用 AI 由邊度開始?揀啱第一條流程嘅五條問題
每個月都有公司嚟問我哋同一條問題:「我哋應該用邊個 AI 工具?」我哋嘅答案永遠係同一個:先揀流程,後揀工具。工具三個月換一代,你揀咗都未必係下個月最好嗰個;但你條流程係你公司真實嘅運作,揀啱咗,任何工具都係喺上面加分。
問題係:一條公司入面有幾十條流程,點知邊條先值得郁?以下五條問題,我哋見客嗰陣會同對方一齊行一次。全部答「係」嘅流程,就係你嘅第一單 AI 項目。
1. 係咪規則清楚、重複發生?
AI 最擅長嘅唔係「諗嘢」,係「照規矩做嘢做得好快」。文件分類、表格資料抽寫、標準回覆、格式轉換 —— 呢啲嘢有明確輸入輸出、有判斷準則,AI 好快學得識。相反,一單 case 一個玩法、連你自己都講唔清「點解咁決定」嘅流程,AI 只會學到你嘅混亂。
2. 而家係咪人手做、做得好悶?
問吓做嗰個人。如果佢答「好悶」「好浪費」,恭喜你 —— 你搵到一條高價值流程。員工覺得悶嘅嘢,通常即係重複、冇判斷、但又要日日做嘅嘢。呢類工作每做一次就燒一次人工,而且永遠做唔快。
3. 做錯一次嘅成本有幾高?
如果你話「錯咗都係重做一次,十分鐘搞掂」,咁呢條唔係好選擇 —— AI 幫你慳嘅時間,隨時唔夠 cover 你管住佢嘅時間。要揀嗰啲「錯一次就麻煩」嘅:漏咗一份文件、覆錯一個客、入錯一個數。AI 唔會攰,先至係呢度嘅價值。
4. 有冇量得到?
「每日大約 200 條查詢」「每個月 1,200 張單」—— 有數字,先有 before/after。我哋會喺項目開始前同你量一次 baseline:做一次要幾耐、幾錢、錯幾多。冇 baseline 嘅自動化項目,做完都唔知成唔成功。
5. 出事嗰陣,有冇人識收手?
AI 輸出要有人最後把關。唔係因為 AI 信唔過,而係任何自動化都會遇到「從未見過嘅 case」—— 嗰陣需要一個人話「呢單我嚟」。流程入面如果本身已經有「核對」呢一步,恭喜,你已經有條現成嘅人機分工線。
五條之後:唔好由零開始砌系統
揀啱流程之後,我哋通常建議由最窄嗰一段做起 —— 唔好一開始就諗「成個部門自動化」,揀一條真實嘅流程、一個真實嘅樽頸,兩三個禮拜行得通,先至擴。行得通嗰陣,你自然會知道下一步要郁邊度。
如果你喺度諗緊公司邊條流程值得郁,同我哋傾兩句 —— 我哋會直接話你知嗰條值唔值得做,唔值得嘅話,你慳返個 consultation 嘅時間。
Where to start with AI: five questions to pick your first workflow
Every month, companies ask us the same question: "Which AI tool should we use?" Our answer is always the same: pick the workflow first, the tool second. Tools get replaced every three months — whatever you choose today may not be the best one next month. But your workflow is how your company actually runs. Pick the right one and every tool that comes along simply adds to it.
The catch: a company has dozens of workflows. How do you know which one deserves your first AI project? These are the five questions we walk through with clients. If one process answers "yes" to all of them, that is your first project.
1. Is it rule-based and repetitive?
AI is not great at "thinking" — it is great at doing rule-based work very fast. Document sorting, extracting data from forms, standard replies, format conversion: clear inputs, clear outputs, clear criteria. AI learns these quickly. A workflow where every case is different — where you cannot even explain why you decided what you decided — will just teach the AI your chaos.
2. Is a human doing it and finding it dull?
Ask the person who does it. If they say "boring" or "such a waste", congratulations — you have found a high-value workflow. Work people find dull is usually repetitive, judgement-free and daily. Every run burns payroll and it never gets faster.
3. How expensive is a mistake?
If the answer is "we just redo it, ten minutes", this is not the one — the time AI saves may not cover the time you spend supervising it. Pick the workflows where one slip causes real trouble: a document missed, a customer mis-replied, a figure entered wrong. AI never gets tired; that is where the value lives.
4. Can you measure it?
"About 200 enquiries a day", "1,200 invoices a month" — numbers give you a before and after. Before any project we measure a baseline with you: how long it takes, what it costs, how often it goes wrong. An automation project without a baseline can never be declared a success.
5. Is there a human who can catch it when it fails?
AI output needs a final human check. Not because AI is untrustworthy — because every automation eventually meets a case it has never seen, and at that moment someone needs to say "I will take this one." If your process already has a review step, you already have the human-machine boundary line built in.
After the five questions: do not build a system from zero
Once you have your workflow, we usually suggest starting with the narrowest slice — not "automate the whole department". Pick one real process and one real bottleneck, prove it works in two or three weeks, then expand. When it works, you will naturally see where to move next.
If you are wondering which of your workflows is worth touching, talk to us — we will tell you plainly whether it is worth doing. If it is not, you just saved yourself a consultation.