Build an LLM automation workflow with Make

1. Trigger: a schedule (cron) or an event (new email, new form row, webhook) starts one execution;
2. Data preparation: fetch the content to process and trim it to the length the model needs — this step directly determines input-token cost;
3. Call the LLM API: send a request to the model endpoint through the platform's HTTP module (Bearer key authentication, JSON body); Make offers a REST-style open API and a custom-app mechanism, and most administrative endpoints require a paid subscription (per official documentation);
4. Parse and validate: check the response is usable (timeouts, rate limits, empty responses) and decide retry or fallback;
5. Follow-up actions: write back to a sheet, send a notification, update a ticket.

Cost control lives in steps 2 and 4: longer inputs and more retries mean higher bills. Unit and cache prices per model are in the table below.

1. Create a new Scenario on the platform and configure the trigger's schedule;
2. Add a data-fetching module (read the form / email / file) and trim fields with length truncation;
3. Add an HTTP module: POST to the model endpoint, authentication in the headers, body = {model, input, parameters};
4. Configure error handling: bounded retries with backoff for 429/5xx, and defaults for empty responses;
5. Trial with a small volume first, record the actual input/output tokens per execution, then estimate the monthly bill with the cost table below;
6. Set platform-side alerts (consecutive failures) so a broken flow fails loudly instead of silently burning quota.

Buttons, module names and billing details change with platform versions; this article deliberately stays at the concept level. Follow Make's official documentation and your actual console.

· Two bills: total cost = platform operation fees + LLM API fees. The former depends on the platform plan and modules consumed per execution — current Make and n8n pricing has not been verified by this site; check each vendor's official site. This article cites no unverified price numbers;
· Retry amplification: if each task retries 0.5 times on average, real API consumption ×1.5; under heavy rate limiting it is worse. Write retry caps, backoff and "fail to manual review" into the flow before shopping for a cheaper model;
· Input trimming: cutting a 20,000-token source to 2,000 tokens at step 2 cuts input cost by an order of magnitude — the most certain optimization in automation scenarios.

If data must not leave your intranet, choose a self-hostable option such as n8n — a fair-code licensed tool with Docker self-hosting support (see the n8n documentation); if the flow is highly customized and needs code-level control, build a custom API; at very high execution volumes with an ops team, self-hosting is usually cheapest — but rate limiting, retries and monitoring become your own job. The comparison framework across the three options is in Make, n8n or a custom API — neither article contains any promotional links.

· Commercial relationship: Make's official affiliate program is in candidate evaluation on this site — not applied for, not approved — and this article contains no promotional links. If it is ever approved, the CTA will appear only near this section with the commission relationship disclosed (rel="sponsored nofollow noopener"), and recommendations will not change because of the partnership;
· Data sources: model prices in the cost table come from this site's single source models.json (official list prices, each with a source link and verification date — 2026-10-28 basis for this page); USD models are converted at the reference rate (CFETS central parity), indicative only;
· Unverified items: Make's per-scenario/operation billing figures and n8n pricing are not verified and are labeled as such instead of guessed;
· This article is documentation research with no hands-on conclusions; price-verification time ≠ page-content update time.

· Recompute with your real token usage: cost calculator;
· Full pricing profiles: model library; two candidates via model comparison;
· Choosing a platform: Make, n8n or a custom API;
· Back to the guide directory.