Zapier, Make, n8n, Workato: the task-pricing trap

A marketing operations manager sent me a Zapier invoice in January with a note: “Is this normal?” It was £740 for a month, up from £69 in September, and nothing about the business had changed except that a form had gone viral for a fortnight. Nothing was broken. The billing model had simply done exactly what it says it does, and nobody had modelled it.

Task-based pricing is the single most misunderstood line in operations budgets. Zapier counts a task for every action step that successfully processes data. Make counts operations, which includes iterator cycles. n8n Cloud counts workflow executions regardless of how many nodes are inside. Those three units are not comparable, and vendors compare them anyway.

The test workflow

I built the same thing five times: webhook trigger, enrichment lookup, conditional branch, an LLM classification call, a CRM write, a Slack notification, an error handler, a retry, a dead-letter path, a daily digest, an audit log write. Eleven steps, 20,000 runs a month, roughly 4 per cent error rate. Deliberately ordinary.

Look at the third column before the fourth. Eleven steps times 20,000 runs is 220,000 billable units on a per-action model and 20,000 on a per-execution model. That is an order of magnitude of difference created entirely by how the meter is defined, before anyone argues about features.

So is Zapier a rip-off? No — and this is where I part company with the n8n evangelists

Zapier has roughly 8,000 app integrations. Nobody else is close; Make is around 2,500, n8n around 1,100 with community nodes filling gaps. If your workflow touches a mid-market SaaS product with an idiosyncratic API, Zapier probably has a maintained connector and the other three do not. A week of engineering time to build and maintain one connector costs more than a year of the difference in subscription. Buy the connector coverage.

Zapier is the right answer when volume is low and integration breadth is high — a revenue-ops team wiring fifteen tools at a few thousand runs a month. It becomes the wrong answer, quickly and expensively, when a single high-volume workflow dominates your usage. The fix is not migration; it is moving that one workflow somewhere metered differently and leaving the long tail where it is.

Make sits in the middle on both price and coverage, and its visual router is the best of the group for genuinely branching logic. Its operations counter still surprises people: an iterator over a 50-item array is 50 operations, not one.

n8n is my recommendation for engineering-adjacent teams. Per-execution billing, a code node when the visual builder runs out of road, and self-hosting under a fair-code licence that is free for internal business use. It is not open source in the OSI sense — the Sustainable Use Licence restricts reselling it as a service — and I wish more coverage said so plainly. Self-hosting means you own upgrades, backups, queue mode and the 3 a.m. page.

Power Automate wins on price-per-user if you are already paying for Microsoft 365 and your work lives in SharePoint, Dataverse and Outlook. Outside that gravity well it is awkward, and the premium connector licensing is a maze.

Workato is enterprise iPaaS: governance, environments, role-based access, audit trails, and a sales process to match. It is genuinely good at what mid-market tools are bad at — multi-team governance at scale — and it is priced for organisations where a failed integration is a compliance incident.

The AI-step problem

Every platform in this list now ships AI nodes and “agent” blocks. Most of them charge you twice: once for the task or operation, and once for the tokens, either via your own API key or a marked-up credit system. Read which. Two of the five bundle inference into opaque credits with no per-token disclosure, which makes cost forecasting impossible and, in my view, disqualifies them for anything you plan to scale.

What I tell people to do

Instrument first. Before you pick anything, count your realistic monthly runs and multiply by your step count — then add error retries, which is where the marketing manager’s £740 came from. Set hard spend caps and volume alerts on day one. Keep the long tail of low-volume integrations on the broadest connector library you can afford, and pull any workflow that exceeds roughly 50,000 monthly steps onto per-execution or self-hosted infrastructure. That split has saved every team I have recommended it to between 60 and 80 per cent, and it takes about a day.