8:42 AM - Inbox triage
I open the dashboard at 8:42, still nursing yesterday's coffee. There's a Slack notification waiting: three new signups overnight, two from the LinkedIn outreach campaign I refreshed last week, one from a Facebook group post our sales person - actually, me, still wearing that hat - dropped into the Small Business Insurance Agents group on Thursday. The third one, Marcus Chen from Blue Ridge Insurance, converted from trial to paid at 2:17 AM. That's forty-seven dollars in MRR I didn't have when I fell asleep.
I pull up the admin panel and scan the "Intake Pending Review" queue. Seven new client intake forms from agents using our platform. These are the templates and follow-up email drafts that our AI engine generates for their prospective customers. I need to spot-check a few before they go live. The AI is usually clean, but I've learned the hard way that "usually" isn't the same as "every time."
First one is from Carol Reyes at Reyes Family Insurance. Her client filled out an intake form for a small business liability quote. The AI drafted a follow-up email that's solid - warm, professional, three questions about the business type that make sense. I approve it. The next three are routine. The fifth one flags something odd in the subject line. It says "We're here to help you protect your assets." The agent is a one-woman shop in Milwaukee who focuses on high-net-worth clients. The phrasing feels generic for her brand voice. I mark it for revision and send her a Slack message in our agent Slack workspace. She's usually responsive.
10:15 AM - A flagged conflict
The Gmail tab chimes. A customer email came in twenty minutes ago, and the Slack alert rule caught it because it contains "refund" or "cancel." It's from Janet Voss at Voss & Chambers Insurance. She signed up six weeks ago, loved the platform for the first month, but her admin just flagged that one of our automated follow-up sequences sent a duplicate email to her high-value client last Tuesday. Not a disaster, but sloppy on her part for not catching it in preview mode, and annoying enough that she's threatening to pause.
I read the email twice. She's not angry, just frustrated. I fire up Linear to check the bug database. Two other agents reported the same duplicate-send issue last month, both in the evening hours, both with multi-part campaigns. I search the system logs in the admin panel and find the root cause: a race condition in our campaign scheduling logic that only surfaces when two follow-up sequences are set to fire at the same time. This is a real bug, not user error.
I open my code editor and fix it - it's a 3-line change in the scheduling queue, adding a mutex lock around the send operation. I test it locally against three test cases, push it to staging, and deploy. Twenty minutes total. Then I write Janet an email: I apologize for the bug, explain what caused it, confirm it's now fixed, and offer her a month free to make up for the frustration. I mention that I fixed it myself because I take these things personally. That's not entirely true, but it is true enough. Then I hit send.
12:30 PM - Lunch and the metrics check
I order a sandwich and pull open our Stripe dashboard. Week-to-date revenue is sitting at four hundred and twelve dollars. That's solid for the second month. Today alone we've added forty-seven. If we hit our pattern, this week will close around five hundred fifty to six hundred. The math is simple: I need to close three new customers a month at one-fifty each to hit my margin targets. Last month was four. This month is trending toward five or six if the LinkedIn outreach continues converting.
The churn rate is my daily worry. Janet's situation isn't full churn yet - she's paused - but it's a leading indicator. I have seventeen paying customers. If I lose two of them this month, I'm back to breakeven on the growth curve. I check the health score dashboard. Thirteen agents are actively using the platform at least three times a week. Four are light users. One, Derek at Mountain View Insurance, hasn't logged in since March 15. That one stings. He closed one deal through the platform, paid for one month, then ghosted. I make a note to reach out tomorrow.
I also check the "agents active in the last 48 hours" metric. Twelve. That's a good sign. The platform is sticky for the ones who use it. The question is distribution: I need to move from selling to early adopters in the insurance space to selling to people who didn't know they needed us.
2:08 PM - Pipeline and a manual handhold
Back to the admin panel. I check the "Scheduled Campaign Reviews" queue. These are prospecting sequences that agents have built using our templates but customized for their own books of business. Before they go live to real leads, I review them to catch tone-deaf pitches or compliance issues. Insurance is regulated, and I can't afford a customer's badly phrased email costing them a license.
One is from Roy Patterson at Patterson & Sons, targeting commercial auto policies. He's good at this - his email is tight, personalized with client names, and hits the right compliance notes. I approve it. The next one, from a newer agent named Jessica Williams, is overly aggressive. She's promising premium reductions that seem unrealistic. I flag it and add a comment: "You'll want to be careful here. Most state regs require you to back these specific numbers with actual quotes. I'd pull the personalized estimates before you send this batch." She'll get the message in Slack.
Also in the afternoon: A customer, Miguel Soto at Soto Insurance Partners, messaged through the chat widget asking how to export client data from our platform. It's not something we've built yet. I explain the current workaround - he can pull a CSV from the reports tab, then import it into his brokerage CRM. I walk him through it in a Loom video, a screencast with narration. I send it to him via Gmail with a note. He responds forty minutes later saying it worked perfectly and asking if we have plans to build a direct Salesforce integration. I tell him yes, it's on the roadmap for Q3, and I add him to the "wants Salesforce" list so I remember to reach out again in June.
4:42 PM - The customer thank-you
A different email arrives. It's from Diane Cho at Cho & Associates, one of my earliest customers. She writes:
"I closed three deals in the last two weeks that I'm confident I would have lost with my old workflow. Your intake forms are cutting my follow-up time in half. I'm running the numbers this afternoon, but I think I'm adding fifteen hundred a month in closed-rate improvement. Wanted you to know it's working. Tell me when you're ready to talk about team accounts."
I read this three times. I screenshot it and add it to a folder on my desktop called "Why I Do This." Then I respond. I ask her about the specific deals, what the intake forms changed about her process, and whether the follow-up sequences were the bottleneck or the initial form itself. She's hinting at upsell - team accounts - but I want to understand her use case properly before I build something wrong.
5:55 PM - Close-out
I scan the Slack channels one more time. No urgent alerts. The agent in Milwaukee who I flagged about the subject line approved my revision and launched the campaign. The LinkedIn lead from this morning, the second one, is now showing up in our "Welcome Email Sent" event log. That means she opened the product tour.
I look at the spreadsheet where I track all my numbers. Week-to-date: five hundred and twenty-three dollars. Current customer count: seventeen active, one paused (Janet, pending her next check-in), one dormant (Derek). Monthly churn rate: 5 percent. Monthly growth rate: about 40 percent. The margins are thin but real. I'm not selling Cadillacs. I'm selling gasoline to people who need to stop wasting it on slow paperwork.
I close the admin panel, then Slack, then Gmail. I stand up and walk away from the desk at 5:55 PM. The work isn't done - it never is - but it's the moment where I stop pushing. Tomorrow there will be new signups to approve, more duplicate-send reports to investigate, customers to onboard, maybe another refund request. Today, I fixed a bug, spotted a customer who was about to churn and turned her around with honesty, closed a contract on honest terms, and helped someone understand what our product actually does.
Second month in, and it still feels like real work. But the work is proportional to the outcome. The AI handles what it handles. I handle everything it can't. And so far, that seems to be the math that holds.