The true cost of automation platforms, measured: a 5–8× spread on the identical workload

Running the same workflow basket 1,000 times a month costs $73.50 on Zapier, $9 on Make, and $0–5 on self-hosted n8n - and that 5–8× Zapier-to-Make spread holds at every volume tier we priced, from 100 to 30,000 runs a month (July 2026, monthly billing, list prices read from our own paid accounts).

Most cost comparisons quote pricing pages. This one doesn't: we run the identical workflows on each platform around the clock, read what the billing meters actually charge, and verified our consumption model against those meters - exactly (45 = 45 tasks) on Zapier, within 1% on Make. Then we priced the ladders from inside our own accounts. Every number below is traceable.

Cost vs volume (the number that matters)

Monthly cost of running the disclosed workflow basket N times per month, choosing the cheapest sufficient plan tier each time:

Curve stops where our first-party-verified price ladder stops - we don't extrapolate prices we haven't read from a real account. Make's annual prices weren't exposed in our account UI, so we only publish its monthly column. n8n self-hosted: $0 platform fees; we run it on a cloud always-free tier VM ($0), a typical small VPS is ~$5/month.

Why the gap: billing models, measured

The same basket run consumes different billable units by construction of each platform's meter:

Two folklore-killers from the measurement: filtered runs do not bill on Zapier (9 of our filter-stopped runs consumed 0 tasks, contrary to a widely repeated claim), and editor-mode test runs are free on both platforms. Both facts came out of reconciling our run ledger against the vendors' own meters.

The standing tax nobody prices in: polling

One of our four workflows is a 30-minute scheduled poll. On Make, every empty poll - the scheduler fires, finds nothing new, does nothing - still consumes 2 credits (the poll request plus the state lookup its dedup logic needs). At a 30-minute cadence that is ~2,880 credits a month for zero work, more than half of Make's entry paid tier before your workflow does anything. The identical poll on self-hosted n8n costs $0. Per-operation platforms charge you for checking, not just for doing - polling-heavy workloads should price this first.

We then ran into the consequence instead of estimating it: Make's free tier cannot keep one scheduled workflow running for a month. The free plan allows 1,000 credits. At 2 credits per empty poll, a single 30-minute poller spends them in about eleven days, and the scenario stops until the monthly reset. Ours has been going dark for roughly nineteen days of every month since August 2026, which we did not notice until October, because a poller that stops does not raise an error. It is simply not there. The measured coverage and the exact dark windows are published alongside the reliability data; the same workflow on self-hosted n8n has run at full coverage over the same period for the cost of the box it sits on.

The practical reading for a buyer: on a per-operation platform, a scheduled check is not a rounding error you can leave running while you evaluate. Price the checking before the doing, and if you are trialling on a free tier, expect the trial to end partway through the month without telling you.

Free tiers: what the pricing page doesn't say

• Zapier Free allows 2-step Zaps only - it cannot run our basket at all (two of the four workflows need more steps), so it does not appear on the curve.
• Make Free (1,000 credits) allows a maximum of 2 active scenarios - enough to sample, not enough to host the full basket plus polling.
• Pipedream Free ran our workflows flawlessly until its credits ran out - then its webhooks kept returning success while silently discarding every delivery. We measured that failure mode; if production data flows through a metered free tier, read it before trusting "generous free tier" framing.

Method and provenance

• The basket (disclosed): equal thirds of three webhook workflows - single action; filter then two actions (50% of events pass the filter); two parallel branches. Identical step-for-step on every platform; both endpoints under our control. Details and delivery statistics are on the reliability scoreboard.
• Consumption is measured from our reconciled run ledger and verified against each platform's billing meter (Zapier exact; Make within 1%, the residue being cycle-boundary minutes).
• Prices are list prices read from inside our own paid accounts on 2026-07-05 (plan-change sliders), anchored to our actual bills. Zapier annual prices are monthly ÷ 1.5, a ratio verified on two rungs. We paid for both subscriptions ourselves.
• Limitations: prices are USD list and change; this page is dated and re-checked when re-published. Consumption is for our disclosed basket - a loop-heavy or AI-step workload will differ (Make bills AI features differently, for example). n8n costs exclude your time to operate it, which is real but not a platform fee.

FAQ