The price displayed by an automation platform represents only part of the actual expense. Make’s recent change to its AI credit calculation provides a useful opportunity to review the full cost of each automated process instead of focusing solely on subscription fees.
What has just happened
On August 25, 2026, Make introduced separate credit rates for input and output tokens used through its AI Provider. Connections using a company’s own API key are not affected. In Make’s published extraction example, a request using 2,000 input tokens and 100 output tokens falls from approximately 0.6 to 0.16 credits, a reported reduction of 73%. Conversely, generating a long response from a short prompt may become more expensive, as explained in Make’s announcement about its updated AI credit model.
Why this affects you — or does not
You are directly affected if a Make workflow uses the integrated AI Provider to extract information from documents, classify requests, summarise records or generate text. Execution costs may then depend on the selected model and the amount of input and output, not only on the number of modules. Make’s credit documentation states that some functions have fixed consumption, while AI and advanced modules may depend on tokens, file size, page count or processing time.
You are less directly affected if you use your own API key, run scenarios without AI modules or rely exclusively on n8n or Power Automate. Their costs still need to be measured, but the variable component may appear in provider invoices, per-user licences, hosting or internal maintenance instead.
The matter is more urgent when a workflow supports invoicing, customer service, order management or reporting. Make states that scenarios stop running when available credits are exhausted and resume after credits are added or the plan is changed. The issue therefore concerns business continuity as well as expenditure. Make’s pricing documentation describes this behaviour and the available options for additional credits.
What this changes in practice
The subscription is not the cost of the process
France Num places common automation subscriptions between €10 and €50 per month, with advanced plans ranging from €50 to €200 per month. It also estimates that a simple, robust first automation may require between half a day and two days to set up. These benchmarks, published in France Num’s automation guide, are useful for elementary workflows but may not include business integrations, exception handling, monitoring or incident recovery.
A management-level cost calculation must include the following items:
| Cost item | What must be counted | Required deliverable |
|---|---|---|
| Process scoping | Business rules, variants, exceptions, inputs and human approvals | Process map and rule register |
| Build and integration | Configuration, connectors, APIs, transformations, tests and error handling | Design file and acceptance record |
| Consumption | Subscription, credits, AI calls, storage and overages | Usage report reconciled with invoices |
| Operations | Alerts, failed-run analysis, retries and data corrections | Operations dashboard |
| Maintenance | Changes to APIs, forms, business rules or AI models | Change register |
| Continuity | Manual fallback, recovery of pending records and process downtime | Recovery procedure |
The full cost is therefore the cost of design and integration, plus usage, operational work, maintenance and interruption risk. A licence count alone cannot support the decision.
The same workflow can produce different costs
Consider an illustrative scenario: a supplier document arrives by email, its content is extracted, the expense is classified, and the resulting data is submitted for approval before being sent to the ERP.
An extraction step with extensive input and a short structured response may benefit from Make’s new calculation. A step asking the AI to draft a lengthy explanation may move in the opposite direction. Retries, unnecessary branches, repeated searches and row-by-row processing also increase the number of actions executed. Make states that an action generally consumes one credit, with dynamic consumption applying to some advanced features.
The relevant decision is therefore not simply whether to keep Make. It is whether the current architecture remains appropriate for the process volume, criticality and available internal skills.
| Model | Main billing unit | Frequently overlooked cost | Appropriate when… |
|---|---|---|---|
| Make | Actions and credits, with dynamic usage for some AI functions | Unnecessary branches, long outputs, overages and connector dependency | Business teams need a visual tool and usage remains observable |
| n8n | Complete workflow executions on hosted plans, or self-hosted infrastructure | Hosting, backups, upgrades and internal support | The company wants greater hosting control or more technical workflows |
| Power Automate | User, process or capacity licences depending on the automation | Premium connectors, unattended RPA and environment administration | The process is closely integrated with Microsoft 365, Dynamics or Power Platform |
n8n prices hosted plans around complete workflow executions rather than individual steps, while Microsoft provides different licence types for users, processes and automation modes. The official n8n pricing page and Microsoft Power Automate licence documentation illustrate why comparing monthly subscription prices rarely produces a reliable decision.
Each manager needs a different indicator
- The process owner monitors completed records, exceptions, human approvals and items requiring manual recovery.
- IT or the automation owner monitors credits per execution, failures, retries, response times and connector changes.
- The CFO reconciles platform invoices, AI provider charges and internal operational work.
- The managing director reviews cost per successfully completed record rather than the advertised price of the tool.
This complements our analysis of Power Automate licence costs in 2026. The next step is to apply a consistent method for measuring automation ROI using observed business volumes.
What to do by October 31, 2026
- By September 30 — IT: export the register of active Make scenarios. For each scenario, document the business owner, trigger, AI modules, connection method and manual fallback procedure.
- By October 9 — process owner: review representative executions and record credit usage, failures, manual recovery and output length. The deliverable is a measurement sheet for each process, not a general impression.
- By October 16 — CFO: reconcile usage reports with Make invoices, AI provider charges and the internal cost of interventions. Separate initial build, normal operations and incident costs.
- By October 31 — managing director and IT: issue a decision note for each critical workflow: retain the architecture, optimise the scenario, use a company-owned API key or assess another platform. Assign an owner, budget and review criterion to each decision.
What you can ignore for now
There is no need to migrate every Make scenario. Make states that no mandatory action is required and that connections using a company-owned API key are unaffected by the change.
You should not assume that every AI automation has become cheaper either. The change primarily benefits workflows with large inputs and short outputs. Long-form generation, verbose agents and scenarios with repeated attempts must be assessed using their own execution logs.
Changing platforms solely to obtain a lower subscription price may simply move expenditure into hosting, administration or maintenance. Our Power Automate, n8n and Make comparison remains useful for platform selection, but process cost is primarily determined by architecture and exception handling.
Our view at D1 Consulting
Make’s change does not create the cost problem. It exposes the weakness of budgets built solely from licence fees. Automation becomes expensive when an unclear process is transferred directly into a tool, failed runs are not monitored or no recovery procedure has been defined.
We begin by representing the real workflow, including approvals, exceptions and failure paths. We use BPMN Studio, our process modelling application, as part of a structured business process audit and mapping engagement. This extends our process mapping playbook into a deliverable that can be used for architecture and cost estimation.
We then reconcile the process map with execution logs, invoices and internal operating effort. Alternative architectures can consequently be compared using consistent criteria: cost per completed record, monitoring capability, platform dependency, fallback procedure and maintenance workload.
Our Automation and Process Optimisation service covers the correction and industrialisation of the selected workflow. For Make environments or self-hosting requirements, our n8n and Make integration service includes integration, monitoring, documentation and operational handover.
👉 Book a free 30-minute diagnostic to calculate the full cost of an automated process and identify the expense that should be addressed first.
