台灣綠色和平 Welcome Workflow 成效檢視與決策建議 TW Greenpeace — Welcome Workflow Review & Decisions 2026 上半年（H1）· 給決策者的版本 / First half of 2026 · decision-maker edition

技術重現：每個數字怎麼來的Reproducibility: how every number was obtained

本頁假設讀者沒有任何本機檔案，只要有下列存取權即可從零重算所有數字。所有查詢皆為唯讀（GET／SELECT）。This page assumes the reader has no local files — with the access below, every number can be recomputed from scratch. All queries are read-only (GET / SELECT).

1. 環境與存取1. Environment & access

2. 來源物件與欄位2. Source objects & fields

• HubSpot v4 流程：GET /automation/v4/flows/{id} → actions[]（actionTypeId：0-1 delay、0-4 send email（fields.content_id）、1-241321279 SMS、0-63809083 add-to-list、LIST_BRANCH 分支）。流程 ID：RG 2843532534／SG 4067609839／Reactive 2866280642／Supporter 3600140512。注意：v4 流程不在 /automation/v3/workflows 清單。HubSpot v4 flows: GET /automation/v4/flows/{id} → actions[] (actionTypeId: 0-1 delay, 0-4 send email (fields.content_id), 1-241321279 SMS, 0-63809083 add-to-list, LIST_BRANCH branch). Flow IDs: RG 2843532534 / SG 4067609839 / Reactive 2866280642 / Supporter 3600140512. Note: v4 flows are not in /automation/v3/workflows.
• HubSpot Email 統計：GET /marketing/v3/emails/statistics/list?startTimestamp=&endTimestamp=&emailIds={cid} → aggregate.counters（sent / delivered / open / click / bounce / dropped）。開信率 = open ÷ delivered；點擊率 = click ÷ delivered。Email 內容/CTA：GET /marketing/v3/emails/{cid}?statistics=true。HubSpot email stats: GET /marketing/v3/emails/statistics/list?startTimestamp=&endTimestamp=&emailIds={cid} → aggregate.counters (sent / delivered / open / click / bounce / dropped). Open = open ÷ delivered; click = click ÷ delivered. Content/CTA: GET /marketing/v3/emails/{cid}?statistics=true.
• Salesforce Case：Category__c='TFR'、Sub_Category__c='New Donor Upgrade'、Market__r.Name='Taiwan'、CreatedDate 落在區間；結果欄位 TFR_Call_Outcome__c（'Not Reached' / 'Yes'…）、TFR_Call_Response_Code__c（'Accepted (Upgrade)'…）。Salesforce Case: Category__c='TFR', Sub_Category__c='New Donor Upgrade', Market__r.Name='Taiwan', CreatedDate in window; outcome fields TFR_Call_Outcome__c ('Not Reached' / 'Yes'…), TFR_Call_Response_Code__c ('Accepted (Upgrade)'…).

3. 三步流程（可直接執行）3. 3-step pipeline (copy-runnable)

步驟 1 — 流程 + Email + 統計（自 monorepo 根目錄執行；完整腳本如下，輸入路徑為相對路徑）：Step 1 — flows + emails + stats (run from the monorepo root; full script below, relative paths):

#!/usr/bin/env python3
# Extract the 4 TW HubSpot Welcome v4 flows -> ordered steps + per-email subject/CTA + H1-2026 stats.
import json, urllib.request, urllib.error, os, re, html
TOK = os.environ["HUBSPOT_PAT"]
WIN_START, WIN_END = "2026-01-01T00:00:00Z", "2026-06-30T23:59:59Z"
FLOWS = {
  "RG":{"id":"2843532534"}, "SG":{"id":"4067609839"},
  "Reactive":{"id":"2866280642"}, "Supporter":{"id":"3600140512"},
}
ACTION = {"0-1":"DELAY","0-4":"SEND_EMAIL","0-5":"SET_PROPERTY","0-31":"SET_MARKETABLE",
  "LIST_BRANCH":"IF_THEN_BRANCH","1-218849820":"DATE_FORMAT","1-241321279":"SEND_SMS","0-63809083":"ADD_TO_LIST"}
def get(url):
    req = urllib.request.Request(url, headers={"Authorization":"Bearer "+TOK})
    try: return json.load(urllib.request.urlopen(req)), 200
    except urllib.error.HTTPError as e: return e.read().decode()[:400], e.code
STAT = ["sent","delivered","open","click","bounce","unsubscribed","dropped","spamreport"]
def email_stats(cid):
    r,code = get("https://api.hubapi.com/marketing/v3/emails/statistics/list"
                 "?startTimestamp=%s&endTimestamp=%s&emailIds=%s"%(WIN_START,WIN_END,cid))
    if code!=200: return {"error":code}
    c = (r.get("aggregate",{}) or {}).get("counters",{}) or {}
    d = {k:c.get(k,0) for k in STAT}; deliv = d["delivered"] or 0
    d["open_rate"]  = round(d["open"]/deliv,4) if deliv else None   # open  / delivered
    d["click_rate"] = round(d["click"]/deliv,4) if deliv else None  # click / delivered
    return d
out = {"flows":{}, "emails":{}}; need=set()
for label,m in FLOWS.items():
    d,_ = get("https://api.hubapi.com/automation/v4/flows/%s"%m["id"]); steps=[]
    for a in d.get("actions",[]):
        t = a.get("actionTypeId") or a.get("type"); f = a.get("fields",{}) or {}
        e = {"type":ACTION.get(t,t)}
        if e["type"]=="SEND_EMAIL" and f.get("content_id"):
            e["content_id"]=f["content_id"]; need.add(str(f["content_id"]))
        elif e["type"]=="DELAY": e["delay"]="%s %s"%(f.get("delta"),f.get("time_unit"))
        steps.append(e)
    out["flows"][label]={"total_actions":len(steps),"steps":steps}
for cid in sorted(need):
    em,code = get("https://api.hubapi.com/marketing/v3/emails/%s?statistics=true"%cid)
    out["emails"][cid] = {"error":code} if code!=200 else {
      "subject":em.get("subject"), "stats_h1_2026":email_stats(cid)}
print(json.dumps(out, ensure_ascii=False, indent=2))

步驟 2 — GA4 sg2rg 捐款 CTA（analytics-mcp run_report）：property 251984551、日期 2026-01-01→2026-06-17（勿用未來日期，totalRevenue 會因幣別換算 400 錯誤）、維度 sessionManualAdContent、指標 sessions+keyEvents、過濾 sessionManualCampaignName = sg2rg。Step 2 — GA4 sg2rg donation CTA (analytics-mcp run_report): property 251984551, dates 2026-01-01→2026-06-17 (no future dates — totalRevenue 400s on currency conversion), dimension sessionManualAdContent, metrics sessions+keyEvents, filter sessionManualCampaignName = sg2rg.

步驟 3 — Salesforce TFR 電訪（唯讀 SOQL）：Step 3 — Salesforce TFR calling (read-only SOQL):

sf data query --target-org 00D2v000001f7liEAA --json --query "
  SELECT TFR_Call_Outcome__c, TFR_Call_Response_Code__c, COUNT(Id)
  FROM Case
  WHERE Category__c='TFR' AND Sub_Category__c='New Donor Upgrade'
    AND Market__r.Name='Taiwan'
    AND CreatedDate >= 2026-01-01T00:00:00Z AND CreatedDate <= 2026-06-30T23:59:59Z
  GROUP BY TFR_Call_Outcome__c, TFR_Call_Response_Code__c"
# Totals: 670 cases; upgrades = (Outcome='Yes') = (Response='Accepted (Upgrade)') = 43;
# Not Reached = 333 (~50%); no outcome / never dialed = 139 (~20%); connected = 198.

4. 逐數字溯源4. Number-by-number lineage

5. 重現時的陷阱5. Pitfalls when recomputing

• v4 流程不在 v3 workflow 清單；要用 /automation/v4/flows（分頁 paging.next.after）。v4 flows are not in the v3 workflow list; use /automation/v4/flows (paginate paging.next.after).
• statistics/list 不帶 emailIds 時只回全域彙總；必須單獨帶一個 emailIds 才能拿到該封的 counters。Without emailIds, statistics/list returns only the global aggregate; pass a single emailIds to get one email's counters.
• 開信/點擊率分母是 delivered 不是 sent；bounce/dropped 率分母才是 sent。Open/click denominators are delivered, not sent; bounce/dropped rates use sent.
• 「約 70% 未接觸」包含「139 尚未撥打」；單講「打不通」(Not Reached) 只有約 50%——兩者要分開。已接通分母 = 670−333−139 = 198。"~70% not contacted" includes the 139 never-dialed; "couldn't connect" (Not Reached) alone is only ~50% — keep them separate. Connected base = 670−333−139 = 198.
• Email 預覽連結用 /email/144636797/details/{id}；/preview/{id}?preview_key= 形式會 404。Email preview link is /email/144636797/details/{id}; the /preview/{id}?preview_key= form 404s.
• GA4 totalRevenue 遇未來日期會 400（幣別換算）→ end date 設到 6/17。GA4 totalRevenue 400s on future dates (currency) → cap end date at 6/17.
• 進件 proxy（首封 sent）會持續累積；不同日重抓數字會略增（snapshot 漂移）。The intake proxy (first-email sent) keeps accruing; re-pulling on a later day gives slightly higher numbers (snapshot drift).
• RG 與 Reactive 共用 6 封信 → 這些信的 sent 是兩條流程合計，不能當單一流程量。RG and Reactive share 6 emails → those emails' sent counts are combined across both flows, not one flow's volume.

原始資料檔（若可取得本機）：HubSpot 流程＋Email＋統計 _exports/20260617-1054-tw_welcome_flows_hubspot.json · Email 預覽連結 _exports/20260617-email_preview_links.json · 擷取腳本 _exports/extract_welcome_flows.py · 詳細版 _reports/20260617-1109-tw_welcome_workflows_review.md。Raw files (if available locally): HubSpot flows+emails+stats _exports/20260617-1054-tw_welcome_flows_hubspot.json · preview links _exports/20260617-email_preview_links.json · extract script _exports/extract_welcome_flows.py · detailed report _reports/20260617-1109-tw_welcome_workflows_review.md.