AI Helped Me Write a Mini Program: After 44 Days Online, 36 People Reached the Checkout, 8 Paid

By 好市民_

On the evening of October 31, I pulled up the backend. ChuCaiJi (Kitchen Ledger) had been live since September 18, 44 days in total. Total revenue: 375.50 yuan.

In the one-month retrospective I wrote about 206.42 yuan. That means in the 12 days after that article, I made another 169 yuan - more than half of what the first 30 days brought in. Sounds good, but the number that really kept me staring this time wasn't revenue. It was 22.2%.

36 membership orders were created, but only 8 paid successfully. In the previous post I said this funnel "needs to be broken down next." These 12 days, it got worse before I could touch it.

Timeline: Eight checkpoints over 44 days

9/18   First version goes live (v1.0.1)
10/06  Ad monetization enabled (day 19, cumulative UV 601)
10/08  AI image generation and text recognition launch, banner ads across the site (v10)
10/12  Membership system launches (v12)
10/13  First membership order arrives; traffic peak 209 people
10/14  Ad revenue single-day 41.80 yuan, eCPM 23.44 (still not beaten)
10/19  One-month retrospective: UV 1,669, total revenue 206.42
10/31  Today: UV 2,726, total revenue 375.50

I went through the git history. 57 commits in 44 days, versions from v2 to v14. Like the one-month post, most commits are bug fixes and copy changes. The real feature work happened in one week, 10/08 to 10/14 - AI image generation, storage expansion incentive videos, membership, merchant needs validation, stickers. Five directions crammed into five days. What happened later proved that week was too packed.

Traffic: Daily average 105, dropping to 80 by month-end

WeChat backend access curve, 9/18 to 10/31:

Average daily visits: 104.95 people. The curve splits roughly into three stages: before National Day it hovered between 30 and 70, early October it started climbing, peaked at 209 on 10/13, then slid steadily, back to around 80 by month-end.

Two details stand out. First, that 10/13 peak was exactly the day v14's login flow table launched - DAU has been calculated by real logins since then. The peak was a coincidence, but that curve stopped being an estimate that day. Second, 10/25 and 10/29 have little bumps - both days I posted on content channels. Post once, bump once, then decline resumes. The organic floor is still those 80 or so people.

In the one-month post I said dropping to 93 on 10/18 made me pause. By month-end, pausing is pointless. This is the norm: after the holiday pulse recedes, the product's natural operating level sits around 100.

Ads: The jagged sawtooth is still there, but I've stopped staring at it

Ad revenue started on 10/06 and by 10/31 totaled 250.30 yuan:

Fetch 96,476, impressions 36,583, impression rate 37.92%, clicks 392, CTR 1.07%, average eCPM 6.84.

I analyzed these two curves in the one-month post. This time, with the full month's shape visible, the conclusion doesn't change, it's firmer: the 41.80 spike on 10/14 dominates the whole month, then revenue sawtooths between 2 and 20 yuan, eCPM falls back to 4-5 with intermittent jumps to 13, 15, but never touches 23 again.

My current routine is to glance at yesterday's revenue once a day, then go about my business. There's no lever I can pull on this money; staring won't change the curve. But the 37.92% impression rate reminds me of something: nearly 100,000 fetches, only a third actually rendered - the ad slots are placed, but more than half of the opportunities never even paint. That's a place I can cut into later.

Membership: Orders created tripled, payment rate dropped to 15%

This is the core of this retrospective, and the part I'm least comfortable calculating.

Full 44-day cycle: 36 orders created, 8 paid, success rate 22.2%, 28 timed out and closed, 0 refunds.

Broken down by time, it looks worse. In the one-month post, the numbers were 9 created, 4 paid, success rate 44.4%. So in the last 12 days: 27 orders created, only 4 paid, about 15% success rate. Order creation tripled - that's good news, it means the ad slot exposure from 10/08 and the membership system from 10/12 really drove entry visibility - but the checkout step barely held anyone.

Looking at the 8 paid orders' distribution:

Package	Orders	Revenue
Changchu annual	2	79.8
Changchu quarterly	1	19.9
Changchu monthly	2	5.8
Qingxiang annual	1	12.9
Qingxiang quarterly	1	4.9
Qingxiang monthly	1	1.9
Total	8	125.2

Two annual cards account for 64% of total revenue. The 4 new orders after the one-month post are all small tiers from 1.9 to 19.9 yuan, 4 orders totaling 30.6 yuan. Translation: people willing to drop 39.9 at once are a tiny minority; most who reach the checkout either don't pay or try the cheapest option.

Zero refunds actually makes me feel solid - it's not buyer's remorse, it's people walking away at the checkout. The problem isn't price backlash, it's what happens in those tens of seconds before payment. I still don't know. Because I still haven't instrumented the funnel.

In the previous post I wrote "break down the payment funnel" as next step, then these 12 days I scheduled it after bug fixes and feedback intake. The data degradation is payback for my prioritization. No excuse.

The money part, one final calculation

Source	44-day cumulative	At 1 month	Last 12 days increment
Ad revenue	250.30 yuan	111.82 yuan	138.48 yuan
Membership	125.20 yuan	94.60 yuan	30.60 yuan
Total	375.50 yuan	206.42 yuan	169.08 yuan

Cost basis unchanged: server 99, certification 30, total 129. Net profit 246.5. The last 12 days earned 14 yuan/day on average, double the first 30 days - modest base, but the trend is real.

User-side increments, aligned with the one-month post:

Metric	10/19	10/31	Increment
Cumulative users	1,669	2,726	+1,057
Cumulative recipes	7,481	11,595	+4,114
Cumulative order placements	310	511	+201
Cumulative ratings	27	52	+25
Current active members	4	8	+4

New users in 12 days exceeded the total of the first 19 days (launch to ad monetization). Recipes added 4,114, meaning each new user actually created something - evidence this product isn't dead yet.

Forty-some days later, I've rewritten one sentence

In the one-month post I wrote "AI takes away the cost of building it, but it can't help you decide whether to build it at all." After 44 days, I want to add half a sentence: It also doesn't stand in line for you.

In those five days from 10/08 to 10/14, I launched AI image generation, membership, merchant validation, sticker integration simultaneously. AI "built" each one, and each was fast. But a person has only one attention budget. The result of shipping five things at once is that the payment funnel deteriorated for 12 days without me noticing - and when I did notice, I had no capacity to address it. AI raised my output to a level where my attention can't keep up. That's the most honest lesson of this month.

Next step is one thing, everything else waits: add payment funnel instrumentation, and go one by one through those 28 timed-out orders. Record time from checkout to abandonment, payment method, failure callbacks - all of it. If I fix this funnel, membership revenue could double; if I can't, I'll admit the membership path is dead for a solo developer, at least for now. I'll use the next 30 days to pay off that account.

Finally

The promised retrospective checkpoint remains: cumulative UV 3,000, or membership reaches 20 orders, whichever comes first - I'll write again. Currently 274 visitors short of 3,000, and 12 paying customers short of 20 orders. At this month's pace, UV will arrive first.

This is the ChuCaiJi home page:

To try it, search for 「厨菜记」 on WeChat, or scan the code for faster access:

First visit auto-creates a default kitchen, pick a few categories and you're in. The order, cook, and rate loop needs at least two people to get lively.

---

One last thing. At the end of the one-month post, I created a WeChat mini-program exchange group. It's now 60+ people, all building their own mini programs - how to survive the eCPM sawtooth, how to break down funnels, where the review red lines are. When someone's stuck, they paste backend screenshots and ask openly. That's much faster than guessing alone in a black box.

If you want in, message me 「进群」 and I'll add you. Also feel free to comment below about which funnel in your product looks the ugliest - I'll reply to everything.