Frame
- Goal: Break out of three flat months and find which of traffic, conversion or repeat purchase is the weak link.
- Constraints: About 5 hours a week, $300 a month, and a small shop with no mention of a team.
- Decision to make: What to try first, and how to tell within a few weeks whether it worked.
- Out of scope: A rebrand, a new platform, and wholesale at scale.
I couldn't look anything up, so the research below rests on what you told me and general e-commerce knowledge. I've used no outside statistics.
Research
| # | Item | Status | Source or how to check |
|---|---|---|---|
| 1 | Sales are flat for 3 months (roughly Jul–Oct) | Known | You said so |
| 2 | Budget is $300 a month and time is 5 hours a week | Known | You said so |
| 3 | It's late October, so the gifting season (Nov–Dec) is close. Candles usually sell best then | Assumed | Compare last year's monthly sales, if you have them |
| 4 | You haven't yet split the problem into traffic, conversion and repeat | Known | You said you're unsure |
| 5 | Your store platform's analytics (Shopify, Etsy, etc.) can give sessions, conversion rate and repeat-customer rate | Assumed | Check the dashboard. Most platforms show all three |
| 6 | The three causes need different fixes, so guessing wrong wastes the budget | Assumed | General principle |
| 7 | Where your current customers come from (search, Instagram, email, marketplace) | Unknown | Traffic-source report |
| 8 | How many past customers you have, and how many bought more than once | Unknown | Order export, 30 minutes |
Flat sales may not be a problem. If the same three months were flat last year, you may just be in the seasonal lull. Checking last year's figures is the first job.
Options
| # | Option | How it works | Main way it fails | Lens |
|---|---|---|---|---|
| 1 | One-hour funnel diagnosis | Pull sessions, conversion rate and repeat rate for the last 90 days and compare with the prior 90 and last year | Data is thin or tracking is broken | Smallest version |
| 2 | Email past customers | Send a "holiday order early" note with a small perk to everyone who bought before | List is tiny | Quiet option |
| 3 | Post-purchase email flow | Automate a thank-you, a care tips email and a reorder nudge at about 6 weeks | Takes setup hours, and the payoff is slow | Time shift |
| 4 | Gift bundles and sets | Package 3 candles as a gift set at a higher average order value | Doesn't help if traffic is the problem | Combine |
| 5 | Product page fixes | Better photos, burn time, scent notes and reviews placed above the fold | Conversion may already be fine | Inversion (what makes a visitor leave?) |
| 6 | Small paid ads test | Spend $150 on Meta or Google retargeting and interest ads aimed at gift buyers | Money burned if the page doesn't convert | Remove a constraint |
| 7 | Local and craft markets | Do one holiday market to get names and emails | Costs hours, and the stall fee eats budget | Borrow from another field |
| 8 | Micro-influencer swaps | Send free candles to 10 small creators in home or wellness | Low response rate, and it's hard to track | Change who does the work |
| 9 | Ask 10 customers why they bought | Short message or call, then reuse their words in your copy | Small sample | Quiet option |
| 10 | Subscription or "candle of the month" | Recurring revenue from your best customers | Too much work before the holidays | Extreme user |
Shortlist
Scores run 1–5, and higher is always better. Effort and risk are scored so that 5 means low.
| Option | Effort | Risk | Impact before holidays | Speed | Total |
|---|---|---|---|---|---|
| 1 Funnel diagnosis | 5 | 5 | 3 | 5 | 18 |
| 2 Email past customers | 4 | 4 | 4 | 5 | 17 |
| 5 Product page fixes | 3 | 4 | 4 | 3 | 14 |
| 4 Gift bundles | 3 | 4 | 4 | 3 | 14 |
| 6 Paid ads test | 3 | 2 | 3 | 4 | 12 |
| 3 Post-purchase flow | 2 | 4 | 3 | 2 | 11 |
- Funnel diagnosis: it costs nothing, takes an hour, and tells you which of the other options to spend money on.
- Email past customers: it is the fastest and cheapest way to get sales, and it tests repeat purchase directly.
- Product page fixes (tied with gift bundles): if the diagnosis shows traffic but few purchases, this is where the effort goes. Let the diagnosis break the tie.
Next steps (for the top option)
-
First step (within a day): Export 90 days of data. Write down sessions, conversion rate (orders ÷ sessions), average order value, the share of orders from repeat customers, and the top traffic sources. Compare with the previous 90 days and the same months last year.
-
Cheapest test: Read the numbers like this.
- Sessions are down or flat, and conversion is normal (often 1–3% for small shops, though that's a rough rule of thumb): traffic is the problem.
- Sessions are healthy but conversion is under about 1%: conversion is the problem.
- Both look fine but few orders come from returning buyers: repeat is the problem.
Then send the past-customer email (option 2) in the same week. It costs $0 and gives a second signal.
-
Depends most on: Assumption #5, that your analytics are accurate enough to read. If they aren't, install basic tracking first and use the customer interviews (option 9) in the meantime.
Rough budget once you know the weak link: keep about $100 in reserve for the gift-bundle packaging or photos, and spend up to $150–200 on ads only if traffic is the problem and your product pages already convert.
Open questions
- Which platform do you sell on? A marketplace like Etsy gives different data and levers than your own site. Your dashboard will show it.
- Were these same months flat or slow last year? Last year's order history will tell you.
- How many customers have bought, and how many bought twice? The order export will tell you.
- Is the holiday gifting window your biggest sales period? Last year's November and December sales will show it.
If you send me the three numbers from the diagnosis (sessions, conversion rate, repeat share), I can go deeper on whichever option they point to.
Real run in an isolated directory with only this skill installed. Tools were limited to Skill and Read, so nothing was searched online and the research phase rests on what the user wrote plus general knowledge, which the output says itself. The made-up shop and numbers are example input. The answer is about 1,200 words, longer than the roughly 900 the skill asks for.
它做什么
分四个阶段带模型处理开放式话题，最后给出建议，而不是一长串点子。
工作方式
- 定框架：写清目标、约束、要做的决定和不在范围内的内容。
- 做调研：把已知、假设、未知分开列，每条都有来源或验证方法。不编造事实，查不到就明说。
- 发散：用至少三种角度（反向思考、去掉一个约束、最小版本、借鉴其他领域等）生成至少八个真正不同的方案，并写明每个方案最主要的失败方式。
- 收敛：按你的目标提炼出评分标准给方案打分，选出前三名，并给出第一名的第一步、最便宜的验证方式，以及它最依赖的假设。
适合场景
- 有几个方向时，决定先试哪个
- 在投入之前探索新想法、产品、项目或计划
- 在混乱的局面里把事实和猜测分开
不适合写代码、规格或 PRD。输出可能偏长，需要的话可以让它缩短。
注意：Skill 本身是英文的。
低风险：纯指令，没有脚本、不联网、不写文件。它自己不会去查资料；除非你的代理能联网搜索或读取你指定的文件，否则调研阶段只基于你提供的信息和通用知识，输出里也会这样说明。它列出的事实和经验数值请当作待核实的线索，不要当成已验证的数据。