8:42 AM - Inbox triage
I open Slack first, before coffee. The integration with our signup webhook pinged four times overnight. Four new trials registered between midnight and 6 AM. I jot down their names in the open doc I keep for this - Marcus Chen, Reyes Family Practice, Blue Ridge Inspection Group, and someone named Donna who filled in the phone field but left the business name blank. I'll follow up with her by 9 AM. Three of the four are likely solid. Donna might be a curiosity click.
The Slack channel also shows our nightly ingestion job completed clean. No errors. The system processed 14 inspection reports yesterday, which means 14 of my trial customers or paying users submitted field notes and photos and got a first draft back from the model. I won't see the details until I open the admin UI, but that baseline feels good.
I pull up Gmail and scan for anything that came in after I closed the laptop at 6:15 PM yesterday. Two customer emails. One is from David Park at Park Home Inspections in Ohio, a paying customer since day 18. He wrote: "Got a report back that was 95% perfect. Fixed two sentences and sent it to the client same day. This is exactly what I needed." I leave a short reply thanking him and asking if there's anything I can improve. These notes matter. They keep me calibrated on what's actually working.
The second email is a billing question from someone named Carol Reyes at Reyes Family Practice. She's on her second week of a trial and got charged an unexpected transaction yesterday. I open the Stripe dashboard and pull up her record. Ah. She hit the trial-to-paid conversion boundary and the system double-charged her for one day of overlap. That's a known issue - I have it flagged in Linear as medium priority. I refund her 99 dollars and send her a manual email explaining what happened. "I apologize. This is on us. You're covered for a full month from yesterday. Thank you for catching it." She's not churning yet. Prevention.
10:15 AM - A flagged conflict
The admin dashboard loads. I click into the pending-review queue and see six reports waiting for my sign-off before they go back to the customer. The model drafts them; I verify for tone, accuracy, and any hallucinated details. Most are solid. Five I approve immediately. One has a flag.
It's from Jennifer Wu, who runs a small solo inspection practice in Seattle. She submitted photos and notes from a residential inspection yesterday. The AI drafted a comment about "possible structural compromise in the east-facing wall" based on a misread of one of the photos. It wasn't a structural issue - it was a shadow from the deck railing. The model got confident about something it shouldn't have. I remove the line, add a note to Jennifer's report that says "Water staining detected in east wall - recommend closer inspection," which is accurate and safer. I hit approve and leave her a note in the system.
Then I do something I know I should automate but haven't yet. I send Jennifer a direct email: "Caught something in your draft - fixed it before it went out. The shadow on the deck railing was throwing off the model. You're all set." She doesn't need to know this detail, but I think it's worth her understanding that a human is still in the loop. Trust matters at 99 dollars a month.
11:30 AM - Metrics and the pipeline
I open a spreadsheet I update daily. It's crude, but it tells me what's actually moving:
- Week-to-date new signups: 12
- Conversions this week: 2 (David Park converted yesterday, someone named Rodriguez converted on Monday)
- Active paying customers: 8
- Week-to-date MRR: 792 dollars
- Current run rate if I hold steady: about 3,400 per month
I'm not at the 200-per-month target number yet, but I'm on the arc. I have a backlog of 18 active trials still in their 30 days. If even one more converts this week, I hit nine customers. At that pace, month two ends around 10 or 11.
I close the spreadsheet and open Slack again. I post a message in the outbound-thread: "Week-to-date: 12 trials, 2 conversions, 792 new MRR. Feeling steady." It's a small ritual. Helps me stay grounded.
1:15 PM - The churn notification
I'm between a sandwich and a coffee when the Slack alert fires. One of my existing customers, Thomas Berkley at Metro Inspections in Denver, just canceled his subscription through the Stripe customer portal. He was on my books for 37 days. I don't know him - I've never talked to him. He didn't send a cancellation email. He just left.
I pull his account and review his report history. He ran four inspections in our system. Got four reports back. No complaints in our email record. I send him a brief email: "Thomas, I see you've canceled. If there's anything we can improve or if you hit a snag, I'd love to hear it. I'm here." He probably won't respond. But I want the data.
I update the spreadsheet: 7 active customers again. The math resets.
3:45 PM - One bug fixed, one email answered
A customer named Marcus Chen, one of the overnight signups, sent a support email. He uploaded a photo that was sideways, and the system failed to process it cleanly. The report came back with missing sections. I recreate his upload scenario in a test account and confirm the issue. It's a rotation-handling bug in the image preprocessing. I file it in Linear, tag it high-priority for tomorrow, and send Marcus a reply: "Found the issue. Working on a fix. In the meantime, if you rotate your photos in your phone's camera app before uploading, we can process them correctly. I'll have this patched by tomorrow afternoon."
He replies within 20 minutes: "Thanks for the quick response. Rotated the photo. Got the full report back. Perfect. When does the trial convert to paid?" I don't know what his conversion rates are yet, but his tone tells me he's going to stay.
I make a note: rotation bug and commit to fixing it first thing tomorrow.
4:30 PM - Pipeline audit
I open the InterNACHI forum thread where I've been dropping our value prop once or twice a week. I have it bookmarked. The thread is about time management and inspection report efficiency. I left a comment six days ago about InspectionAI. It has 14 replies and 47 likes. Three of those replies are from people interested in the tool. Two of them asked for a trial link in DM. I haven't reached out yet. I send them both personal messages, keep the tone casual, include the trial signup URL.
I also scan the Facebook home inspector groups where our main GTM motion lives. One group moderator flagged our last post for being too promotional. I pulled it down and reframed our next post as a question instead: "What part of an inspection takes the longest to write up?" Twelve comments so far. No sales pitch. I'm learning that this audience recoils from directness. They want to discover the tool.
6:15 PM - Closing the laptop
I close the admin UI, close Stripe, close Gmail. I check Slack one final time. No new alerts. The overnight ingestion will run again in six hours. Fourteen more draft reports, probably. Fourteen more moments where the owner of this business - me - will decide whether the machine did its job or needs a human override.
David Park sent me a thank-you note that I'll still be thinking about when I wake up. Carol Reyes probably feels seen because I caught her billing issue before it became a churn. Marcus Chen is going to convert, I'm fairly confident. Thomas Berkley canceled for reasons I'll never fully know, but I left the door open.
This is real work. The AI doesn't run itself. But it runs so much faster because of what it does, and that's the whole point. Two months in, I'm holding eight paying customers and a waiting list of trials. The model makes mistakes. I catch them. The customers notice that I exist. I think that's how you get to month three.