AI for Automation

Relay.app — The Complete Guide

Relay.app is a workflow automation platform with a standout feature no other tool matches as smoothly: human-in-the-loop. Pause any workflow for a human to approve, review or decide — then continue automatically. 100+ integrations, built-in AI steps. Free tier: 200 steps per month. Professional: $19 per month.

Workflow AutomationHuman-in-the-loopFree tier availableProfessional: $19/monthLast reviewed: April 2026

What is Relay.app?

Relay.app is a workflow automation platform that sits between Zapier's simplicity and Make's power — with one feature that neither has by default: human-in-the-loop. Relay is built around the idea that some workflows should not run fully automatically. You can insert approval steps, manual decisions and collaborative reviews into any automation, so a human can check or approve before the next action fires.

It connects over 100 apps including Gmail, Slack, Notion, HubSpot, Airtable, Google Sheets and more. You build workflows visually, add AI steps using GPT, Claude or Gemini, and choose where a human needs to stay involved. The result is automation that is reliable but not rigid — it can handle the 80% of cases that follow a pattern, and pause for human judgment on the 20% that do not.

Relay is used by teams at companies including Cursor, Ramp and Motion. It has a 4.9/5 rating on G2 from over 80 reviews, with ease of use consistently the top-cited reason. Non-technical users regularly report building workflows they could not get working in Make or Zapier.

What human-in-the-loop actually means

In most automation tools, a workflow either runs completely or fails completely. Relay adds a third option: it pauses and waits for a human. You can add a step that sends a Slack message asking "Should I proceed?" and the workflow holds until someone clicks Yes or No. Or it can wait until a specific field in your CRM is updated before continuing. Or it can route to a different team member based on the data in the workflow.

This matters most for workflows where an automatic action gone wrong is costly — sending emails to customers, approving refunds, updating billing records, publishing content. Relay lets you automate the preparation while keeping the human in control of the final action.

Who Relay.app is for

Relay is well-suited for small to medium teams that want automation without the setup complexity of Make or n8n. It is particularly good for teams that need an approval layer — finance, legal, customer success, operations — where fully automatic execution is not appropriate for every step.

Non-technical users consistently praise Relay's interface as the easiest in the category. If you have tried Zapier and found the conditional logic limiting, or tried Make and found it overwhelming, Relay is the most likely to click.

Why not just use Zapier?

Zapier's strength is breadth — 8,000 integrations and the largest template library. Relay has around 100 integrations, which is enough for most business workflows but a smaller selection. Where Relay beats Zapier is in human-in-the-loop capability, AI step quality (Relay's AI actions are more flexible than Zapier's AI Fields), and a cleaner interface that users consistently find easier to build complex multi-step workflows in.

If you need a niche integration that Relay does not support, Zapier is the answer. If you need human approval steps baked into workflows, Relay is the answer.

Is Relay.app free?

Yes. The Free plan includes 200 automated steps and 500 AI credits per month, with access to all features and integrations. Professional is $19 per month billed annually for 750 steps and 5,000 AI credits. Team is $69 per month for up to 10 users. Enterprise pricing is custom. Note: steps and AI credits are separate meters — steps count each action in a workflow, AI credits count AI model calls specifically.

Getting started with Relay.app

Step 1 — Sign up at relay.app

Go to relay.app and create a free account. The dashboard shows your workflows (called "plays" in Relay). Click New play to open the builder.

Step 2 — Choose a trigger

Select what starts the workflow. Common triggers: new email in Gmail, new row in Google Sheets, new deal in HubSpot, form submission in Typeform, webhook from any app. Authenticate the app when prompted.

Step 3 — Add steps

Click the + button below the trigger to add the next step. Choose from: app actions, AI steps (ask GPT/Claude to classify, summarise or extract data), conditional paths (if/else logic), or human-in-the-loop steps (request approval or input from a team member). Chain as many steps as needed.

Step 4 — Add a human-in-the-loop step

For any step where a human should review before the workflow continues: click + and select "Human review" or "Request input". Choose where to send the notification (Slack, email). The workflow pauses at this step until the assigned person responds. This is Relay's signature feature — use it wherever automatic execution is risky.

Step 5 — Test and activate

Click Run once to test with sample data. Check that each step produces the expected output. Turn the play on when tests pass.

15 Relay.app workflow ideas

With human-in-the-loop steps

Content approval before publishing
When a new article draft is added to [Notion/Google Docs], use an AI step to check it for: spelling errors, missing sections from the template, and brand voice compliance. If issues are found, notify the author in Slack with the specific issues. Add a human review step: only proceed to publish if the editor clicks Approve. Then publish to [CMS] and post the link to [Slack channel].
Customer refund approval
When a customer submits a refund request via [Typeform/form], extract: customer name, order number, amount, and reason. Check the order in [e-commerce system]. If the refund is under $50 and within 30 days, auto-approve and process. If over $50 or outside 30 days, pause workflow and send to [customer success lead] in Slack for human approval with all context. Process the refund only after approval.
New vendor onboarding
When a new vendor is added to [spreadsheet/Airtable], use AI to check their submitted documentation is complete. Send to the [compliance team] for human review with a checklist of what to verify. When approved, automatically: create vendor account in [accounting tool], add to approved vendor list, send welcome email with payment instructions, and notify the requesting department.
Unusual expense flagging
When a new expense is submitted in [expense tool], use AI to check if it is unusual — over $500, an unfamiliar vendor, or a category that is not standard. Normal expenses: auto-approve and sync to accounting. Flagged expenses: pause and send to [finance manager] in Slack with all details for human approval. Log the decision (approved/rejected and reason) to the expense tracker.

AI-powered workflows

Lead qualification with AI
When a new lead fills out [contact form], send their company name and role to [GPT/Claude] and ask: is this an ideal customer profile based on [describe your ICP]? Score them 1-10 with reasoning. Route high-score leads (8+) immediately to senior sales rep via Slack. Medium scores (5-7) to standard CRM sequence. Low scores (under 5) to nurture email sequence. Log score and reasoning to CRM.
Support ticket classification
When a new support ticket arrives in [help desk], send the subject and first 200 characters of body to AI. Ask it to: classify as Bug/Feature Request/Account Issue/Billing/General, estimate urgency 1-5, and suggest which team should handle it. Use this classification to: assign to correct team, set priority in the ticket, and post to the relevant team Slack channel with a one-sentence summary.
Meeting notes to CRM
When a meeting ends and the transcript is added to [Notion/Google Doc], send to AI and ask it to extract: company name, key pain points mentioned, budget indicators, next steps agreed, and deal stage recommendation. Update the CRM record automatically with all extracted data. Create a follow-up task for the sales rep with the agreed next steps and a suggested deadline.

Standard automation workflows

Waitlist to onboarding
When someone joins the waitlist via [form], add them to [email list] tagged as waitlist. When a new batch is ready to onboard, a human-in-the-loop step notifies the team. On approval, send personalised invite emails to the next [N] people on the waitlist, move them to active users in the CRM, and remove from waitlist. Update the waitlist count in the team dashboard.
Social proof collection
7 days after a customer's first purchase (or key milestone), send them an email asking for a review with a direct link to [G2/Capterra/Google]. When they submit a review, use AI to assess the sentiment and star rating. If 4-5 stars: add to testimonials database, notify marketing team. If 3 stars or below: notify customer success for follow-up, and do not add to testimonials.
New employee check-in sequence
30 days after a new employee starts (trigger from HR system), send them a check-in survey via [Typeform]. When they respond, use AI to identify any concerns or issues mentioned. If concerns are detected, create a task for their manager and HR with the relevant quotes. Send the overall sentiment score to [HR dashboard]. Flag if the score is below [threshold] for urgent follow-up.
Project status updates
Every Friday at 4 PM, for each active project in [project management tool], gather: tasks completed this week, tasks overdue, blockers logged. Use AI to write a 3-sentence status summary. Send a draft to the project manager for human review. After approval, send the summary to stakeholders and post to the [project Slack channel]. Log to the weekly report spreadsheet.

Tips for Relay.app

Use AI steps before decisions. The most powerful Relay pattern is: AI analyses incoming data → routes to appropriate path or person → human approves if needed → automation executes. The AI step does the triage, the human step catches edge cases, and the automation does the mechanical work. Neither AI nor human alone is as effective as both in sequence.

Start with the human-in-the-loop step, then remove it. When building a new workflow, start with a human review step before every consequential action. Let the workflow run for a week, approving everything manually. Once you see the patterns and are confident in the logic, remove the review steps for the cases that are always approved. You end up with automation that is both reliable and only asks for human input when genuinely needed.

Watch your AI credit consumption. Relay's AI credits are consumed by AI model calls. Complex AI tasks — web scraping, image generation, multi-step research — use significantly more credits than basic text classification or summarisation. Monitor your credit usage for the first two weeks and estimate monthly consumption before choosing a plan tier.

Technical background

Relay.app is built by Relay.app Inc., founded in 2021. The platform is SOC 2 certified, per the official Relay.app website. It integrates with over 100 business apps via their official APIs. All integrations are included on all plans — there is no premium tier required to access specific connectors.

Steps vs AI credits

Relay uses two separate usage meters. Steps count each app action in a workflow — sending an email, creating a CRM record, updating a spreadsheet. AI credits count AI model calls — each time Relay calls GPT, Claude or Gemini on your behalf. Simple automations without AI steps only consume steps. AI-heavy workflows consume both. Per Relay's official pricing page, triggers do not count as steps, and some workflow elements (wait steps, path evaluations, data transforms) also do not consume steps.

Human-in-the-loop architecture

Relay's human-in-the-loop feature works by suspending workflow execution at a designated step and sending a notification (via Slack, email or in-app) to a specified person or team. The workflow stores its state and resumes only when the designated person responds — approving, rejecting or providing requested input. According to user reviews on G2, this is the feature most frequently cited as Relay's primary reason for choosing it over alternatives.

Pricing (verified April 2026)

  • Free: 200 steps/month, 500 AI credits, 1 user, all features and integrations
  • Professional: $19/month (annual) — 750 steps, 5,000 AI credits, 1 user, priority polling
  • Team: $69/month (annual) — 2,000 steps, 5,000 AI credits, 10 users, shared workflows
  • Enterprise: Custom — SOC2, GDPR, priority support, custom integrations
  • Additional AI credits: Available as add-ons, 5,000 credits for $11/month annual
Primary sources cited in this guide