WorkCord 使用狀況分析WorkCord adoption analysis

20 位使用者的聚合分析 · 資料截至 2026-08-31Aggregate analysis of 20 users · data through 2026-08-31

不是使用者玩太複雜，是基本使用就不穩Not users overreaching — basic use is what is broken

20 位使用者累計撞到 160 個問題，近九成發生在「送一則訊息、傳一個檔案、錄一場會、設一條排程」這種第一天就會做的事情上。真正需要自己組裝的進階用法，只佔一成。Across 20 users, 160 problems were catalogued. Nearly nine in ten occur in things anyone does on day one — send a message, upload a file, record a meeting, set one schedule. Advanced work accounts for barely a tenth.

88%的問題出在基本使用
（160 個裡的 141 個）of problems are in basic use
(141 of 160)

20／20每一位都在基本使用上撞到
沒有人只撞到進階問題every user hit a basic-use failure
none hit only advanced ones

12%才是進階用法
agent 互叫、自建技能、外接金鑰is advanced work
agents calling each other, skills, API keys

基本使用Basic use141 · 88%

進階用法Advanced19 · 12%

已停用的 10 位，基本使用佔 90%For the 10 who stopped, 90% basic

最常見的錯誤是假的：答案早就到了，半小時後才補一句「失敗」The most common error is false: the answer arrived, then a failure notice followed 30 minutes later全平台 807 筆逾時錯誤重算：0 筆在 1 秒內跳出，中位數 31 分鐘（等於逾時期限本身）。真正的缺陷是——87% 的逾時錯誤，在它被寫進房間之前，同一個 agent 早就把答案給了。使用者讀完答案，半小時後才看到一句「失敗」。Re-measured across all 807 timeout rows: none fired within a second; the median is 31 minutes — the deadline itself. The real defect is that for 87% of them the same agent had already delivered an answer before the error was written. The user reads the answer, then half an hour later is told it failed.

使用者要做的事都很普通What users were doing was ordinary把會議變成排播順序、把公開資料做成圖卡、把素材庫貼標籤、每天早上收一份摘要、把影片交出去拿回完成片。全都是「交代一件事、等一個結果」。真正需要組裝的進階用法只佔 12%。Turn a meeting into a running order, turn public data into on-air cards, tag an asset library, receive one briefing each morning, hand off a video and get back a finished cut. All of it is "ask for one thing, wait for one result". Genuinely advanced work is only 12%.

修好了，但使用者不知道It was fixed — users were never told平台修好 18 個，只有 11 個有人回頭告知。20 位裡有 15 位一次都沒被通知過——對他們來說，修好等於沒修。另有 5 個是使用者自己寫規則、自己糾正 agent 才解決的。The platform fixed 18; only 11 were communicated back. 15 of the 20 were never told anything — so for them, fixed and broken look identical. A further 5 were solved by the users themselves.

問題型錄：使用者當下在做什麼Problem catalogue: what the user was doing

問題型態	20 人中幾人撞到	已修	未修	未確認
Problem type	Users hit (of 20)	Fixed	Not fixed	Undetermined
送出訊息後沒有回覆，或收到假的失敗通知	20 人	3	35	8
No reply after sending, or a false failure notice	20	3	35	8
傳檔案給 agent，讀不到或檔案遺失	13 人	10	2	11
Files handed to an agent are unreadable or lost	13	10	2	11
agent 產出的內容錯誤或憑空編造	10 人	0	4	12
Agent output is wrong or invented	10	0	4	12
排程沒準時、沒送到、或停不掉	9 人	1	8	8
Schedules fire late, never arrive, or cannot be stopped	9	1	8	8
錄音、逐字稿或會議報告不能用	8 人	1	6	6
Recording, transcript or meeting report unusable	8	1	6	6
其他	8 人	1	4	7
Other	8	1	4	7
系統內部訊息／錯誤原文外洩給使用者看	7 人	0	1	6
Internal system text leaks into the room	7	0	1	6
做出來的東西使用者拿不到	6 人	1	3	2
Work is produced but the user cannot retrieve it	6	1	3	2
agent 之間無法協作（互叫、派工、跨房）	6 人	0	4	2
Agents cannot work together (calling, dispatch, cross-room)	6	0	4	2
App 介面本身的問題	3 人	1	0	4
Problems in the app interface itself	3	1	0	4

每一列代表一種機制，不是一個人。同一個人可能出現在多列。Each row is a mechanism, not a person; one person can appear in several rows.

兩組使用者的對照The two cohorts

代號	單位	訊息量級	活躍天數	問題	平台修好	有被告知
Code	Unit	Messages	Active days	Problems	Fixed	Told
A1	TVBS 內部單位	300+	26 天以上	11	2	6
A1	TVBS internal	300+	26+ days	11	2	6
A2	TVBS 內部單位	300+	15–25 天	12	1	0
A2	TVBS internal	300+	15–25 days	12	1	0
A3	TVBS 內部單位	300+	26 天以上	16	1	0
A3	TVBS internal	300+	26+ days	16	1	0
A4	TVBS 內部單位	300+	26 天以上	10	1	0
A4	TVBS internal	300+	26+ days	10	1	0
A5	TVBS 內部單位	300+	15–25 天	8	0	0
A5	TVBS internal	300+	15–25 days	8	0	0
A6	集團關係企業 IT	150–300	15–25 天	9	2	0
A6	Group affiliate — IT	150–300	15–25 days	9	2	0
A7	集團關係企業 IT	150–300	6–14 天	6	1	0
A7	Group affiliate — IT	150–300	6–14 days	6	1	0
A8	TVBS 內部單位	150–300	6–14 天	8	1	0
A8	TVBS internal	150–300	6–14 days	8	1	0
A9	TVBS 內部單位	150–300	6–14 天	7	0	1
A9	TVBS internal	150–300	6–14 days	7	0	1
A10	TVBS 內部單位	150–300	15–25 天	10	1	2
A10	TVBS internal	150–300	15–25 days	10	1	2
C1	TVBS 內部單位	150–300	15–25 天	9	1	0
C1	TVBS internal	150–300	15–25 days	9	1	0
C2	TVBS 內部單位	150–300	6–14 天	7	1	0
C2	TVBS internal	150–300	6–14 days	7	1	0
C3	TVBS 內部單位	50–150	6–14 天	7	1	1
C3	TVBS internal	50–150	6–14 days	7	1	1
C4	TVBS 內部單位	50–150	1–5 天	7	1	0
C4	TVBS internal	50–150	1–5 days	7	1	0
C5	TVBS 內部單位	< 50	1–5 天	7	1	0
C5	TVBS internal	under 50	1–5 days	7	1	0
C6	TVBS 內部單位	< 50	1–5 天	8	0	0
C6	TVBS internal	under 50	1–5 days	8	0	0
C7	集團關係企業 IT	< 50	1–5 天	5	2	1
C7	Group affiliate — IT	under 50	1–5 days	5	2	1
C8	TVBS 內部單位	< 50	1–5 天	5	1	0
C8	TVBS internal	under 50	1–5 days	5	1	0
C9	TVBS 內部單位	< 50	1–5 天	4	0	0
C9	TVBS internal	under 50	1–5 days	4	0	0
C10	TVBS 內部單位	< 50	1–5 天	4	0	0
C10	TVBS internal	under 50	1–5 days	4	0	0

A 開頭＝仍在使用，C 開頭＝曾經活躍但已停用。活躍組活躍日中位數 19–21 天、已停用組 5 天；已停用的 10 位裡有 7 位一輩子只用了 5 天以內。Codes starting A are still active; C are once-heavy users who stopped. Median active days: 19–21 for the active cohort, 5 for the churned; 7 of the 10 churned users were active on five days or fewer in total.

方法與界線Method and limits

項目	說明
Item	Detail
資料來源	房間資料庫與執行紀錄（唯讀），加上 8/29 與 8/31 在專用測試房的實地測試。
Data sources	Room database and execution records (read-only), plus live tests in a dedicated test room on 8/29 and 8/31.
「已修復」的判準	分兩級：實測＝跑同一條路徑通過（18 個裡有 14 個）；觀察＝沒重跑，但生產環境同一路徑連續多天成功。
What counts as fixed	Two grades: live-tested — the same path was re-run and passed (14 of the 18); observed — not re-run, but the same path succeeded in production for several consecutive days.
「未修」的判準	8/26 之後仍觀測到，或當日實測仍失敗。只有舊紀錄一律歸「未確認」。
What counts as not fixed	Observed on or after 8/26, or failed a live test that day. Anything with only older records is filed as undetermined.
一次重大更正	本分析原本的頭條數字讀錯了資料欄位（把「使用者送出時間」當成「錯誤寫入時間」），經平台工程團隊複核後，160 列全部重新查證並更正。
One major correction	The original headline figure in this analysis read the wrong database field (treating the user's send time as the error's write time). After a line-by-line review by the platform engineering team, all 160 rows were re-verified and corrected.
已知限制	已停用的使用者在修復上線前就停止產生紀錄，因此他們的部分問題無法從紀錄判定，只能靠實地重現測試。
Known limit	Users who stopped did so before some fixes shipped, so parts of their experience cannot be judged from records alone and rely on live reproduction instead.

統計截至 2026-08-31。Data through 2026-08-31.