-scenario cap. On n8n, inactive workflows should not execute.

Leave a runbook

One page: what it does, where it lives, who owns the vendor login (their email, not a contractor), what Slack channel gets the fail, what to do if X fails, where the cost shows up (tasks, credits, executions, tokens). If you cannot write that page, you have a demo.

When you want a third party to hold that page, Care is $1,000/month: maintenance, not a new platform. Migrations stay a quote. AI Solutions.

When NOT to migrate

The bill is under a few hundred dollars and the flows work. 1,000 runs on Zapier Pro 5,000 yearly at $89/month is a correct answer. Two days of rebuild to “save $80” on Make Core is not a save.

Nobody can maintain the destination. Community n8n with no owner is an expired certificate and a silent form. Make with no one who reads a router is the same Zapier problem in a new UI.

You do not own the accounts. If the last freelancer is still the billing email, migration copies a hostage. Move billing first.

You are mid-incident. Fix the red flow on the current tool. Autoreplay, incomplete executions, or a manual replay. Migrate next month.

The license does not fit. You cannot take Community n8n and sell it as your product without a deal. Fair-code license.

The only goal is “add agents.” Agents are a second meter on Zapier, a token risk on Make, and a Tools Agent plus provider invoice on n8n. Changing platforms does not create a safe agent.

You have no inventory, or you expected a converter. You cannot migrate 40 unnamed Zaps. There is no converter that preserves Paths, error handlers, and Code. Do the half-day map. If you want that map done for you, that is Takeover, not a migration project.

FAQ

Can I export Zaps into n8n or Make? Not as something you should run in production. Document triggers, map fields, rebuild. A one-click convert is a first draft.

Does Make still use operations? No. Credits replaced operations on 27 August 2025. Credits. Ignore 2024 prices that still say operations.

Is n8n free? Is there a 50-execution Cloud plan? Community self-host has no license fee for own internal business or non-commercial use. Cloud is paid from Starter. Business self-hosted is paid. There is no hidden Free 50-execution Cloud plan in the public pricing we used. n8n pricing, license.

Why is Zapier more expensive in the table? Will $19.99/month yearly Pro cover me? The sample flow is 5 tasks per run. Yearly Pro at $19.99 is 750 tasks — 150 runs of that flow. At 1,000 runs you are on Pro 5,000 ($133.50 / $89). At 10,000 runs you are on Pro 50,000 ($433.50 / $289). Make is 60,000 credits if the trigger counts (Core 80,000 at $64.71 / $55). n8n is 10,000 executions — price Pro 50,000 at €145, not the €60 plan on the cap. Task usage.

Should I turn on Autoreplay and Make incomplete executions? Autoreplay: yes on Pro/Team/Enterprise if failures are transient; owner or super admin, account-wide. Bad mappings plus Autoreplay spend tasks. Autoreplay. Incomplete executions: yes on production scenarios you intend to retry; they are off by default. Incomplete executions.

Make says Break and Ignore in one place and Skip in another. Which is it? Error-handler docs: Retry, Resume, Commit, Rollback, Skip. The pricing FAQ still says Break/Ignore. Trust error handlers and test one scenario.

How long does a migration take? Inventory 0.5–1 day for 20–40 flows. Simple flow 1–2 hours. Branching 0.5–1 day. AI/Code 1–3 days. Our estimates, not a vendor SLA. Parallel run and one cutover day on top.

Should I self-host n8n to save the €145? Only if someone owns the box and you have read the license. Community is $0 license plus server plus hours, and it is missing the Cloud/Business features listed above.

Do you migrate stacks as the default engagement? No. We take over what exists, put alerts on failures, and leave a runbook. Takeover is $600–$1,200. Care is $1,000/month. Migrations are quoted when the inventory says they pay back. AI Solutions.

Conclusion

Zapier vs Make vs n8n in 2026 is a billing-unit decision and a staffing decision. Tasks vs credits vs executions. Ops clicker vs graph reader vs someone who can own a host.

At 1,000 clean runs of a 5-action flow, Make Core and n8n Starter are cheap; Zapier Pro 5,000 is the catalog premium. At 10,000 and 50,000 the bills diverge — use the table above, not a screenshot. Add an LLM step only with a known rate. Add maintenance hours every time.

Stay if the current tool matches the person and the volume. Fix the bill shape before you move. Inventory before you rebuild.

If the stack is already there and unowned, do not buy a 12-flow rebuild. Send the tool and what is failing. Takeover ($600–$1,200) is inventory, alerts, and a runbook. Care ($1,000/month) is the month they stay up. A migration, if the map says it is worth it, is a separate quote.

Tell us the tool and what is red.