Sep 3, 2026 · Building products
The do-things-that-don't-scale playbook不能規模化的事，怎麼做才對
A short playbook for doing the unscalable work that actually moves early products: from Paul Graham’s advice to Superhuman’s PMF engine and The Mom Test.新產品一開始就是一連串「不能規模化的事」。從 Paul Graham、Superhuman 到 The Mom Test，整理一份好好去做的實用清單。
Every early product is held together by unscalable work: hand written emails, custom setups, one by one support. The point is not to suffer. It is to learn fast enough that you earn the right to automate later.
What Paul Graham really meant by “do things that do not scale”
In Paul Graham’s essay, doing unscalable work is how you pull product market fit into existence. You recruit users manually, onboard them one by one, and stay close enough to see their real struggles instead of dashboard averages. The unscalable part is the microscope that lets you see problems at the resolution where they can still be fixed.
Superhuman: unscalable service, scalable signal
Superhuman did not start by chasing a huge funnel. They personally onboarded people, watched them use the product, and then asked how they would feel if they lost it tomorrow to quantify real attachment. Behind the white glove service they built a repeatable engine, segmenting users and iterating the product until the percentage who said “very disappointed” crossed a clear threshold.
The Mom Test: unscalable conversations, non fake learning
The Mom Test reminds you that not every user conversation produces truth. If you pitch your idea and ask if people like it, you get flattery instead of data, even from strangers. Unscalable conversations only matter when you ask about their real past behavior and let them complain about their own life instead of reacting to your mockup.
A short prep list before you go unscalable
- A narrow definition of “who this is for now” and the first twenty names you will contact
- A concrete promise for each new user, for example personal setup or weekly check in calls
- One place to log every manual action and the learning you got from it
- A single metric that your unscalable work should move, such as activation or “very disappointed”
- A stop rule for each activity, for example when to automate, delegate, or delete it
The takeaway
- Unscalable work is a microscope for product market fit, so stay close enough to users to see every crack
- Wrap white glove service around a simple metric, and you can turn artisanal work into a repeatable engine
- Conversations only count when they reveal past behavior and real pain, not opinions about your idea
所有早期產品本來就靠一堆不能規模化的事撐起來：手寫信、客製開通、一個一個盯著用。重點不是受苦，而是用最快的速度學到足夠多，之後才有資格把東西自動化。
Paul Graham 口中的「不能規模化」，其實在說什麼
在 Paul Graham 的文章裡，不能規模化的事，是你把產品市場契合拉出來的方式。你親自找使用者、一個一個帶上手，靠近到可以看見每個卡關點，而不是只看平均數的報表。這些吃力的細活，就是那支放大鏡，讓問題還小的時候就有機會被修好。
Superhuman：人力服務換來可規模化的訊號
Superhuman 一開始沒有追求大漏斗，而是用真人一對一帶新手上手，現場看他們怎麼用、卡在哪裡。結束後只問一題「如果之後不能再用 Superhuman，會多失望」，把真實黏著度變成可以追蹤的數字。這層白手套式服務下面，其實是一套可重複的引擎：分群、改產品、一直拉高「非常失望」的比例。
The Mom Test：不能規模化的對話，才要換真學習
The Mom Test 提醒你，並不是每一場使用者訪談都能得到真話。如果你一邊推銷、一邊問「這樣好不好」，對方多半只會客氣配合。既然都要花時間面對面，不如只問他過去怎麼做、怎麼被卡住，讓他抱怨自己的生活，而不是對著你的稿圖給意見，這樣才能換到真的學習。
下去做這些「不能規模化的事」前，先準備好這些
- 一句現在版本「只為了誰而存在」的描述，加上你要親自聯絡的前二十個名字
- 給每位新使用者一個具體承諾，例如親自設定或每週一次電話回顧
- 一個集中地方，紀錄自己做過的每個手工動作和從中學到什麼
- 一個你希望被拉動的指標，例如啟用率或回答「非常失望」的比例
- 每項手工工作的停損規則：什麼時候自動化、什麼時候交給別人、什麼時候直接砍掉
總結
- 不能規模化的事是一支看產品市場契合的放大鏡，要靠得夠近，才看得到裂縫
- 把白手套服務綁在一個簡單指標上，手工活才能慢慢被變成一套引擎
- 對話只有在說出過去行為和真實痛點時才算數，對你點子的意見頂多是噪音