Human-in-the-Loop on Zapier: 3 Tasks Per Run, 600 Tasks a Month

Say you run one approval-gated workflow on Zapier. A new lead hits your inbox, AI drafts a reply, you review on your phone, you approve or reject, the email sends. It runs 200 times a month. That one workflow is where most people first bump into the real human-in-the-loop automation cost, and the cost isn't the time you spend reviewing. It's the per-task tax your platform charges for the pattern itself.

On Zapier's Pro plan, every successful Human in the Loop action counts as one task. The AI draft is a task. The send is a task. That's three tasks per run, 600 a month before you add a trigger or a filter. You're already burning most of the 750-task Pro ceiling on one workflow, and a second one pushes you up to the Team plan at $69/month.

The human part is technically free. You're not charging yourself to review a message. What's billable is every step the platform has to touch, and "approve" is one of those steps. Per-task pricing quietly taxes caution.

Rills makes a different bet. Logic is free, approvals are free, paused workflows cost $0. Build all the oversight you want, and you still only pay for the outbound actions: the AI call and the email send. The rest of this post runs the actual math on a Rills Hobby plan ($29/month) and shows what changes once approvals don't cost anything.

What You're Actually Paying For on Zapier and Make

Zapier updated its pricing in January 2024. Delay, Filter, Paths, Looping, and Digest no longer count as tasks, which was a real improvement over the old model. But actions inside third-party apps still count (one task per successful API call, one per AI step), and Zapier's own Human in the Loop step counts one task per successful approval. The waiting is free, but the tap is billable.

Make is worse if you want a genuine approval flow. The native Human in the Loop module is Enterprise-only and in closed beta. Everyone else builds the pattern by splitting a workflow into two scenarios: one that pauses by writing a row to Airtable or Google Sheets, another that resumes when a human clicks a link and fires a webhook. Each scenario has its own trigger and modules, all billable as operations. A simple approve-before-send pattern ends up around 5 to 8 operations per run.

There's also the Sleep module, which some people use to fake a "wait and check" pattern. It counts as one operation per call and caps at 300 seconds. For a full-day wait, you'd be chaining 288 Sleeps. Nobody actually does that, but the constraint exists because the pricing model doesn't accommodate long pauses. Make has no concept of "paused and costing nothing."

Why Human-in-the-Loop Automation Cost Climbs Fast

Here's the math with real numbers. You're a solopreneur approving 200 AI-drafted client replies a month.

Zapier Pro ($19.99/mo, 750 tasks included): AI step (200) + approval action (200) + send (200) = 600 tasks. You fit on paper, but almost nothing is left for anything else. Add a second workflow and you're on the Team plan at $69/month.

Make non-Enterprise (Core plan, $9/mo, 10K operations): Roughly 6 operations per run through the two-scenario workaround, 200 runs, so 1,200 operations. Comfortable on paper, but the Sleep cap pushes you toward external delay tools, and operations compound fast once you add routing, enrichment, or retries.

Rills Hobby ($29/mo): Two separate credit buckets, and the approval itself costs nothing in either.

• Workflow credits: 200 native email sends = 200 workflow credits, well inside the 10,000 included.
• AI credits: Each credit is $0.01, and an AI step consumes however many credits the underlying model call actually costs. A short email draft on a small model runs about 1 AI credit per call, so 200 drafts ≈ 200 AI credits, well inside the 1,000 included.
• Approvals, pausing, filters, and routing logic: 0 credits in either bucket.

Total on Rills: $29 flat, no overage. One workflow, two workflows, or ten workflows at that volume all stay inside the plan.

That gap isn't a percentage discount. Zapier and Make penalize the pattern of adding human oversight. Rills charges nothing for it. Workflows that feel expensive on one platform feel free on the other.

A note on the two Rills buckets: workflow credits cover the outbound action steps (a native email send is 1 workflow credit), and AI credits cover the model call itself, billed at actual cost rather than a flat per-call fee. A cheap draft might be 1 AI credit, while a longer analysis using a pricier model might be 3 to 5. You pay for the AI model work you actually did, nothing more.

How action credit pricing works has the full breakdown. The one-line version: you pay when something leaves your system, and only then.

What Free Approvals Let You Build

Once approvals are free, the cost of oversight goes to zero, and you stop optimizing workflows to minimize human steps. You start optimizing for trust.

A few patterns you can now afford:

Dual approval for high-stakes actions. Before sending anything to a paying customer, have two people approve independently. On per-task pricing, this doubles your approval tax. On free approvals, it costs the same as one.

Approval on every edge case. Configure your human review step to auto-approve when AI confidence is above 85%, and require manual human review otherwise. With Zapier or Make's per-task pricing, the filter itself is free but each approval it produces is a task. On Rills, both are free and you keep fine-grained oversight without a cost penalty.

Review checkpoints mid-workflow, not just at the end. Approve the draft after step 3, before the workflow spends credits on step 4. You catch a bad trajectory before it burns the expensive AI call downstream, which turns the approval into a cost-saving step rather than a cost-adding one.

Shadow-mode while you build trust. Run a new AI workflow in parallel with your manual process, approving every AI proposal just to see what it would have done. On per-task pricing, the shadow costs real money. On Rills, the shadow's approvals are free, and only the outbound actions on the version you actually ship consume credits.

Free approvals aren't a small discount. They change which workflows are economical to build at all. The Rills, Zapier, and Make comparison maps these patterns against real workflow shapes.

The Compromise Most Skeptics Carry

Skeptics of AI automation often object less to the AI itself than to being bill-shamed for being careful. That hesitation comes from the platform's pricing model, not from anything inherent to human-in-the-loop design.

On Rills, you design the approvals you want, pause as long as you want, and pay only for the actions that leave your system.

If you'd like to see what that feels like in a real workflow, try a live demo and swipe through a few pending approvals on your phone. The waiting is free, the approving is free, and the only thing that costs anything is the action you actually wanted to pay for.