AgentSkillsCN

falsification-driven-user-research

将用户研究从“形式上的验证”转向“战略性学习”。当你忍不住想要发起一项“快速调研以验证假设”，当研究工作似乎拖慢了产品交付进度，或当你亟需弄清A/B测试结果背后的“深层原因”时，不妨尝试这一方法。

SKILL.md
--- frontmatter
name: falsification-driven-user-research
description: Shift user research from "performative validation" to strategic learning. Use this when you are tempted to ask for a "quick study to validate assumptions," when research feels like it's slowing down shipping, or when you need to understand the "why" behind an A/B test result.

Falsification-Driven User Research

Traditional user research often falls into "user-centered performance"—the act of conducting studies to signal customer obsession without actually intending to change a decision. To drive real business impact, research must shift from seeking validation (looking for proof you are right) to falsification (actively looking to be proven wrong).

The Three Altitudes of Research

Align your research requests to one of these three levels to avoid "middle-range" research that is interesting but un-actionable.

  1. •Macro Research: Strategic, forward-looking, and business-focused. Use this for annual planning, identifying TAM (Total Addressable Market), or "concept car" projects looking 3-5 years out.
  2. •Micro Research: Tactical, high-precision usability and optimization. Use this for A/B test "why" analysis, perfecting CTA language, or fixing specific friction points in a funnel.
  3. •Middle-Range Research (The "Danger Zone"): General questions about "how users feel" that aren't pointed at a specific business problem. Avoid this unless it is explicitly tied to a conversion funnel or a specific strategic pivot.

How to Move from Performance to Impact

1. Adopt the Falsification Mindset

Stop asking, "Can you validate this design?" instead ask, "What evidence would prove this approach is the wrong way to solve the user's problem?"

  • •Goal: Expose blind spots and biases in your intuition.
  • •Success Metric: You shouldn't want to hold a decision-making meeting without your research partner present because their insights are critical to the "go/no-go" choice.

2. Connect Research to Profit

Researchers must speak the language of the business to be impactful.

  • •Quarterly Alignment: Read the latest quarterly report and shareholder calls. Align research questions with the current OKRs and conversion funnel.
  • •Unified Metrics: The PM, Designer, and Researcher should share the exact same metrics for success. Research is not a separate workstream; it is a component of the product's success.

3. Deploy the "Five Tools"

A high-impact researcher (or a PM doing their own research) should utilize these five technical tools:

  • •Formative/Generative: Ethnographic field work to find new opportunities.
  • •Evaluative: Rapid usability testing to find functional bugs.
  • •Survey Design: Rigorous, unbiased scaling of user feedback (prefer CSAT over NPS).
  • •Applied Statistics: Understanding significance in the context of A/B testing.
  • •Data Querying: Using SQL, dashboards, or AI prompting to pull your own behavioral data.

Metrics to Use (And What to Avoid)

  • •Use CSAT (Customer Satisfaction): Ask "Overall, how satisfied are you with [Experience]?" It has better data properties and correlates more strongly with business outcomes.
  • •Avoid NPS (Net Promoter Score): The "likelihood to recommend" question is often irrelevant to the product (e.g., enterprise software) and the 11-point scale is statistically noisy and prone to formatting bias on mobile.

Examples

Example 1: Micro-Research (The Multi-Million Dollar Button)

  • •Context: An Airbnb checkout button had low conversion. PMs suspected the UI was ugly.
  • •Application: Evaluative research showed users were afraid the button would charge them immediately.
  • •Output: Changed the text (7 characters) to clarify it was just the "next step." This drove a 1% lift in conversion, worth millions in revenue, in under 48 hours.

Example 2: Falsifying Intuition (The "Super-Hider")

  • •Context: Facebook engineers assumed that "hiding a post" was a signal of low quality. They planned to deprioritize all hidden posts.
  • •Application: Formative research (think-aloud study) observed a power user who hid every post she liked.
  • •Output: Researchers discovered she used "Hide" as an "Inbox Zero" marker. The team falsified the "Hide = Bad" assumption and saved the feed from a broken ranking algorithm.

Common Pitfalls to Avoid

  • •Treating Research as a Service: Calling a researcher at the end of a project to "bless" a design. This is 97% performance and 3% learning.
  • •Relying on "Post-hoc" Logic: Saying "we knew that already" after a study. This is hindsight bias. Acknowledge that while it seems obvious now, you didn't have the conviction to act until the data arrived.
  • •The "Henry Ford" Trap: Avoiding research because "users don't know what they want." Great researchers don't ask users what they want; they observe behavior to identify latent needs.
  • •A/B Test Speculation: Seeing a stat-sig drop in a test and guessing why. Use a 48-hour qualitative study to get the "why" instead of running three more "guess-and-check" experiments.
  • •Biased Walkthroughs: Dogfooding your own product and assuming your experience is universal. You are not the user; you lack their constraints, priorities, and technical limitations.