The Workflow
I ran a real purchase. $49, live money, my own card, buying my own Meeting Intelligence Kit the way a customer would. I wanted to see what actually shows up on the way in, not what I assumed showed up.
Step 1 — Buy it for real: Bought the kit through the actual checkout. No test mode, no shortcuts.
Step 2 — Read what arrived: One email landed. A generic welcome from the membership platform, naming a different kit than the one I’d just paid for. Reassuring.
Step 3 — Trace the ownership: I had Claude find out why. The membership platform’s welcome email is scoped to the whole membership collection, not the individual kit, because that’s how the payment gateway had to be wired months earlier to make the storefront unlock work. Correct architecture. Wrong customer experience.
Step 4 — Check the other system: The kit is also supposed to trigger a separate, kit-specific sequence in my email platform. I had Claude check that sequence directly against the live purchase. It fired. On time, to the second.
Step 5 — Find the actual gap: The email inside that sequence had a one-day delay built into it, written back when the copy assumed “you bought this yesterday.” The kit-specific email existed. It worked. It still left a buyer with nothing useful for a full day.
Step 6 — Fix it, then check for repeats: Claude rewrote the email, dropped the merge fields that only resolve on the other platform, and changed the delay to zero. Then checked Kits #3 and #4 for the same bug. Same bug, both of them. Same fix, both of them.
One kit’s purchase experience is fixed. The other two are fixed the same way. And checking them surfaced something worse: neither sequence has ever added a single subscriber, which means nobody has confirmed the automation fires for those two at all.
What Broke
Nothing was down, which is what made it hard to find. Every system involved did exactly what it was built to do. If I’d asked Claude to just check whether the automation was active, it would have come back clean: sequence live, trigger firing, no errors. That’s a true statement, and it’s also useless, because it isn’t the same question as what a customer actually receives. The one-day delay and the wrong-kit welcome email only showed up because I bought the kit myself and sat in the inbox waiting for what a real buyer gets.
Once I could see the actual gap, the fix was routine. I had Claude rewrite the email, change the delay from one day to zero, swap “you bought this yesterday” for “you just bought this,” and drop the {member_password} and {login_url} merge fields, since those only resolve on the membership platform and show up blank in ConvertKit.
AI is good at telling you a system is working. It can’t tell you whether working feels right to the person on the other end, because it has never been that person. The check that caught this wasn’t a smarter prompt or a better audit. It was going through my own checkout like a stranger would, and noticing what a stranger would notice.
Time Ledger
- Time saved: Not the axis this week. Once the actual gap was visible, Claude fixed all three kits’ delay bugs in one sitting.
- Time added: About 25 minutes end to end: buy the kit, read what arrived, trace it, fix it, check the other two.
- Net: The real number isn’t minutes. It’s 51. Fifty-one days between Kit #2 going live and the first time anyone actually bought it the way a customer would.
Fifty-one days is a long time for a gap like this to sit unnoticed, and it sat there because “the automation is active” and “the automation feels right to a buyer” get checked by two completely different methods, and only one of them is fast to run.
The Prompt File
You are auditing a purchase or signup experience that's delivered by more than one connected system (payment processor, membership platform, email/CRM tool, or any automation linking them). I need to know whether what these systems technically do matches what a real buyer actually receives, not just whether each system reports itself as working.
SYSTEMS INVOLVED: [list each one and what it's responsible for, e.g. "ThriveCart — payment," "ProductDyno — membership/login," "ConvertKit — onboarding emails"]
PRODUCT OR OFFER: [PRODUCT_NAME]
**Step 1 — Map ownership:** For [PRODUCT_NAME], list every system involved in delivering the purchase experience and exactly what each one is supposed to send or trigger. Flag anything where two systems might both assume the other is covering a piece.
**Step 2 — Trace the live automation:** Check the actual automation, sequence, or webhook connecting these systems for [PRODUCT_NAME], not its documentation. Report what it's scoped to, any send delays, and any copy that assumes a specific timing ("you bought this yesterday," "next week you'll," etc.).
**Step 3 — Flag inherited assumptions:** Look for merge fields, delay values, or scoping decisions that were correct when an earlier fix was made but may not still be accurate. Name the fix that likely introduced each one, if you can find it.
**Step 4 — Check for siblings:** If [PRODUCT_NAME] is one of several similar offers using the same wiring pattern, check every other one for the identical issue. Report each one separately. Don't assume a fix found on one applies to the rest until you've verified it.
CONSTRAINTS: Do not report "the automation is active" or "no errors found" as evidence the customer experience is correct. Those are different questions. Flag anything you cannot verify without an actual purchase or signup, and tell me plainly that I need to go do that myself before this audit is complete.The Owner’s Question
Whoever owns this workflow, including you if it’s just you: when did you last go through it yourself, start to finish, as the person on the receiving end?
Most people who own a process can tell you whether the automation behind it is running. Almost nobody can tell you whether it feels right to the person on the other side of it, because that answer doesn’t live in a dashboard or an audit log. It only shows up when someone actually sits in the seat of the customer, the new hire, or the buyer and experiences the handoff the way they would. Solo or corporate, the reason this gets skipped is the same: checking that the automation runs takes two minutes, and going through it as a stranger takes twenty.
The fix isn’t a bigger QA checklist. It’s putting “go through it yourself” on a real calendar, on a real cadence, the same way you’d schedule a recurring task. Once a quarter, whoever owns the process actually becomes the customer for ten minutes and reports back what they noticed that the system didn’t.
Field Notes
- The ProductDyno welcome email’s subject line is hardcoded to “Your Solo Department Head Starter Kit,” Kit #1’s name, no matter what someone actually buys. A Marketing Vault purchase gets that subject. So does everything else. I caught it by accident, mid-session, while already annoyed about the other bug.
- “Delay: 0 days” in ConvertKit doesn’t mean instant. The sequence fires on a fixed clock, 11am Eastern, so a purchase at 11:01am still waits almost 24 hours. Zero is a setting, not a promise.