8:42 AM - Inbox triage
I open my laptop on the kitchen table and the Goodword dashboard loads. Seven Slack notifications are waiting. Three are customer responses from yesterday - including one from Carol Reyes at Reyes Family Practice saying the response we drafted to a one-star Google review actually brought her in for a consultation. That one lands different today. The other four Slack alerts are flagged reviews that need human approval before going live. The system is working exactly like it should.
My Gmail is quieter. One support email from Marcus Chen, a restaurant owner in Portland who signed up two weeks ago. Subject line: "Questions about billing." I make a note to get back to him by lunch. The other emails are the usual: two customers replying to the drip campaign emails I sent on Friday, both asking about the 14-day trial. Both look like solid leads.
Week-to-date I've added eighteen new customers. Revenue so far is tracking to $1,650 this week. Two customers have already cancelled - one said the platform was too hands-off, they wanted templates they could customize more, and one's restaurant closed. The churn stings more than I expected. But the pipeline report shows forty-two businesses still in trial, and the conversion rate from trial to paid is holding at around 35 percent, which is better than the models I ran before launch.
9:15 AM - The approval queue
I click into the dashboard's flagged responses section. Goodword has drafted replies to 146 new reviews overnight across all my active customers. Of those, nine didn't pass the confidence threshold and are waiting for me to either approve them as-is, edit them, or reject them entirely.
The first five are straightforward. A four-star review for David Park's dental practice in Sacramento that the AI thinks was lukewarm. The draft response is genuinely good - acknowledges the gap, thanks them for the feedback, offers a specific solution around appointment scheduling. I approve it. The next three are similar quality. Click, click, click. I'm through them in five minutes.
The fourth one makes me pause. It's for a Vietnamese restaurant in Fresno. A customer left a one-star review saying the food was "authentic but too salty." The AI's draft response is sharp. It says: "We appreciate your honesty. We use traditional recipes, but we're always listening. Come back and ask for less salt - we'll make it your way." I read it twice. It's collaborative, not defensive. It's exactly what I'd want someone to write if I left a review like that. Approve.
The last four all come from the same customer - a dental office in Las Vegas. Three of them are decent. The fourth one is weird. A two-star review that just says "waiting room was cold." The AI drafted a response that says "We're sorry you felt uncomfortable. We'll adjust our temperature settings." It's the kind of response that's technically fine but sounds robotic. I edit it manually: "Thanks for letting us know. We'll bump the heat - our team runs cold anyway." Hit approve. It's a small thing, but I can feel the difference.
10:50 AM - A real conflict
I notice a red flag alert in the dashboard. A response has been flagged as potentially too aggressive. It's from a real estate office in Phoenix. A one-star Google review from someone who says: "They sold me a house in a neighborhood that's gone downhill. Bad decision."
The AI's draft response is trying to be helpful, but it crosses a line. It says: "We're sorry you're unhappy with your neighborhood choice, but our job is to find properties that match your criteria, not predict neighborhood trajectories. We did our part." It's defensive. It's technically accurate but tone-deaf. This is the exact kind of thing that would make someone angrier if they read it.
I delete the draft and write a new one from scratch: "We understand this is frustrating. While we can't predict neighborhood changes, we're here to support you. If you'd like to discuss alternatives or next steps, please reach out." It's shorter, it's humble, it holds the line without being sharp. I approve it and move on.
12:35 PM - Lunch and the run-rate check
I open the Stripe dashboard while eating a sandwich. Today's revenue so far is $348. I can see the line of payments from yesterday - twelve customers at $49 each, two at $99 (annual). The recurring revenue is actually sticky so far. Only two refunds this week, both from the free trial period.
I click over to the metrics dashboard I built in Google Sheets. Week-to-date: eighteen new signups, 6 churn, 12 net new, $1,650 revenue. If I can hold this pace for a year, I hit the $108k ARR target. That's the whole business model right there.
I text back Marcus Chen, the restaurant owner with the billing question. He's asking why he was charged twice in his first month. He wasn't - what happened is the trial ended on the 15th, we charged him for the month, and he re-triggered a trial somehow and got charged again on the 23rd. I explain it in an email and offer a credit for the duplicate charge. It's about $49. I issue it manually through Stripe. Problem solved, but it shows me I need a clearer trial-to-paid transition in the product. I make a note to fix it.
2:15 PM - Customer success
Carol Reyes emails me back. She says the one-star review that worried her turned into actual business. The person came back to her practice, and they booked two fillings. She's sending me a photo of her team. I read it three times. This is why I built it. I hit reply and tell her I'm thrilled for her, that this is exactly what we're here to do, and that if she ever needs anything, to reach out directly.
That email takes ten minutes. It's the kind of thing the system can't automate, and I don't want it to.
3:45 PM - The edge case
Back in the dashboard, I notice something. One of my customers - a dental practice in Mesa - has had responses go out to the same person twice. Once on Monday morning, once on Tuesday morning. The customer left a three-star review. Goodword sent a draft response, it got approved, it went live. Then forty-eight hours later, a new three-star review from the same email address came in (different content, actually), and we sent another response.
It's not wrong, technically. But it's noticeable. I call the customer directly. Lisa at Desert Smile Dental picks up. I tell her what happened and that I'm going to add a check to prevent duplicate responses to the same customer within thirty days. She says it's fine - the customer actually appreciated the follow-up. But I'm already planning the fix. This is the kind of thing that matters for retention.
4:45 PM - Pipeline and the week ahead
I open Linear, where I track feature requests and bugs. Seventeen tickets open. Three are blocking: the duplicate response thing I just found, a billing automation gap for customers who downgrade mid-month, and a request from two customers asking for Yelp auto-response scheduling (right now everything goes out immediately).
The second month matters for retention. I can feel it. I prioritize the duplicate-response bug for Friday. The Yelp scheduling feature is lower priority - it's nice-to-have for most customers. The billing edge case is the one that keeps me up. I need to fix it this week or I'll risk churn.
I pull up my simple pipeline in a Google Sheet. Forty-two customers still in the fourteen-day trial. Twelve of them have actually integrated Goodword into their workflows (they're using it more than once a day, which is the signal I watch). Of those twelve, I estimate eleven will convert. The other thirty? I'm not sure. Some of them signed up, looked at the dashboard once, and disappeared.
5:50 PM - Wrap
I close Linear and look at the time. It's almost six. I've been on since eight-forty and there's no hard line where this work ends. I could spend the next three hours optimizing response templates or building better reporting. Instead, I close the laptop.
What worked today: the approval flow is smooth. Four customers happy. The metrics are holding. The core product is doing what it's supposed to do - draft good responses, save people time, occasionally change the outcome of a customer interaction.
What's hard: the churn is real. Even with an excellent product, some people don't stick. Marcus Chen's duplicate charge never would have happened in a more mature product. The pipeline is uneven - some customers are all-in, others are ghosts.
What I'd change: I need better onboarding for those forty-two trial customers. I need to call more of them, not just wait for them to use the product. I need that billing automation fix yesterday. And I need to ship something that makes this job less binary - either we're perfect and the customer loves us, or we're not and they leave.
But also: Carol Reyes brought in new business because of something Goodword drafted. That's real. That's the game. Tomorrow is Wednesday, and I've got work to do.