Day 1 operating Referral AI · Wishdeal Factory

8:42 AM - Opening the dashboard

I open my laptop at the kitchen table with a cold brew. The Slack notification from overnight shows three automated alerts: two new trial signups came through the Shopify App Store listing, one customer renewed their annual plan. I click through to the Referral AI dashboard.

The numbers from yesterday: 47 new referral requests sent by the system, 8 customers interacted with them, 2 actual referred signups came in. Not spectacular, but the pipeline is $1,240 in MRR from those two alone if they convert. Today's target is to hit 50 referral sends and review whatever the engine flagged overnight.

My inbox has 12 new messages. I scan past the Stripe receipts and Shopify notifications. There's an email from Marcus Wu, a paid customer who's been on the platform for six weeks, asking about the referral reward tiers. That's a manual response. Below that is a red flag from the system: the AI drafted an outreach email to Carol Reyes at Reyes Family Practice and marked her as a "high engagement" referrer, but when I look at the details, she downgraded her plan two weeks ago. The system didn't catch it. I add a note to fix that query logic.

10:15 AM - The flagged draft

I pull up the email the engine composed for James Chen at Altimetry, a SaaS metrics startup. The draft reads: "Hi James, we've noticed you've referred two customers successfully. Those two referrals brought in $4,200 in annual revenue. Want to keep going. Check your referral dashboard for your reward balance and let's talk about scaling this."

Three problems: The grammar is broken at "Want to keep going." The math is wrong. He referred one customer, not two. The tone is too pushy for someone who cares about precision.

I rewrite it in Gmail: "Hi James, I noticed you referred a customer last month and they've been active. That adds about $2,100 in ARR for us. I wanted to check in directly instead of letting the system handle it. If you know others who might benefit, I can make sure they get the right tier for their size. Let me know if you have questions."

I approve the rewritten version and send it manually. That's 15 minutes this morning just on email quality. This is the constant work nobody talks about. The engine handles volume. I handle the handshake.

12:30 PM - Metrics and the hard conversation

Lunch is a sandwich at my desk. I open Stripe to see where we stand year-to-date.

Current MRR: $3,680. Week-to-date revenue: $1,094. That's better than last Tuesday. I cross-reference with the dashboard. Four new paid trials started yesterday, three converted from Product Hunt upvotes last week. The CAC looks reasonable if they have a 12-month payback window.

Then I notice something: Rachel Kim, a customer we onboarded eight weeks ago at $149 per month, has a failed payment notification pending in Stripe. I click through. Her card expired and she didn't update it. It's now been three days. She hasn't opened her dashboard in a week. I know what that pattern means.

I send her a manual email instead of letting the automated reminder handle it: "Rachel, your card on file expired. Want to help troubleshoot, or should I pause your referrals while we sort it out." Small message, but it matters. Sometimes a real person reaching out stops a churn.

2:08 PM - An escalation

Rachel responds within an hour. New card is updated. But she asks a question I can't answer with a canned response: "Is it possible to exclude certain customer cohorts from the referral requests. I don't want to ask my smallest customers to refer because they aren't stable."

This is a product question that touches on the engine's logic. I log into the admin panel and trace through the customer segmentation rules. Turns out we can set filters by MRR cohort, and Rachel just didn't see the advanced settings in her onboarding.

I record a two-minute Loom video showing her exactly where the toggle is and send it over: "This should let you target only customers over a certain spend threshold. Holler if you hit other edge cases." Another 20 minutes where the difference between me and a pure SaaS product is that I noticed she didn't read the docs, and I fixed it instead of assuming she would.

4:30 PM - The outbound push

I spend 45 minutes on LinkedIn identifying three ecommerce founders in the $5-15M ARR range who aren't in our trial system yet. I write personalized messages referencing their recent funding round or feature launch. One mentions Finley Cosmetics and their new loyalty tier: "That kind of cohort usually gets 30-40 percent referral adoption once they see who their advocates are."

Stella Okonkwo from a supplement company replies within 10 minutes asking for a demo. I schedule it for tomorrow and add it to Linear as a task. That's probably one of my 2-3 monthly conversions starting.

Back to the dashboard. I check the referral pipeline: $6,840 in potential monthly revenue from customers who opened the referral request last week but haven't closed yet. I approve the automatic reminder sequence and scan the customer names to make sure nobody's getting spammed. All clean.

6:15 PM - Closing

I close Slack and Gmail. Before I shut down, I pull up the weekly numbers one more time.

This month to date: $10,420 in revenue across 34 paying customers, 16 new trials, 4 conversions to paid. That tracks to about $44,000 ARR at this run rate. Not bad for month two of owning it. The Product Hunt launch is pulling, but it's slowing. I need to get better at the LinkedIn side.

One customer, David Park at a fitness app called Vitality, sent an unsolicited thank-you email today: "Our referral rate climbed to 8 percent after two weeks using this. My team actually likes sending these out instead of hating a sales process." That's the good part. That's why I'm doing this.

The hard part is that I worked a solid seven hours today, and I barely left the dashboard. The system does the heavy lifting: it identifies customers, drafts emails, tracks conversions, sends reminders. But I'm here for the signal in the noise. I'm the editor who makes sure volume doesn't mean garbage. I'm the person who remembers Rachel Kim's card expired instead of watching her churn silently.

By August, I want to hire someone to handle half of this. By December, I want the engine to handle 80 percent of the outreach with me just reviewing the exceptions. For now, I'm the amplifier. The system is the megaphone.