8:47 AM - Inbox triage
I open the Almsbury dashboard and the first thing I see is Slack pinging me with overnight alerts. Three new trial signups came in from LinkedIn outreach, and the Stripe webhook logged them correctly. $0 revenue yet from these three, but two of them filled out detailed intake forms, which is good. The third one, from a small homeless services nonprofit in Denver, just signed up and hasn't logged in yet. I make a mental note to check on that one by end of week.
I pull up Gmail and scan the inbox. Seven customer emails since I left yesterday at six. One is a thank-you note from Carol Reyes at Reyes Family Practice, a small health equity nonprofit in Oakland. Their grant application went out last week and they scored 87 out of 100 on the funder's preliminary review. That's exactly the kind of win I need to see early and often. I file that one into a folder called "love" where I keep customer testimonials. I'll weave that into a case study in a few weeks.
The other emails are a mix: one is a question about whether Almsbury integrates with Salesforce (it doesn't, and I need to add that to the FAQ), two are password resets, one is a churn notice from a nonprofit that ran out of grant cycle funding and is pausing their subscription, and the last one is from Marcus Thompson at Thompson Community Development Corp asking about a refund. That one makes my stomach drop a little. I flag it and set a reminder to investigate.
10:22 AM - A flagged conflict
I walk through my review queue in the Almsbury admin dashboard. This is the engine's nerve center. There are six grant applications waiting for me to approve before they get marked as "ready to submit" and sent back to the users' inboxes.
The first five are clean. The system's drafted them correctly, the tone is right, and the budget justification aligns with the nonprofit's stated mission. I approve each one with a click.
The sixth one is from Jennifer Wu at Westside Youth Initiative. They're applying for a $50,000 workforce development grant. The application is strong, but I notice the AI has made an assumption about their target population size. It estimated 400 youth served annually, but the latest form Jennifer submitted last month says 650. The application text reads smoothly, but it's now inconsistent with their own intake data.
This is exactly the kind of thing that would get flagged if Jennifer submitted it as-is. I don't approve it. Instead, I open the editor, flag the discrepancy in the notes, and send Jennifer a direct email explaining what I found. It takes me seven minutes. The email is short and specific: "Hi Jennifer, loved the direction of this application, but I caught a number mismatch on page two that we should fix before you submit. Your intake form shows 650 youth served, but the narrative says 400. Easy fix. I've flagged it in your dashboard. Let me know if you want me to adjust it, or if I'm missing context."
That's the real work. The system generates the draft, but I'm the quality gate. Without this step, Jennifer would submit something that contradicts her own data, and the funder would either ask for clarification or dock her score.
12:18 PM - The metrics check
I take a late lunch, leftover chicken and rice at my desk, and pull up the Almsbury admin analytics tab. Week-to-date, we're at $2,847 in MRR from active subscriptions, plus $423 from the three trial periods that converted to month-to-month plans. That's on track.
Today specifically: three signups, one new monthly conversion (Jennifer Wu is upgrading from her trial to annual billing at $1,248 per year), and one churn. The churn one is the homeless services nonprofit from Denver. They lasted three weeks, so I know it wasn't a product issue. They probably ran out of planning bandwidth. That stings a little, but it's not unexpected with understaffed nonprofits.
I open Stripe to check the payment processing. One failed payment from a nonprofit in Chicago. Their card was declined. I add a note to my task list to reach out and get an updated card on file. These manual follow-ups are tedious but necessary. I've learned that automated retry emails work maybe half the time. Better to call.
2:34 PM - Marcus Thompson and the refund
I finally follow up on Marcus's email. I open Slack and message him directly to set up a quick call. We connect via Zoom. I almost never do long email back-and-forths with customer issues anymore. It's faster to talk.
Marcus's issue: Thompson Community Development Corp has been with us for six weeks. They used the tool heavily for the first three weeks, generated three applications, got one funded, then didn't log in for two weeks. Now he's asking for a refund on this month's subscription because he feels like they've got what they need and don't need it anymore.
I listen to his story. He's not angry. He's thinking like a nonprofit CFO, which makes sense. I explain our money-back guarantee is within seven days of purchase. They're now at 42 days. I also talk through what they did get: the one funded application paid out at $25,000 and probably would have scored lower without Almsbury's feedback on the narrative. That ROI is good, even if they don't renew.
We land on this: I offer them 50 percent off month-to-month if they come back for their next grant cycle, probably Q3. Marcus agrees. I update his subscription in Stripe immediately, and he promises to reach back out in July. He's not a lost customer. He's a hibernating one. That's a win in my book.
3:15 PM - The bug
I open Linear, where I track issues. There's a bug report from last week that I've been putting off. A customer noticed that when they upload a PDF of their existing 990 form, the system sometimes misreads the organizational revenue figure and inflates it by a factor of 10. I've been dreading this one because I'm not sure if it's a problem with the PDF parser, the OCR library, or the prompt I'm sending to Claude.
I dig in. I pull the test case PDF and upload it through Almsbury myself. The system overestimates the revenue. I check the debug logs and trace the issue. It turns out the OCR is correctly reading "Revenue: $1.2M" but the prompt I wrote is asking the model to extract it as a raw number, which is causing it to sometimes return 1200000 instead of 1200000. It's a phrasing problem, not a model problem.
I adjust the prompt to be more explicit: "Extract the organization's total revenue as reported in the 990. Return only the numeric value in dollars, with no formatting." I test it three times with different PDFs and it works. I deploy the fix and close the issue in Linear. Total time: 23 minutes. Small win.
4:52 PM - Pipeline and close-out
I check my LinkedIn outreach. I've got a warm lead from an AFP conference last month, someone named Sarah Chen from a housing nonprofit in Portland who said maybe when I demoed Almsbury to her. I send her a quick message: "Hi Sarah, following up on our chat at AFP. I know January was busy. When you're ready to explore grant writing tools, would love to show you how Almsbury helped us get to 87-point applications in Q1. No pressure, let me know if this spring is a good time."
I review the trial funnel. Out of the three signups from this morning, one has already logged in and started their intake form. That's good conversion velocity. I make a note to check on the other two by Thursday if they haven't moved.
I close Stripe, close the analytics dashboard, and shut down Linear. It's 5:47 PM.
5:51 PM - Reflecting
I lean back and think about the day. The work is real. It's not set it and forget it. I spent time fixing a bug, handling a customer escalation, and quality-checking AI-generated content. But the system moved: three new customers came in, one converted cleanly, one became a future deal instead of a lost one, and I built goodwill with a testimonial from Carol.
If I scale this from one owner-operator to two people, me plus an ops person to handle customer emails and trial follow-ups, the economics work. That's when the leverage kicks in. Right now, at month two, it's still me doing the customer work, the quality control, the bug fixes, and the pipeline. But the AI is handling the grunt work: drafting applications, responding to customers with insights, running the system.
I wouldn't change the core product or the go-to-market. What I might change: I'd invest earlier in a simple outbound email sequence to nonprofits who've used free trials but haven't converted. The manual follow-up on those three Denver people was probably too late to save.
I close my laptop at 6:04 PM. Tomorrow, more of the same, but if Jennifer Wu's email comes back with a fix, I'll have another clean approval. That's the rhythm.