Action Credit Pricing: Logic and Approvals Are Free

You set up an AI workflow to handle something important (maybe it drafts responses to inbound leads, or categorizes support requests, or processes refund requests). You add a human approval step because you are not ready to let AI send messages to your customers unsupervised. Smart move.

Then you check your automation bill and realize you are paying for every approval. The platform counts "human reviewed this action" as a billable operation, right alongside the actual work. You are being charged for caution.

This is the fundamental problem with per-task pricing, and it is why we built Rills around a different model.

The Perverse Incentive in Per-Task Pricing

Per-task pricing means every step in a workflow costs money. It does not matter whether that step sends an email to a customer or checks whether a number is greater than 10. Both count. Both cost. Zapier's pricing tiers illustrate this well: every action step in every Zap consumes from your monthly task quota, regardless of whether it touches an external system or just moves data around internally. Make's operations model is similar: each module execution counts, whether it's a router deciding which branch to take or an API call doing meaningful work.

This creates a specific, predictable problem for anyone building supervised AI workflows, and the same dynamic shows up when you map real Zap and scenario volume in our Rills vs Zapier vs Make comparison: the more oversight you add, the more you pay.

Think about what that incentivizes. You want to add a review step before AI sends a client email? That is another task. You want to route high-value leads through manual review? Another task. You want to add a conditional check so only certain cases get flagged? More tasks.

Per-task pricing tells you, in dollar terms, that being careful is expensive. It pushes you toward fewer steps, less logic, and less oversight, which is the opposite of what you actually want when AI is making decisions on your behalf.

What Action Credit Pricing Changes

Action credit pricing draws a clear line: logic is free, actions cost credits. Human approvals are always free.

Here is the breakdown:

Free (no credits consumed):

• IF/ELSE branches
• Data filtering and transformation
• Variable formatting
• Conditional routing
• Loop iterations
• Human approvals (always)

Costs credits:

• AI model calls (GPT-5, Claude, etc.)
• External API calls
• Sending emails
• Webhook deliveries
• SMS and notification sends
• File processing
• Database writes to external services

The principle is straightforward: if a step moves data around, makes a decision, or asks a human for input, it is free. If a step reaches out to an external service or runs an AI model, it costs credits.

Why Free Approvals Matter More Than You Think

This pricing changes how you design workflows.

When approvals cost money, you minimize them. You set up your workflow, hold your breath, and let it run. If something goes wrong, you find out after the fact when a wrong email was sent, a lead miscategorized, or a refund processed incorrectly. The cost of oversight pushed you into a monitoring-and-damage-control pattern.

When approvals are free, you design workflows the way you actually want them to work. You add approval steps wherever the stakes are high. You route edge cases through manual review. You keep humans in the loop for anything customer-facing until you are confident the AI handles it well.

This connects directly to how Rills is designed. Workflows in Rills start supervised. AI steps handle classification, drafting, and analysis, then you review and approve before consequential actions execute. Over time confidence scoring tracks the AI's accuracy for each specific type of decision. As confidence rises, you can let proven patterns run autonomously while keeping novel situations in your approval queue.

That progression from supervised to autonomous only works if the supervised phase does not penalize you financially. If every approval is a line item on your bill, you are incentivized to skip it and jump straight to autonomous execution before you are ready. Action credit pricing removes that pressure entirely.

What You Actually Pay For

Credits go toward the things that cost real resources to execute: AI inference, external API calls, and outbound communications. These are the operations that create tangible value in your workflows and have real compute or service costs behind them.

Everything else -- the logic that decides which path to take, the filters that determine whether a workflow should continue, and the approval steps that keep you in control -- run without consuming credits.

This means you can build workflows with as many decision branches, filters, and approval checkpoints as you need. A workflow with 3 logic steps and one that has 30 logic steps cost the same in credits, as long as they take the same number of external actions. Complexity in your logic is not punished. It is expected.

How This Plays Out in Practice

Say you build a workflow that handles inbound support requests:

1. AI reads the message and classifies the issue type
2. Checks the customer's account status
3. Routes based on urgency: critical issues go to your approval queue, routine ones proceed
4. For critical issues: pauses for your review (you swipe to approve on your phone)
5. Drafts and sends a response
6. Updates your help desk

Under per-task pricing, that is six billable operations every time it runs. Under action credit pricing, you pay credits for three things: the AI classification (step 1), the email send (step 5), and the help desk update (step 6). The routing, the account check, and your approval are free.

Now imagine you want to make this workflow smarter. You add a second approval gate specifically for refund-related issues. You add conditional routing that escalates anything marked urgent. You add three more filter steps to catch edge cases. Under per-task pricing, your bill just went up. Under action credit pricing, your bill stays the same; approvals and logic are free regardless of how many you add.

The pricing rewards exactly the behavior you want: building thorough, well-supervised workflows.

The Confidence Scoring Connection

Here is where action credit pricing and Rills' confidence scoring system reinforce each other.

Every time your workflow runs, each step gets a confidence score based on the specific input data. High confidence means the AI is sure about its decision. Low confidence means it is uncertain, and the action pauses for your review.

In the early days of a new workflow, confidence is low across the board. You review most actions. As the system learns from your approvals and rejections, confidence rises for patterns it handles consistently well. Those patterns graduate to autonomous execution.

Because approvals are free, you are not paying extra during the supervised learning period. You can take your time. Review everything for a week. Two weeks. A month. Let confidence build naturally. The workflow gets smarter without your bill getting bigger.

With per-task pricing, that learning period costs money; every approval is a billable operation. There is financial pressure to skip the supervised phase and go straight to autonomous, which means skipping the phase where the system actually earns your trust.

Comparing Pricing Models

Most automation platforms use one of two pricing models:

Per-task pricing charges for every operation in a workflow. The more steps your workflow has, the more you pay. This is simple to understand but creates bad incentives: it punishes logic-heavy workflows and treats oversight as overhead.

Per-workflow or per-execution pricing charges each time a workflow runs, regardless of step count. Better for complex workflows, but it still treats human approval steps the same as automated ones, and it can make simple, frequent workflows disproportionately expensive.

Action credit pricing only charges for steps that reach outside your workflow: AI calls, API requests, outbound messages. Everything internal to your workflow (logic, routing, approvals, data transformation) is free. This means your costs scale with the actual value your workflows produce, not with their internal complexity.

The right model depends on what you are building. If your workflows are simple two-step automations with no human oversight, per-task pricing might be fine. But the moment you start building workflows that involve AI decision-making, conditional logic, and human review (which is what supervised AI workflows are), action credit pricing aligns your costs with your actual goals.

The Bottom Line

Pricing models are not neutral. They shape what you build and how you build it. OpenView's guide to usage-based software pricing explains why consumption-based models often align what customers pay with the value they receive, compared with flat seat licenses where cost can drift away from actual use.

Per-task pricing tells you that every approval, every filter, and every branch costs money. So you build simpler, less supervised workflows. You skip the human review step to save a few cents. You let AI run unsupervised before you are confident in it.

Action credit pricing tells you that oversight is free and logic is free. So you build workflows the way they should be built: with thorough logic, appropriate safeguards, and human review where it matters. You pay only when your workflows take real-world actions that deliver real value.

If you are a solopreneur building AI workflows and you want to stay in control without being penalized for it, that distinction matters. You can see the model in action in our step-by-step guide to building your first workflow: the example walkthrough shows exactly which steps consume credits and which are free.

See how it works. Check out our pricing page to see action credits in detail, or start building a supervised workflow, with human approvals included at no extra cost.