growth-pulse
拉各通路、漏斗、活動和評論資料，回答成長引擎有沒有在轉並給三個本週行動
中文翻譯
--- name: growth-pulse description: > 替中小企業老闆回答一個問題：成長引擎到底有沒有在運轉？ 從任何已連接的來源，拉出各通路的銷售趨勢、漏斗轉換、行銷活動回報、 表現最好和最差的產品，以及客戶情緒， 最後以本週值得採取的三個成長行動作結。這是 business-pulse 在成長面的 孿生兄弟（business-pulse 負責財務面）。只要老闆問行銷或業務表現如何、 某個活動值不值得、潛在客戶從哪裡來、為什麼 pipeline 停滯、哪些產品在動， 或說了類似「is my marketing working」「growth check」 「how's the funnel」或「where should I put my money next month」的話，就用它。 即使老闆只點名單一通路，而不是整體成長，也要用。 allowed-tools: Read, WebFetch --- # Growth Pulse 一頁回答成長引擎有沒有在運轉，以及這週該怎麼做。 老闆們說，他們得從三個儀表板拉數字，然後自己想辦法找出背後的故事。在這裡，故事才是交付物，不是儀表板。 ## 步驟 1：平行拉取所有資料 把所有 connector 呼叫放在同一批送出。指標對應到工具的方式，請看 `reference/data_sources.md`。一個一個依序拉，會變成沒人願意等第二次的等待。 - **HubSpot**：依階段的 deal、新聯絡人、lead 來源、得標率、週期時間、停滯的 deal - **PayPal、Stripe、Square**：營收趨勢、交易筆數、退款 - **Shopify**：訂單、SKU 層級的營收、回購率 - **QuickBooks**：已實現的營收，帳本認可的那個數字 - **TikTok Ads**：從原生 connector 取得上週的花費、曝光、點擊、成果，以及每個活動的單次成果成本（`report_integrated_get`，最近 7 天對比前 7 天），讓付費廣告花費和它宣稱帶動的銷售並排呈現 - **Mailchimp**：活動、電商和受眾成長的分析，附活動的營收歸因 - **Gmail 或 Microsoft 365**：進來的詢問量和回覆時間 - **公開評論**：透過網路研究取得近期的評分和評論文字 如果某個來源出錯或缺少，就記錄下來然後繼續。這個 skill 的設計是：只有一個 connector 時，也能說出有用的東西；有六個時，能說出很出色的東西。 **零個 connector 也是支援的路徑。** 請對方提供匯出檔，例如 CRM 的 deal 匯出、銷售報表、廣告成效 CSV，並用檔案做出同樣的 pulse。直接說明，不要停住。 ## 步驟 2：計算五個檢視 門檻值請讀 `reference/thresholds.md`。計算每個檢視並指派一個狀態。 1. **通路趨勢**：這一期各通路的營收，對比上一期，以及有資料的話對比去年同期 2. **漏斗轉換**：進來的 lead、合格的、opportunity、成交；每個步驟之間的轉換率，以及最大的流失點在哪 3. **行銷活動回報**：每個活動的花費對比歸因營收，以及單一 lead 成本 4. **產品表現**：依營收和依成長率排序的前三名和後三名 5. **客戶情緒**：近期的評論分數、評論主題、爭議、流失訊號 沒有資料的檢視要標成「n/a」，並在附錄中點名。不要讓它默默消失。 ## 步驟 3：找出真正的故事 這是這個 skill 發揮價值的地方。五個檢視本身仍然只是儀表板。工作是把它們串起來。 最重要的幾種連結： - **花費上升、單一 lead 成本也上升**，代表這個通路正在飽和，不是在擴張 - **lead 量很強、得標率卻下降**，指向的是 lead 品質或回應速度，不是行銷 - **一個產品撐起整個趨勢**，代表成長比總數看起來的更脆弱 - **營收撐住、情緒卻下降**，是兩個月後會反映在營收上的早期警訊 - **漏斗急遽下降的某個階段**，通常比任何新活動都更有價值 找出能解釋這一期的那一句話，並把它放在最前面。 ## 步驟 4：寫出三個行動 每份 pulse 都以剛好三個行動作結。不是五個，也不是列出所有注意到的事。 每個行動需要三件事，缺一個就不算行動： - **做什麼**：具體到今天就能開始 - **為什麼**：這份 pulse 裡能證明它的那個數字 - **值多少**：機會或風險的大致規模 依影響大小排序，不是依容易程度。老闆自己可以判斷什麼容易；只有這份 pulse 能告訴他們什麼是大的。 如果誠實的答案是什麼都不用改，就這樣說。每週都硬湊出三個行動的 pulse，會訓練老闆去忽略它們。 ## 步驟 5：撰寫 使用 `reference/output_template.md` 裡的確切結構。只放有真實資料的段落。 寫作規則： - 數字領頭，文字跟隨。不要寫「paid social performed well」，要寫「paid social drove 41 leads at USD 18 each, down from USD 31 last month.」 - 每個數字都附帶比較。沒有基準線的數字，是錯失的洞察。 - 指名具體的活動、產品、通路或客戶。「Some campaigns underperformed」無法使用。 - 不要用老闆平常不用的行銷術語。不要寫「top of funnel velocity」，要寫「how many new leads came in and how fast.」 - 如果某個段落沒有值得說的，就寫「No material change」然後繼續。 ## 步驟 6：交付 pulse 頁面 除了聊天裡的 pulse 之外，另外交付完整的 pulse（是額外附上，絕對不是取代），並依老闆儲存的輸出偏好，絕對不要預設用 markdown 檔案。檢查 `## Business context` 區塊的 `Output preference`（共用風格指南的規則，`../../shared/artifact-style.md`）： - **視覺化 artifact（預設）：** 以 house style 把 pulse 渲染成 HTML 頁面。五個檢視各有一個面板；營收趨勢、單一 lead 成本和得標率是統計 tile，並以它們的比較當成脈絡行；活動回報和產品表現是表格，花費和營收欄位用 tabular-nums；每個檢視的狀態變成狀態標籤（good、warn、critical，或以純文字寫 n/a）；三個行動排好名次，放在自己的面板，作為頁面的結尾。 - **docx / md / notion / canva 偏好：** 用那種形式交付相同內容：DOCX 或 markdown 檔案、透過 connector 建立的 Notion 頁面（指定目的地，絕對不覆寫）、或透過 Canva connector 建立的 Canva Doc（每次執行都是新的設計，以日期命名；表格變成清單）；如果沒有連接 Notion 或 Canva，就退回用視覺化 artifact，並說明原因。 - **Best for skill：** 用視覺化 artifact，因為五個檢視並排在同一個畫面上最好比較。 ## 步驟 7：提議儲存，只提一次 呈現之後，只提議一次把 pulse 存成檔案或貼到 Slack。貼到 Slack 是對外寫入，所以只有在明確說「好」時才會發生，這個提議本身就是核准關卡。說好就做。說不或沒回應就繼續往下。不要問兩次。 然後用一行總結 pulse 的發現，再加上最相關的一個下一步，以及附近最多另外兩個： - 如果行動指向內容或推廣：「what should I promote」會執行 `content-strategy`。 - 如果問題出在漏斗：「fill my funnel」會執行 `/grow-pipeline`。 - 想看財務面的孿生 skill：「how's the business doing」會執行 `business-pulse`。 最多三項提議。這個 session 裡老闆婉拒過的提議，絕對不再重複。 ## 範圍變體 老闆常常只想要其中一塊，不是整份 pulse： - **「How did the campaign do」**：只做活動回報，再加情緒檢查 - **「How's the pipeline」**：只做漏斗轉換和停滯的 deal - **「What's selling」**：只做產品表現 - **「Are we getting reviews」**：只做情緒 - **「Where should I spend next month」**：通路趨勢和活動回報，三個行動偏重預算 給他們問的那一塊。不要拿完整的 pulse 硬套在一個狹窄的問題上。 ## 不要做什麼 - **拉資料之前不要先問許可。** 這個 skill 已經被呼叫了。直接執行。 - **不要編造歸因。** 如果活動的營收追蹤不到，就說歸因無法取得，不要用猜的把營收分配過去。假的歸因曾經挪動過真實的預算。 - **不要把 CRM 的預測金額當成營收回報。** Deal 金額是期望；發票是事實。要標明哪個是哪個。 - **不要跨來源重複計算。** 一筆在 Stripe 結算的 Shopify 訂單，算一筆銷售。選一個營收來源並點名它。 - **不要在同一個來源內重複計算。** 依客戶彙總的銷售報表，對同一個客戶會同時回傳一列上層摘要列和另一列單獨的總計列。在你替客戶排名或計算集中度之前，要先過濾掉標為總計的列，否則同一個客戶會在清單裡出現兩次，集中度的數字也會偏高。 - **不要製造急迫感。** 如果成長平穩，pulse 讀起來就該是平穩的。 - **不要從 QuickBooks 的損益摘要讀費用或利潤率。** 要把各列加總（`../../shared/quickbooks-report-traps.md`，Trap 2）。 ## 參考檔案 - `reference/data_sources.md`：connector 對應到指標，附備案 - `reference/thresholds.md`：五個檢視各自的狀態門檻 - `reference/output_template.md`：確切的輸出結構 - `reference/gotchas.md`：會產生「很有自信、卻是錯的」成長建議的失敗模式 ## 使用沒有列出的工具 這個 skill 裡提到的 connector 是測試過的路徑，不是一道牆。如果老闆想讓這個流程用某個沒連接、也沒列出的工具，就提供 `build-connector`：它會先檢查 connector 目錄，沒有的話再透過 Zapier 連接，絕對不要對著原始 API 手工打造。連線建立之後，這個工具就跟其他任何選用的 connector 一樣加入這個 skill，並受同樣的核准關卡約束。
英文原文
--- name: growth-pulse description: > Answers one question for an SMB owner — is the growth engine actually working? Pulls sales trend by channel, funnel conversion, campaign return, best and worst performing products, and customer sentiment from whatever is connected, then ends with the three growth actions worth taking this week. The growth-side twin of business-pulse, which covers the finance side. Use this whenever the owner asks how marketing or sales is performing, whether a campaign paid off, where leads are coming from, why the pipeline is stalling, which products are moving, or says anything like "is my marketing working," "growth check," "how's the funnel," or "where should I put my money next month." Reach for it even when the owner names a single channel rather than growth overall. allowed-tools: Read, WebFetch --- # Growth Pulse One page that answers whether the growth engine is working, and what to do about it this week. Owners describe pulling numbers from three dashboards and trying to find the story themselves. The story is the deliverable here — not the dashboards. ## Step 1 — Pull everything in parallel Dispatch all connector calls in a single batch. See `reference/data_sources.md` for the metric-to-tool mapping. Serial pulls turn this into a wait nobody sits through twice. - **HubSpot** — deals by stage, new contacts, lead source, win rate, cycle time, stalled deals - **PayPal, Stripe, Square** — revenue trend, transaction counts, refunds - **Shopify** — orders, SKU-level revenue, repeat purchase rate - **QuickBooks** — realized revenue, the number the books agree with - **TikTok Ads** — last week's spend, impressions, clicks, results, and cost per result per campaign, from the native connector (`report_integrated_get`, last 7 days against the prior 7), so paid spend sits beside the sales it claims to drive - **Mailchimp** — campaign, e-commerce, and audience-growth analytics, with revenue attribution on campaigns - **Gmail or Microsoft 365** — inbound inquiry volume and response times - **Public reviews** — recent ratings and review text, via web research If a source errors or is missing, record it and move on. This skill is designed to say something useful on one connector and something excellent on six. **Zero connectors is a supported path.** Ask for exports — a CRM deal export, a sales report, an ads performance CSV — and build the same pulse from files. Say it plainly rather than stalling. ## Step 2 — Compute the five views Read `reference/thresholds.md` for the cutoffs. Compute each view and assign it a status. 1. **Channel trend** — revenue by channel this period versus last, and versus the same period last year where available 2. **Funnel conversion** — leads in, qualified, opportunities, won; the conversion rate between each step and where the biggest drop-off sits 3. **Campaign return** — spend versus attributed revenue per campaign, and cost per lead 4. **Product performance** — top three and bottom three by revenue and by growth rate 5. **Customer sentiment** — recent review scores, review themes, disputes, churn signals A view with no data gets marked "n/a" and named in the appendix. It does not silently vanish. ## Step 3 — Find the actual story This is where the skill earns its keep. Five views on their own are still a dashboard. The job is to connect them. The connections that matter most: - **Spend rising while cost per lead rises** means the channel is saturating, not scaling - **Strong lead volume with falling win rate** points at lead quality or response speed, not at marketing - **One product carrying the whole trend** means the growth is more fragile than the total suggests - **Sentiment dropping while revenue holds** is the early warning that shows up in revenue two months later - **A stage where the funnel drops off sharply** is usually worth more than any new campaign Look for the single sentence that explains the period. Lead with it. ## Step 4 — Write the three actions Every pulse ends with exactly three actions. Not five, not a list of everything noticed. Each action needs three things or it isn't an action: - **What to do** — specific enough to start today - **Why** — the number from this pulse that justifies it - **What it's worth** — the rough size of the opportunity or the risk Rank them by size of impact, not by ease. Owners can decide what's easy; only this pulse can tell them what's big. If the honest answer is that nothing needs to change, say that. A pulse that manufactures three actions every week trains the owner to ignore them. ## Step 5 — Compose Use the exact structure in `reference/output_template.md`. Include only sections with real data. Writing rules: - Numbers lead, words follow. Not "paid social performed well" — "paid social drove 41 leads at USD 18 each, down from USD 31 last month." - Every number carries its comparison. A figure with no baseline is a missed insight. - Name the specific campaign, product, channel, or customer. "Some campaigns underperformed" is unusable. - No marketing vocabulary the owner doesn't use. Not "top of funnel velocity" — "how many new leads came in and how fast." - If a section has nothing worth saying, write "No material change" and move on. ## Step 6 — Deliver the pulse page Alongside the chat pulse — additive, never a replacement — deliver the full pulse per the owner's stored output preference — never default to a markdown file. Check the `## Business context` block's `Output preference` (shared style guide rule, `../../shared/artifact-style.md`): - **Visual artifact (the default):** render the pulse as an HTML page in the house style. Each of the five views gets a panel; revenue trend, cost per lead, and win rate are stat tiles with their comparison as the context line; campaign return and product performance are tables with tabular-nums spend and revenue columns; each view's status becomes a status pill (good, warn, critical, or n/a in plain text); the three actions close the page in their own panel, ranked. - **docx / md / notion / canva preference:** deliver the same content in that form — a DOCX or markdown file, a Notion page created via the connector (named destination, never overwriting), or a Canva Doc created via the Canva connector (a new design each run, named with the date; tables become lists); fall back to the visual artifact if Notion or Canva is not connected — and say that is why. - **Best for skill:** use the visual artifact — five views compare best side by side on one screen. ## Step 7 — Offer to save, once After presenting, offer once to save the pulse as a file or post it to Slack. Posting to Slack is an outward write, so it only happens on that explicit yes — the offer is the approval gate. If yes, do it. If no or no response, move on. Do not ask twice. Then close with what the pulse found in one line, plus the single most relevant next step and at most two others nearby: - If the actions point at content or promotion: "what should I promote" runs `content-strategy`. - If the funnel is the problem: "fill my funnel" runs `/grow-pipeline`. - For the finance-side twin: "how's the business doing" runs `business-pulse`. Max three offers. Never repeat an offer the owner declined this session. ## Scope variants Owners often want a slice rather than the whole pulse: - **"How did the campaign do"** — campaign return only, plus the sentiment check - **"How's the pipeline"** — funnel conversion and stalled deals only - **"What's selling"** — product performance only - **"Are we getting reviews"** — sentiment only - **"Where should I spend next month"** — channel trend and campaign return, with the three actions weighted toward budget Give them the slice they asked for. Don't force the full pulse on a narrow question. ## What not to do - **Do not ask permission before pulling data.** The skill was invoked. Run it. - **Do not invent attribution.** If a campaign's revenue can't be traced, say the attribution is unavailable rather than assigning revenue on a guess. Fake attribution has moved real budgets. - **Do not report CRM forecast value as revenue.** Deal values are hopes; invoices are facts. Label which is which. - **Do not double-count across sources.** A Shopify order settled in Stripe is one sale. Pick one revenue source and name it. - **Do not double-count inside one source.** The sales-by-customer summary returns both a parent summary row and a separate total row for the same customer. Filter rows marked as totals before you rank customers or compute concentration, or the same customer lands in the list twice and the concentration figure comes out too high. - **Do not manufacture urgency.** If growth is steady, the pulse should read as steady. - **Do not read expenses or margin from the QuickBooks P&L summary.** Total the rows (`../../shared/quickbooks-report-traps.md`, Trap 2). ## Reference files - `reference/data_sources.md` — connector to metric mapping, with fallbacks - `reference/thresholds.md` — status cutoffs for each of the five views - `reference/output_template.md` — exact output structure - `reference/gotchas.md` — the failure modes that produce confidently wrong growth advice ## Using a tool that isn't listed The connectors named in this skill are the tested paths, not a wall. If the owner wants this flow to use a tool that isn't connected or listed, offer `build-connector` — it checks the connector directory first and connects through Zapier otherwise, never hand-building against a raw API. Once the connection exists, the tool joins this skill like any other optional connector, under the same approval gates.