做产品 PMaker
空格的键盘
他说的
能不能加个导出 Excel 的按钮
表层诉求
≠
他真正要的
每周一要向老板汇报
现在得手动抄二十行数字
抄错过一次,被说了
真实动机
≠
你该给的
一份每周一自动生成的
汇报视图,可以直接截图
解决方案

三样东西经常被混成一句「用户要导出功能」。混了之后,你做的是按钮,他要的是不再挨骂。

What Is a Need

A need is the state a user wants to reach in a given situation. It isn't the sentence the user says, nor the feature you plan to build.

What you'll run into
  • 照着用户说的做完了,他还是不满意
  • 功能一个个加上去,产品越来越重,问题还在
  • 两个用户提的需求互相打架,不知道该听谁的

What the user says is the solution they can think of, not their problem. They're used to Excel, so they ask for export; they've seen this button in another app, so they want it too. Treat that sentence as the need, and what you build will forever trail behind someone else's imagination.

同一件事的三种说法,只有中间那句是需求 用户说的话 「能不能加一个导出 Excel 的按钮」 他能想到的解决办法 他想达成的状态 这才是需求 每月对账那天,把散在各处的改稿记录 变成一份能直接发给客户的清单 你打算做的功能 一键生成对账单,也可以导出 Excel 你能想到的解决办法
Pay special attention to the last group. Say "build a search" and the AI will definitely do the functional part; "return results within one second across three thousand records"—that kind of non-functional requirement, you have to say, or it won't occur to it.

Six traits of a need

These six traits explain why needs are so hard to handle.

特征对你意味着什么
源于现实需求受技术和环境限制。通讯的需求没变过,但从书信变成了即时消息
因人而异同一个功能,两个人要的东西可能完全不同。所以要问具体的人
不断发展满足了会长出新的。今天的惊喜是明天的标配
相互关联一个需求满足不了时,人会用别的方式补偿。这是替代方案的来源
有层次低层的没满足,高层的就不成立。刚失业的人不会先考虑自我实现
会冲突一批人要功能多,一批人要打开快。冲突不可能全解决,只能选边
Pay special attention to the last group. Say "build a search" and the AI will definitely do the functional part; "return results within one second across three thousand records"—that kind of non-functional requirement, you have to say, or it won't occur to it.

Which categories actually matter

There are a dozen-plus ways to classify needs. The three that actually change your decision day to day are these.

按有没有说出口
显性 已经明确提出来的
潜在 他自己都还没意识到
显性需求你在竞品上都能看到,潜在需求才是差异化的来源
按会不会掏钱
有效 有意愿也有能力买单
无效 想要,但不会为它付出
「我会用的」和「我会付钱」之间隔着一整条鸿沟
按软件怎么落地
功能性
这个东西要能做什么。写需求时大部分人只写这一半
非功能性
多快、多稳、多好用。漏掉这一半,AI 生成的东西演示能过、上线就垮
Pay special attention to the last group. Say "build a search" and the AI will definitely do the functional part; "return results within one second across three thousand records"—that kind of non-functional requirement, you have to say, or it won't occur to it.