多數轉型的失敗，是敘事的失敗Most transformations fail as narrative problems

過去一年，我近距離看了很多 AI 導入專案。技術幾乎從來不是死因：工具會愈來愈好，供應商會愈來愈多，預算遲早會到位。死因幾乎都一樣——組織說不清楚自己為什麼要變、要變成什麼樣、誰會因此不一樣。三位主管，三種說法；第一線聽到的版本，又是第四種。 Over the past year I have watched many AI adoption programs up close. Technology is almost never the cause of death: the tools keep improving, vendors keep multiplying, budgets eventually arrive. The cause of death is almost always the same — the organization cannot say why it is changing, into what, and who will be different because of it. Three managers, three versions; what the front line hears is a fourth.

這不是溝通不良，這是敘事沒有被當成基礎設施。訊息架構、術語體系、故事線——這些東西和機房與產線一樣，需要被設計、被維護、被驗收。沒有人會讓工廠「自然長出」動線，但多數組織讓轉型敘事自然長出，然後在第八個月發現大家走散了。 This is not a communication problem; it is narrative not being treated as infrastructure. Message architecture, terminology systems, storylines — these need to be designed, maintained and accepted, exactly like server rooms and production lines. Nobody lets a factory "grow" its own workflow, yet most organizations let their transformation narrative grow wild — and discover in month eight that everyone has wandered apart.

敘事不是包裝，是承重結構。 Narrative is not decoration. It is load-bearing structure.

三個起點Three starting points

如果你的組織正要導入 AI——或任何需要人改變行為的東西——我會建議先做三件小事。第一，一句共同的「為什麼」：讓三位主管在同一句話上簽名，貼在所有簡報的第一頁。第二，一條基線數據：挑一件事，量它現在要花多久；沒有基線，一年後你將無法回答「有沒有用」。第三，一條先跑起來的流程：不是試點簡報，是真的有人每週在用的流程——曲線會從那裡開始長。 If your organization is about to adopt AI — or anything that requires people to change behavior — I would suggest three small things first. One: a single shared "why," a sentence three managers will sign, printed on page one of every deck. Two: one baseline metric — pick one task and measure how long it takes today; without a baseline, a year from now you cannot answer "did it work." Three: one workflow that actually runs — not a pilot deck, but something someone uses every week. The curve grows from there.

這封信每月一封，寫 AI 導入的現場、敘事與數據的交界，以及值得留下的判斷。下一期見。 This letter arrives once a month — on AI adoption in the field, the border between narrative and data, and judgments worth keeping. See you next issue.

林政威 敬上 — Rockie Lin