8:42 AM - Inbox triage
I open the Auto-AI dashboard. My coffee is still warm. Six overnight signups from the Facebook groups post I shared yesterday - mechanics in Arizona, Vegas, Sacramento. Good. I mark them for the team to onboard this week.
In the customer email queue, I see three drafts the AI agent prepared. First one is to Marcus Chen at Chen's Mobile Automotive in Portland. His shop used Auto-AI for thirty days, and the agent flagged that his last appointment was March 18th. The draft is gentle: "We noticed your team hasn't booked a job through Auto-AI in a few weeks. Is there something we can improve?"
I read it twice. The tone is right - not salesy, not desperate. I approve it and move on.
The second draft is for Westside Dental in Phoenix. I have to stop. Wrong business type. Someone in our system has been miscategorized, or the intake form had an error. This is the kind of mistake that kills trust. I flag it in Slack to our product person and make a note to myself to spot-check our intake. The AI shouldn't be drafting appointment reminders to dental offices.
The third draft is better. It's to Sarah Kleinfeld at Valley View Auto Repair in Fresno. She signed up seven days ago, imported her customer list, and last ran a batch of appointment reminders yesterday. The agent sees activity and is drafting a follow-up: "Your reminders brought in two same-day bookings yesterday. Here's how you can set up SMS reminders for next week." I approve this one too. She's using the product. This email might help her use it more.
Stripe is open in my other tab. I check last night's revenue: 847 dollars from thirty-four charges. Good Tuesday. YTD we're tracking at 156K. The margins on this are real.
10:45 AM - A flagged conflict
The Slack alert comes through while I'm reviewing the week's pipeline. Red flag. Escalation queue. Dmitri Petrov at Petrov's Collision in Brooklyn scheduled two jobs for the same bay at 1:00 PM today using Auto-AI's booking agent. The system didn't catch it because Dmitri configured two separate appointment types that both map to "service bay 1," but the system saw them as different resources.
This is on me. We didn't validate the bay setup during his onboarding. I pull up his account and send him a direct message: "Hi Dmitri. I caught a double-book for today at 1 PM. I've deleted one of the appointments and sent the customer an automated reschedule. Can you confirm both cars are actually there, or do we need to call the customer back?"
He replies in forty minutes. One car is real. The other customer had called this morning to reschedule, and Dmitri had forgotten to mark it in the old system. The bot picked it up and rebooked automatically. I walk Dmitri through the bay configuration properly this time. It's a fifteen-minute conversation in the Slack channel, but it's the kind of work that keeps churn down.
1:15 PM - Lunch and the metrics check
I order Thai food to my desk and open the analytics dashboard. Today is Tuesday. End-of-month close is in three weeks.
Today's new signups: 11. That's decent for a Tuesday. I check the source tags. Three from Google Ads. Five from Facebook. Two from one of the trade show booth visitors from last month who finally set up an account. One direct - someone probably found us in a Google search.
Weekly signups so far: 38. That's tracking ahead of the monthly pace. If we hit 160 plus this month, we're on a 1,920-customer annual run rate. At ninety-nine dollars average, that's way above the 144K Year 1 midpoint. But I know sign-ups aren't closes. Lots of people kick the tires.
I check conversions. Of the 38 signups this week, 6 have actually used the product. Started scheduling or imported a customer list. Conversion from sign-up to active is around 15 percent, which matches what I saw last month.
Revenue this week: 3,284 dollars from the live cohort. Plus the 847 from yesterday. Plus what should come in today if no one churns.
Pipeline is 124 demos scheduled with prospects over the next fourteen days. The sales team is booked solid.
I eat a green curry. The thing that's hard to explain to people is that this is real business work. I'm not delegating everything to the AI. I'm running a company.
2:30 PM - A customer escalation
An email comes in from Bethany Cross at North Bay Auto Group in Sonoma. Subject line: "Billing issue need to talk to real person."
She signed up two months ago. Four locations. Paid on a per-location basis, ninety-nine dollars per location per month. So 396 dollars a month. In the Stripe dashboard, I can see her last three charges all went through. But her email says the system charged her twice in March for location three.
I pull up her account. March: four charges on the 1st, and yes, a duplicate charge on the 15th. It looks like a system bug from our end - we had an issue with the automated billing retry logic that we patched in April. She got double-billed on one location for fifteen days.
I Slack our finance person to confirm, then I write to Bethany directly: "You're right. We had a billing issue in March that's since been fixed. I'm issuing a credit for the duplicate charge today - ninety-nine dollars - to your account. It'll show on your next invoice. I'm also adding thirty days of free service to location three as an apology. Can you let me know if anything else looks wrong?"
She writes back in twenty minutes. She says thank you and asks if the system will pick up the credit automatically. It won't - not yet. I have to manually note the credit in Stripe. I do that. I also make a note to fix the billing docs so customers understand how refunds appear.
This is the kind of customer work that doesn't scale. It's unglamorous. But it's the difference between a 10 percent monthly churn and a 3 percent churn.
4:15 PM - Code issue
I'm reviewing the list of bugs from this week. One of them is marked high priority: the SMS reminder functionality sometimes drops the customer's name from the template and sends a generic message instead.
It's a parsing issue. I look at the code. Someone in a past version forgot to null-check the customer name field before inserting it into the template string. I write a two-line fix, test it against a sample of last week's messages, and mark it as resolved.
It's the kind of thing I can do in forty minutes because I built a lot of this system myself. Not every operator will have a technical background, but I do, and it saves us a day of developer time when I can make small fixes.
5:45 PM - Wrap
I close the laptop at 6:15 PM. Today:
What worked: Three customer escalations turned into retention and a quick fix that might have cost us two more cancellations. The email approvals are taking me about ninety seconds each. The metrics are trending right. The product is actually being used by people.
What's hard: I'm still the person who handles the edge cases. The billing double-charge. The double-booked bay. The categorization error that sent an email to a dental office. I can't scale on manual work forever. But I also know that right now, six weeks in, I don't have the volume to hire someone else to do this. So I do it myself and I stay up late sometimes.
The AI agent handles the pattern work - the emails to people in churn, the follow-ups to active users, the scheduling optimization. That actually works. But I'm still here, approving, catching errors, talking to Dmitri and Bethany and Marcus. That's the real business.
I'll probably spend the next hour reading customer feedback before I truly step away.