How I Used AI to Find the Hidden Behaviors That Predict Which Readers Will Buy

The simple metric that tells you who will purchase—before they do

How would you know if your reader will click the buy button?

Before last month, I tracked opens and engagement for my posts.

That was a mistake.

After pulling 69,522 Substack events and 1,244 Gumroad transactions into an analytics database and running it through AI, one clear signal stood out: 88.9% of purchasers clicked at least one link before buying — compared with 15.6% of non-purchasers.

In plain English: link clicks are 4.5× more predictive of future purchases than opens, comments, or total engagement.

I was surprised and a little embarrassed.

That’s what pushed me to trace every purchase back through my content, timelines, and clicks.

What I found changed how I publish, what I optimize for, and how I think about revenue from this newsletter.

Part 1: The Methodology — How I Analyzed My Data

Creators talk a lot about engagement but rarely about behavior.
I wanted to know what my readers actually do before they spend money.

So I tracked everything.

I fed all my Substack and Gumroad data into a single analytics database so AI could analyze every behavior leading up to a purchase.

What I Tracked

Here is a shortened list of data I used for this analysis.

I used the Cursor app connected to a DuckDB SQL database using an MCP server.

Summary of the data collected

Connecting the Two Data Sources

I then matched email addresses across Substack and Gumroad. This gave me a unified behavioral timeline for both purchasers and non-purchasers.

Note that many people use different email addresses for their Substack and Gumroad accounts, so this method has some limitations.

Most people analyze sales in isolation.
I wanted to understand the pathway to the sale.

The Substack backend tracks each step, but manually extracting subscriber events is cumbersome. This is one of the reasons why I’m building the StackContacts CRM tool.

Substack Subscribers Events View (by author)

For data nerds only: All these 69,522 events have timestamps and details that AI can use to map relationships across 563 separate database tables. The total size of this database is 302.5 Mbytes.

This story explains the why and how I built the tool I’m using:

Why I’m Building a CRM That Remembers Your Readers Better Than You Do

The AI Analysis Process

I asked AI to analyze my posts, subscriber behavior patterns based on events, and Gumroad product sales transactions.

The AI created SQL queries and pulled three sets of data:

1. 90 purchasers with complete behavioral histories

2. 876 non-purchasers with high-engagement histories

3. The entire event sequence of each group

I cannot emphasize the importance of item #3 enough.

Substack backend collects a lot of data on subscribers’ activities. All these events, and even the clicked links' URLs, are available on the web pages, but it is just complicated for AI to access them for a large group of subscribers.

My StackContacts tool transforms this data into an accessible analytics database.

AI then analyzed:

• Time between events

• Order of events

• Repeated patterns

• Link-click behavior

• Post-level conversion

• Product-specific pathways

Surfacing Purchaser Behavior Patterns

AI quickly surfaced and summarized the patterns that humans would miss:

• Purchasers behave differently long before they purchase

• There are unique timing patterns

• Some posts predict purchases

• Link clicks form a semi-consistent sequence before revenue

I learned more about my audience in 2 hours than in 2 years of guessing.

Why This Matters

Most creators optimize for the wrong things:

• Email opens

• Total views

• Comment count and likes

But these tell you nothing about revenue.

This data flipped my assumptions:

Email opens are NOT predictive of purchases.
Link clicks ARE.

Creators guess what works.
Data reveals what actually works.

Small changes — like moving a CTA link higher in a post — can produce disproportionately large revenue increases.

Part 2: The Three Buyer Personas I Discovered

After analyzing hundreds of thousands of data points, AI surfaced three buyer personas with distinct patterns.

Here’s what surprised me most:

These personas aren’t abstract “marketing avatars.”
They are behavioral profiles derived from actual actions prior the purchase.

Summary of 3 Buyer Personas

Let’s review the details.

Persona 1: The Quick Buyer (47% of purchasers)

The Quick Buyer moves fast — much faster than I imagined.

Characteristics

• Timeline: 0–24 hours from first engagement to purchase

• Behavior:

  • Post seen

  • Link clicked

  • Purchase

• Median time between click and purchase: 0 hours

• Products: low-cost $5–$29 tools, scripts, Chrome extensions

• Conversion rate: 60–70% within 24 hours of a link click

This is the reader who sees something they want and buys it immediately.

Real Example

How To Schedule Notes on Substack

• 19 purchasers from the above post

• 36 link clicks

• Median time from click → purchase: 0 hours

• $475+ revenue (of $10,819.30 total)

The moment they see the product, they decide.

What This Means

• These buyers are already problem-aware

• They will buy NOW if you make it easy

• Shorter copy and clearer CTAs boost conversions

• Sending them directly to the checkout page works

Urgency helps — not fake urgency, but clarity:

• “Join 500+ creators”

• “Limited-time upgrade”

• “Available today”

Actionable Takeaways

• Put your CTA early in the post (I added a product link)

• Make the checkout two clicks or less

• Don’t bury the core value

• Add a P.S. CTA for fast movers

Now I finally understand why

Justin Welsh

has these CTAs at the end of every email he sends to his 175,000 subscribers. It just works for this Quick Buyer persona.

Example CTA in email I received from Justin Welsh

Persona 2: The Considered Buyer (30% of purchasers)

This group moves slower — but more intentionally.

Characteristics

• Timeline: 25–41 days from first engagement to purchase

• Behavior:

  • Reads multiple posts

  • Clicks multiple links

  • Waits

  • Buys

• Price points: $41–$79+

• Conversion rate: 38–45% within 24 hours of final click

These buyers need proof.

Real Example

Substack Pro Studio User Guide

• 18 purchasers

• 22 link clicks

• Median time from click → purchase: 531 hours (22 days)

• $1,020+ revenue ( of $4,045.25 total)

This audience researches.
They compare.
They return later — when the timing is right.

What This Means

• Higher-priced products require multiple touchpoints

• They need reassurance

• They want to see how the product works

• Long-form guides help convert them

These are your most thoughtful customers.

Actionable Takeaways

• Create tutorials & videos

• Add case studies

• Use post-to-post linking

• Send follow-up offer emails to link clickers

• Add detailed product guides

Persona 3: The Research Buyer (23% of purchasers)

This is the “deep engagement” persona — the one who consumes everything before buying.

Characteristics

• Timeline: 1–30 days

• Behavior:

  • Reads 5+ posts

  • Clicks 3+ links

  • Often comments

  • Buys after intense engagement

• Products: Mixed price points

• Highly analytical

These readers don’t impulse buy.
They want to understand the creator and the product.

Real Example:

How Often Should You Publish Notes to Grow Your Paid Subscribers?

• 11 purchasers

• 18 link clicks

• 95 comments

• Avg link clicks per purchaser: 2.87

• $275+ revenue

Their comments are thoughtful.
They’re already halfway sold — they need confirmation.

What This Means

• These buyers respond to education

• They want community and conversation

• Comments signal their readiness

These are the readers who follow your work closely.

Actionable Takeaways

• Create an educational content series

• Invite discussion at the end of posts

• Engage with commenters, answer questions

• Show your expertise (not your hype)

Part 3: The Surprising Findings That Changed My Strategy

As I dug deeper into the data, four findings fundamentally changed how I publish — and what I optimize for.

Summary of Findings

Here are the details.

Finding #1: Email Opens Do Not Predict Purchases

This one shocked me.

The Data

• Low-signup posts have higher open rates (33.5% vs 28.3%)

• But high-signup posts have 6.3× higher click-through

• Email opens = interest

• Link clicks = intent

I had always assumed high open rates = good performance.

Wrong.

What I Learned

• Don’t optimize for opens

• Optimize for what happens after the open

• Subject lines should drive action, not vanity

Actionable Takeaway

• Track link clicks per email

• Not open rates

• Write subject lines designed for the click

Finding #2: Link Clicks Are Everything

This is the single most significant predictor of revenue.

The Data

• 88.9% of purchasers clicked links

• Purchasers click 4.5× more links (2.65 vs 0.59)

• Median time from click → purchase: 0–11 hours

What I Learned

It’s almost impossible to overstate this:

If a reader clicks your link, they are signaling purchase intent — even if they don’t buy that day.

Clicks reveal curiosity.
Clicks reveal readiness.
Clicks reveal the future.

Actionable Takeaways

• Use link clicks your #1 KPI (Key Performance Indicator)

• Put CTAs early and clearly

• Track click → purchase conversion

• Tag readers who click links

Finding #3: Product Guides Convert 3× Better

This one wasn’t intuitive until I saw the numbers.

The Data

• All top 3 converting posts = product/user guides

• “How To Schedule Notes” → 19 purchasers

• “Substack Pro Studio Guide” → 18 purchasers

• Conversion: 2–5% of engagers vs 0.5–1% for other posts

People buy what they understand.
Guides show the tool in action — and people buy confidence.

What I Learned

One year ago, I believed I needed to create the product first, get a few beta testers, build a Gumroad product page, and then start writing user guides and customer-facing materials before the launch.

Wrong.

After seeing this data, I changed my approach. I started writing about the StackContacts CRM tool way before it was ready for prime time, with a focus on results and outcomes.

This article is an example - I’m sharing these results before the product is ready.

Actionable Takeaways

• Show results and outcomes, not features

• Write a guide for each product

• Include real screenshots

• Add link clicks throughout the post

Finding #4: Comments Signal Quality, Not Revenue

This was a surprise.

The Data

• High-signup posts: 3.6% comment rate

• Low-signup posts: 0.9% comment rate

• But only 22 purchasers came from 654 commenters

Comments = community
Clicks = revenue

It’s easy to misinterpret this.
Comments feel good — they just don’t correlate with product sales.
If you are selling access to the community, it might be a different story for you.

Actionable Takeaways

• Encourage comments for the community

• But track link clicks for revenue

• Don’t let comments mislead you

Do you agree or disagree? Leave a comment

Part 4: The Metrics That Actually Matter

I came away with a short list of metrics that predict product revenue—and those that don’t.

Note that these KPIs are relevant to my business model:
my Substack newsletter is free,
and I sell tools and products on Gumroad.

Your business model may be totally different, so the metrics that matter for you might be different as well. This discovery process will identify these buying patterns from the data and determine which metrics are most predictive of revenue in your case.

Primary KPIs (Most Predictive)

1. Link Click Rate

• Purchasers: 48.7% click rate

• Non-purchasers: 15.6%

• Target: 20%+

This is the #1 metric to track.

2. Click → Purchase Conversion

• Lower-priced Chrome extensions: 60–70%

• Higher-priced products: 38–45%

• Target: 50%+ overall

If this is low, fix your product pages.

3. Time From Click to Purchase

• Median: 0–11 hours

• Goal: <24 hours for lower-priced tools

The shorter the gap, the stronger the intent.

4. Post-to-Purchase Conversion

• Top posts convert 2–5% of engagers

• Aim for 5%+ on product guides

This is the ultimate measure of post quality.

Secondary KPIs (Helpful but not primary)

5. Post Views per Purchaser

Purchasers view 2.77 posts on average —
5.2× more than non-purchasers.

6. Multiple Clickers

27% of purchasers click 3+ links
Only 6.6% of non-purchasers do.

Metrics to Avoid Optimizing For

• Email opens

• Total event count

• Views alone

These metrics feel important.
They’re not.

Part 5: How to Implement This for Your Newsletter

Once I saw how strongly link-click behavior predicted purchases, something simple became obvious: creators don’t need more content — they need clear behavioral signals and an easy way to act on them.

Below is a tiered path so you can pick the level that best fits your skills and available time.

Tier 1 — Beginner (10 minutes) — Do this first

Quick wins that expose intent right away:

• Add UTM parameters to every Gumroad product link.

• Track link click-through in Gumroad (or use any click-tracking tool).

• Identify which posts have the highest click rate.
  If you only do one thing this week: add UTMs + check your top 3 posts’ click rates. That single action will tell you where your revenue is hiding.
  
  If you’ve never looked at Gumroad’s Analytics → Links page, you’ll be shocked at how informative it becomes once UTMs are in place — source, medium, campaign, click count, conversion rate, and revenue per link.

  

  Gumroad Analytics has a Links page with UTM data

Tier 2 — Intermediate (1 hour)

Start answering the questions that matter: which posts actually lead to purchases and how fast.

• Identify your top 5 converting posts.

• Compare click → purchase timing for those posts.

• Map simple buyer personas (fast movers, considered buyers, deep engagers).
  These steps let you repeat what works.

Substack Control Center can help export this post data.

Tier 3 — Advanced (analytical creators)

Stitch behaviors together for predictive power.

• Export Substack Subscriber Events data.

• Export Gumroad Sales data.

• Combine Subscriber Events with Gumroad/Stripe sales data using the email address as the linkage.

• Analyze event sequences and time gaps.

• (Optional) Use AI to surface patterns and create segments.
  This is where you’ll find the subtle timing and sequence behaviors that reliably precede purchases.

This requires a lot of Excel gymnastics if you start from scratch.

See the example screenshot for the Subscriber Events data in your Substack dashboard.

Subscriber Events data example (by author)

Tier 4 — Automated (if you want to skip the manual work)

Doing this manually taught me a lot of the available data fields and Substack APIs — but it’s a fiddly process.

That’s why I’m building the StackContacts tool: unified event timelines, behavioral segments, click intelligence, and early intent signals — all in one neatly organized database.

If you want to run this analysis without spreadsheets, StackContacts automates these four tiers with a single sync.

Quick wins you can apply today

• Move CTAs earlier in your posts.

• Publish more product guides and “how-to” walkthroughs.

• Add one follow-up email sequence for readers who click product links.

These small moves compound quickly when you focus on clicks rather than vanity metrics.

Tools you can use right now

• Google Sheets or a simple database (DuckDB, Postgres, SQLite)

• An AI assistant (Claude, Cursor) to run queries/segment data

• Gumroad’s Analytics → UTM Links for immediate click and conversion data

Part 6: The Revenue Impact

Once I started implementing these insights, everything changed.

Before Data Analysis

• I was guessing

• I optimized my posts for email opens and signups (free subscribers)

• I had no idea what drove conversions for product sales

After Data Analysis

• I understood my buyer personas

• I started optimizing for link clicks

• I found 40–50 link clickers/month not converting

Those 40–50 readers represent the most significant short-term revenue opportunity in my entire business.

The Potential Impact

• Converting 40–50 extra link clickers =
  $1,125–$2,250/month

• That’s a 25%–50% revenue increase

• With the same audience size

The ROI

• 10 hours of setup.

• 2 hours/month to maintain the database

• Lifetime benefit: up to 50% more revenue, less guessing

With the StackContacts tool, my design goal is to reduce setup time to 2 hours and automate database maintenance to a single “Sync” button click each month.

Conclusion: Start Tracking Right Metrics, Start Optimizing

If there’s one thing this analysis taught me, it’s this:

Creators underestimate how much revenue they’re leaving on the table
because they track the wrong things.

Key Takeaways

1. Link clicks predict product purchases

2. Speed matters — most purchases happen within 24 hours

3. Product guides and “How To” guides convert best

4. Optimize for behavior, not vanity metrics

Your Next Steps

• This week: Add UTMs to Gumroad product links, and upsells on your Gumroad store's Checkout page

• This month: Identify your top 5 converting posts

• This quarter: Build a data-driven content strategy

The Bigger Picture

AI makes this type of analysis accessible to every creator.

You don’t need a data degree.
You need a willingness to look at the numbers.
Minor improvements in the right metrics compound quickly.

This is how you go from “hoping people buy” → predictably generating revenue.

Thanks for reading.

— Finn

---

Join StackContacts Beta - Only 5 Spots Open

I have a limited Alpha/Beta for creators who feel the same pain I did: conversions slipping through the cracks, and no tool that truly puts customer relationships first.

This isn’t for everyone. It’s for a small group of Substack writers who:

• Use a Mac

• Do not be afraid to use a command-line tool on a Mac

• Are curious enough to experiment with AI tools like Claude

• Are you willing to spend several hours setting up your environment to run these tools

• And, most importantly, care about building deeper connections with their audience

If that sounds like you, I’d love to build StackContacts with you.

You’ll get:

• Early access to the creator-first StackContacts CRM tool I’ve been testing on my own Substack

• A head start in building a system that helps you remember, follow up, and stay connected as you grow your business

• Ability to run deep dive analysis like in this article using AI

👉 Join the StackContacts Beta — for creators who want to stop guessing and start making data-driven decisions.

Spots are intentionally limited — I’d rather co-build with a handful of committed creators than spread myself too thin. If you’ve ever felt the pain of slipping connections, this is your chance to fix it before growth overwhelms you.

PS: Other Useful Links

1. If you want more consistency in posting Notes, consider checking out the Substack Pro Studio.

2. If you are publishing consistently, but growth still feels random, the Substack Control Center tool helps you navigate through metrics that matter.

3. DM for help

4. Want to see how I set this up? (see this post)