When Zapier is not enough: the signs, Zapier’s own fixes and the other options

Start with the fair version. Zapier connects more than 9,000 apps, and a person in operations can build, change and fix a workflow in an afternoon without joining an engineering queue. For moving a form entry into a CRM, posting an alert to Slack or filing an invoice attachment, that speed is the point, and a custom build would cost more to write and more to keep running.

The same speed explains how, in a larger organisation, Zapier can end up carrying processes that finance or customer operations depend on, built one sensible Zap at a time by whoever needed them. Nothing is wrong with that, but it does mean the question of whether Zapier is still enough deserves a proper answer, workflow by workflow.

Zapier also does more than it used to. Everything below comes from its pricing page or help centre, read on 29 September 2026, and several items answer the usual objections before anyone needs to leave.

If you need someone to build or repair Zaps rather than replace them, Zapier runs its own directory of Solution Partners, tiered and with client reviews. When none of the signs below apply, hiring one of those partners is usually the cheaper answer.

No single sign here is a reason to leave. Each one has a first check inside Zapier in the right-hand column, and I would run that check before anyone scopes a rebuild. The case for moving is usually two or three signs at once, on a workflow that touches money, personal data or a customer record.

A classic Zap is deterministic. The same input takes the same path every time, so once it works it tends to keep working until one of the apps changes. An AI step behaves differently: it can give a different answer to the same input, and it can be confidently wrong in a way a filter never is. That changes what the workflow needs around it, whichever platform runs it.

The cost model moves too. An AI by Zapier step uses one task on the Standard tier, three on Advanced and five on Premium, the default for new steps on paid plans; bringing your own model key keeps it at one task, tool calls included, and each Zapier MCP tool call uses two. A workflow that cost one task a step can cost several times that once each run reasons and calls tools, so the AI version of a workflow needs its own cost forecast before it goes live.

What the step needs is the same on any platform: a record of what it saw and what it did, a test set of real past cases run before any prompt or model change, and a deterministic check between the model’s proposal and anything that writes to money, identity or customer records. That last idea is our published pattern where the model proposes and a deterministic check decides, and it applies inside Zapier as much as outside it.

There are four realistic routes, and the first is to stay. Which one fits depends on the signs you are actually seeing rather than on which vendor anyone prefers.

I would try the split before a full rebuild. It leaves the parts that operations people change often on a tool they can change themselves, and spends engineering time only where a failure costs money or trust. A full rebuild earns its cost when most of the workflow is the risky part.

The build route is what engagement file D-04 shows: an identity provisioning system for more than 1,000 users across more than 600 groups, built rather than configured because off-the-shelf group tooling did not fit a model where one person belongs to many groups. Each run writes only the difference between intended and actual access, and a dry-run mode shows the change first, because a single run can alter access for hundreds of people.

If the next question is who should do the work, our explainer on what an AI automation agency delivers compares it with a consultancy that builds, including what you own when the contract ends.

Answer these in writing, one workflow at a time. For a single workflow they take an afternoon, and the answers are useful whoever does the work, including when the answer is to stay.

If most answers point back to Zapier, stay and fix it there. If two or three point away, start with the split described above.

UK organisations sending personal data to Zapier rely on a transfer to the United States. Zapier says it is certified to the UK Extension to the EU-US Data Privacy Framework, and since 12 October 2023 UK businesses can send personal data to US organisations certified to that extension without further safeguards. Whether that suits a particular data set is a question for your data protection officer or legal adviser.

The custom end of this is what we build, through our AI integration and automation service: the part of a workflow that writes to a ledger, an identity system or a customer record, with dry runs, rollback and an audit log. Where Zapier or another platform already handles the orchestration well, we keep it and add the missing guarantees around it.

We take small fixed-scope statements of work as well as larger builds, so moving one risky step out of a Zap can be a job of its own, and once the scope is signed the build can start within a day. If the job is already scoped, we can resource it on contract from the collective of associates who work with 1AYM, held to the same standard. If the open question is which workflows are worth automating at all, start with a fixed-scope opportunity and feasibility sprint, which typically runs two to four weeks. Either way, the first step is a 30-minute call.