How to find the trial users most likely to convert

If you run a SaaS with free trials, not every signup deserves the same attention. Some trial users are showing signs that they may convert. Others look like valuable accounts but are getting stuck before they reach the point where they can see the product’s value.

Here’s a workflow that checks trial users at key points during the trial, scores their behavior, and helps you decide who may need sales attention, onboarding help, or no action yet.

The tools

• Supabase: Stores the data for each trial user.
• n8n: Runs the workflow, calculates the scores, and sends each user down the right path.
• OpenAI: Looks at the trial data and writes a short explanation and email.
• Gmail: Creates the email as a draft for you to review.

The workflow assumes your product already sends basic usage data to Supabase. If your data is stored somewhere else, use that system instead of Supabase in these steps.

Step 1 — Decide what behavior matters

First, decide which actions show that a user is getting value from your product. For a reporting tool, you could track:

• Import data
• Create a report
• Connect an integration
• Invite a teammate

Next, choose actions that can show buying interest:

• Viewed pricing
• Started checkout
• Invited teammates
• Asked a billing or plan-limit question

Don't rely on weak signals such as logins. Someone who logs in five times may be very engaged. It might also mean the user cannot work out what to do. Focus on actions that give you useful info.

Step 2 — Create one row for every trial user

In Supabase:

• Open: Table Editor
• Click: New table
• Enter: Name → trial_watch

Add these columns (or customize them based on the signals you want to track):

• user_id
• email
• trial_started_at
• trial_ends_at
• sessions_7d
• core_actions_7d
• key_feature_used
• activation_events_completed
• teammate_invited
• pricing_viewed
• Checkout_started
• support_activity
• account_fit
• last_active_at
• converted
• last_agent_action
• activation_score
• intent_score
• route
• last_checked_at
• outcome

Use Boolean for true/false fields, integers for numbers, timestamps for dates and times, and text for labels. activation_events_completed can be a text array.

Make sure your product sends the events you want to track to Supabase. The workflow cannot know about an event that your product doesn't record.

The sessions_7d and core_actions_7d fields should cover the last seven days only. For account_fit, use high, medium, or low based on info collected at signup. Don't use AI to guess this from an email address.

Now, all the trial data the workflow needs is in one place.

Step 3 — Make n8n check the trials every morning

Create a workflow in n8n.

Add: Schedule Trigger

Set:

• Trigger Interval → Days
• Days Between Triggers → 1
• Trigger at Hour → your preferred hour
• Set: Trigger at Minute → 0

Make sure the workflow timezone is correct, since n8n uses the workflow timezone — or the instance timezone if one is not set — to determine when the trigger runs.

The Schedule Trigger runs workflows at fixed times. Scheduled workflows must be saved and published to run automatically.

Next:

• Add: Supabase
• Select: Row
• Select: Get all rows
• Choose your trial_watch table
• Turn on: Return All

Now, every morning, n8n receives the latest trial data it needs for the rest of the workflow.

Step 4 — Give each trial a simple score

Give each trial two scores before bringing in AI:

• activation_score
• intent_score

In n8n:

• Add: Code
• Select: Run Once for Each Item

Calculate:

• activation_score
• intent_score

Use simple rules to start:

• Core feature used → +2 activation
• Teammate invited → +1 activation
• Pricing viewed → +1 intent
• Checkout started → +3 intent

Treat these as high scores for now:

• Activation score of 3 or more
• Intent score of 2 or more

Then set the route:

• High activation + high intent → sales
• High-fit account + low activation → onboarding
• Everything else → ignore

Also calculate _trial\day in this Code node. Then create _should\check and set it to true only for users who are on day 3, 5, or 7, have not converted, and have not reached the end of the trial.

You can change these rules later as you see which behaviors lead to conversions.

Step 5 — Only inspect users who need attention

Only send users to OpenAI if the user:

• Is on trial day 3, 5, or 7
• Has not converted
• Still has time left in the trial

In your Code node, create a _should\check field that returns true only when all three conditions are met.

Then:

• Add: Filter
• Select: should_check
• Choose: is true

Only those users continue to OpenAI.

Step 6 — Let AI explain what is happening

Now, let AI look at the trial data and explain what is going on.

Add: OpenAI

Select:

• Resource → Text
• Operation → Generate a Model Response
• Model → gpt-5.6-luna
• Output Format → JSON Schema

Create four string fields:

• summary
• missing_activation_event
• email_subject
• email_body

Use this prompt (or something similar):

You are reviewing a SaaS trial.

Use ONLY the supplied data. Never EVER invent activity.

The route has already been decided. Don’t change it.

In a single sentence, explain why this user received this route:

If route = sales, focus the message on helping them make a buying decision.

If route = onboarding, identify the most important activation step they haven't completed.

Return:
Summary
Missing_activation_event
Email_subject
Email_body

Pass in the user's trial data, including _activation\_events\completed, the scores, and route.

That's it. AI interprets the data and writes the message. Your scoring rules still control the route.

Step 7 — Split the trials by route

Next, separate the trials based on the route from your Code node.

• Add: Switch
• Select: Rules

Set the Switch rules using the route created in your Code node, not a new route generated by OpenAI.

Create three outputs:

• route = sales
• route = onboarding
• route = ignore

The Switch node supports multiple conditional routes. For ignore, stop there. For sales and onboarding, continue to Gmail.

Step 8 — Create drafts before you automate sending

Start with drafts, not automatic emails.

Add: Gmail

Select:

• Resource → Draft
• Operation → Create

Map:

• To Email → trial user's email
• Subject → AI's email\_subject
• Message → AI's email\_body

For the first few weeks, review the drafts before sending them. Look for wrong recommendations, awkward emails, and incorrect routes. Once you're happy with the results, automate the sending.

Step 9 — Close the loop

The last step is to feed the results back into the system.

After the Gmail step, add another Supabase node:

• Select: Row
• Select: Update a row

Update the record using _user\id. If a user can start more than one trial, use a unique trial ID instead.

Save:

• activation_score
• intent_score
• route
• last_checked_at

When a user converts, or their trial expires, have your app or a separate n8n workflow update outcome. This workflow only checks users at specific trial checkpoints, so let the app or separate workflow track the final outcome.

Then review your results every 20–30 completed trials:

• Are users with high activation actually converting?
• Are certain features showing up often among converters?
• Are high-fit inactive users responding to onboarding?

Don't start with a smarter AI. Start with better signals.