The glue layer nobody budgeted for

What the meter actually counts

Zapier bills in tasks. Every successful trigger or action step in a Zap costs one task, while Zapier's own built-in tools (Filter, Paths, Formatter, Delay, Looping, Storage, Tables, Forms) cost zero. Put those two rules side by side and the model is legible: you are charged for crossing an application boundary, not for logic. A workflow that touches six apps costs six times one that touches a single app, at identical business value.

Make counts the same shape under a different name. One module action is one credit. Custom JavaScript or Python runs at 2 credits per second of execution time, the only place either vendor prices raw compute, and it prices it steeply. Data transfer scales with the plan at 5 GB per 10,000 monthly credits, and unused credits expire at the end of the term.

The AI steps changed the arithmetic again. On Zapier's published rates an advanced model step costs 3 tasks and a premium model step costs 5, with each tool call the agent decides to make billed separately at the same rate. An MCP tool call is 2 tasks. A triage workflow that used to be one task becomes five or fifteen, before your model provider sends its own invoice.

The bill tracks volume, not headcount

The entry prices are genuinely cheap, and that is not a trick. Zapier's free tier gives 100 tasks a month with two-step Zaps and 15-minute polling. Professional starts at $19.99 a month, Team at $69 for 25 users. Make's free plan allows 1,000 credits and 2 active scenarios; Core is $12 a month at the 10,000-credit tier, Pro $21, Teams $38.

Sticker price was never the problem. Slope is. Double your orders and you double your credits, because every order walks the same six steps. Zapier moves you to pay-as-you-go automatically when you pass the limit unless you switch that off, and Make's credits expire at term end, so you size for the peak month and eat the trough. Neither behaviour is hidden. Both mean your integration cost is a function of business volume you do not control.

A queue worker on infrastructure you rent costs roughly the same whether it moves 10,000 records or 200,000. The marginal cost of the 200,001st is a few milliseconds of CPU. That difference in slope is the whole argument, and it only starts to matter above a certain volume.

At the top of the ladder the prices stop being published

Workato lists four editions today (Standard, Business MCP, Enterprise MCP, and Workato ONE) and not one price. The page offers a quote request instead. What it does publish is the shape of the ration: Standard includes 1 concurrency and 30-day log retention, Enterprise MCP includes 15 concurrency and flexible 90-day retention, Workato ONE includes 30.

Concurrency as a purchasable unit is worth sitting with for a second. You are buying permission for your own business processes to run at the same time as each other. That is ordinary enterprise procurement and Workato is not doing anything unusual by it. It is still a strange thing to negotiate once the processes in question are the ones your revenue runs through.

What this looked like in practice

When we moved Stay World Class off Webflow and Xano onto Next.js, NestJS and Supabase, the automation layer stopped being a separate vendor. A 4-stage ETL mapped every legacy identifier to a PostgreSQL UUID so nothing lost its history, and the workflows that had been metered scenarios became scheduled jobs inside the NestJS app, priced as compute. The front end came along for the ride: Lighthouse went from 55.91 to 91 out of 100 and LCP dropped 77%.

Here is when staying put is the right answer, and it covers more cases than the rest of this post might suggest. If your automations are low volume, if they connect two SaaS products you do not own either end of, if an operations lead who does not write code owns them and rewrites them most weeks, then a platform is exactly the correct tool. Rebuilding a Gmail to Slack notification in NestJS buys you a maintenance burden and nothing else.

The test I would use is cheap to run. Pull last month's task or credit count and ask which workflows produced it. If a handful of scenarios moving your primary business object account for most of the meter, you are paying a metered rate to run your own logic. If it is fifty small conveniences spread thin, leave them where they are.