Day 1 operating Tax Planning AI

8:42 AM - Inbox triage

I open the dashboard on my second monitor while my coffee cools in the blue mug from Anthropic's office gifts. The Tuesday morning light is hitting my desk wrong, but I'm not moving.

Three things have happened since I went to bed last night:

1. Stripe dashboard shows two new subscriptions at $199/month each came through Google Ads. Both are small CPA firms in Ohio and Nevada. I won't know if they stick beyond the trial until Thursday, but I screenshot it anyway and drop the image into Slack.

2. The agent has generated four outbound email drafts to CFOs on my list. I open the first one - to Michael Chen at Chen Manufacturing, a $4.2M revenue operation I found through LinkedIn - and the AI's voice feels right. It's specific: references his recent expansion into three new facilities, mentions the tax implications of inventory restructuring in a $4M revenue band, offers the first 30 days free to run a hypothetical scenario. The draft doesn't oversell. I read it twice. I make one change: I add a reference to a specific case study about manufacturing cost segregation that I know will land with his type. I approve it and set it to send at 2 PM, when CFOs in the Midwest are usually checking email.

3. One customer, Patricia Gomez at Gomez & Associates CPA, sent a note at 6:47 PM last night: "Just closed a deal with your system's rec on entity structure. Client saved $23K in taxes. This is real." I don't need this for anything, but I read it three times and sit with it for a minute. That's why I built this. That's not the agent talking. That's a real firm using my product to tell a client something that saves a client $23K.

I put that one in a note document I'm keeping: "Validation moments."

10:15 AM - A flagged conflict

At 9:30 AM the agent flagged something in the system that I needed to review. I only see these when the confidence threshold drops below 85%.

It's Marcus Liu at Liu & Associates. His business is a digital marketing agency doing $7.8M revenue. The agent has been running scenarios on his business for three weeks now, and yesterday it generated a recommendation around qualified business income (QBI) deductions that conflicts with what Marcus told me in a phone call last week. He said his CPA already told him he doesn't qualify for full QBI because of the services business limitation. The agent's system doesn't know that constraint yet because I hadn't manually entered it in his business profile.

This is exactly the kind of moment where the AI is useful but not sovereign. The agent can run numbers, but I'm the one who has to know which numbers matter.

I open a Slack draft to Marcus, and I don't let the agent write this one. I write it myself, acknowledging what his CPA said and offering a specific hour on Thursday to talk through the nuance. I explain that there might be a planning strategy they haven't considered - restructuring his service delivery model slightly to open up QBI opportunities - but I want to make sure it's worth his time before we dig in.

This is the work you don't see in the marketing copy. This is the work that makes the difference between a tool and something real.

12:30 PM - Lunch and the metrics check

Lunch is a sandwich at my desk because I want to check the week-to-date dashboard before my 1 PM call with a referral partner.

Week-to-date numbers as of 11:42 AM:

7 new signups (vs. 4 last week)
$1,348 in revenue recognized today (two subscriptions, one one-time consultation setup fee)
23 qualified prospects in pipeline
Customer churn this week: 1 (a bookkeeper who signed up by accident and realized she needed something else)

The pipeline view is sorted by likelihood to convert. I see Carol Reyes at Reyes Family Practice at the top. She's a $3.2M dental practice, has been in the free trial for 18 days, and opened the email I sent on Sunday with a case study about healthcare practices. The agent predicts 78% likelihood she converts in the next week. I make a note to call her on Thursday to ask what specific questions she still has.

One customer did churn this morning: Keisha Martinez at Martinez Consulting. She'd been a customer for 19 days. I opened her account and saw she'd never logged in after the first login. I sent her a personal note asking what went wrong, not pushy, genuinely asking. The bounce from 19 days to churn is the kind of signal that matters. I'm going to add a check to the onboarding sequence: if someone doesn't log in by day 3, an automated Slack notification hits me.

My referral partner call at 1 PM goes well. The CPA firm I'm working with has four potential referrals for me in the next month. I confirm the commission structure (they get 30% of year-one ARR), and we talk about what kinds of businesses they see that are good fits. That referral channel is the reason I think this can grow past the ad spend math.

2:08 PM - Customer escalation

Derek Thompson at Thompson Landscaping Enterprises calls at 2:07. I see his name pop up in Slack and I take it.

He's a $9M revenue operation and he's been with me for six weeks. He's angry, and I can tell before he says much: his tone is tight. He's saying the monthly fee is coming out of his account and it hit on a day when his payroll went out and he got a double-charge notification from his bank.

This is the kind of problem where the AI can't help. The charge is correct on my end - he did authorize it - but his bank's timing algorithm made it look like a double charge. I open the Stripe dashboard, find his account, and I see the exact issue. One charge on the 8th for $199. The bank notification hit twice but it was one charge.

I walk him through what Stripe shows. I don't make excuses. I apologize for the stress and I offer to have the charge processed differently next month to avoid the same bank timing issue. He settles. He says the product is working and he doesn't want to leave, he just wanted to know we'd handle it right.

I flag this in Linear as a UX improvement: "Show bank processing details to customers in dashboard before they see it through their bank."

4:30 PM - Pipeline review and a small win

I pull up the Gmail thread with Patricia Gomez to re-read that note one more time. "Client saved $23K." I realize I never sent her a thank-you or asked for permission to use her story as a case study. I send an email asking if I can write it up as a case study with her name and the numbers, offering to feature her firm prominently. This is a genuine ask, not transactional. If she says yes, I'll have a real proof point for the next marketing push.

I update the three approval emails the agent drafted this morning and queue them to send. The first to Michael Chen is going out in about 45 minutes. The other two are to dentists - one in Austin, one in Denver - also in the $3M to $5M revenue band. I notice the agent has been increasingly accurate about which service businesses are good fits and which aren't. The pattern recognition is getting tighter.

I walk through the pending signups from today and yesterday. Of the 7 from this week, I personally reach out to 2 of them: high-touch on the ones who came through LinkedIn ads, which means they're CFOs or business owners, not accidental traffic. The first one signs on for a 20-minute call on Thursday. The second one doesn't respond, which is normal.

6:15 PM - Wrap

It's 6:08 PM and I'm running through the checklist before I close the laptop.

Stripe is green. Slack shows no urgent flags. The agent's processing queue for tomorrow is loaded with three more outbound emails. One customer conversation is scheduled for Thursday. One customer is waiting for my follow-up. One customer sent me something that reminds me why I'm doing this.

What worked today: the system moving at velocity while I stay in the loop. Two new customers coming in before noon. A customer with a real problem getting a real solution in real time. A referral partner call that confirmed the secondary motion is working. Patricia's note about that client saving $23K.

What I'd change: I need to stop manually reviewing every outbound email draft. The agent is eight for eight on tone and relevance. I'm confident enough to set the ones that hit my approval threshold to auto-send. That'll free up 45 minutes a day that I'm wasting on things that work.

Derek's bank charge issue is now a Linear ticket. That's a product bug, not an operator problem, and I'm building a list of these.

I close the laptop at 6:17 PM. Tomorrow I have a call with a CPA firm that cold-called me yesterday after reading about the product on Capterra. Growth is not coming from nowhere.