Avoiding Bad Data: Compliments, Fluff, and Ideas skill

Avoiding Bad Data: Compliments, Fluff, and Ideas

Bad data is worse than no data because it gives you false confidence. You build the wrong thing, launch to crickets, and can't figure out what went wrong because "everyone said they loved it." This reference covers the three types of bad data, how to recognize them in real-time, and specific techniques for deflecting each type back to useful information.

The Three Types of Bad Data

Type 1: Compliments

Compliments are the most common and most dangerous form of bad data. They feel amazing in the moment and are completely worthless for decision-making.

What compliments sound like:

• "That's a really cool idea!"
• "I can definitely see myself using that."
• "You guys are going to crush it."
• "This is exactly what the market needs."
• "I love it. When can I get it?"
• "You're solving a real problem."
• "My team would love this."

Why compliments are dangerous: Compliments contain zero information about whether someone will change their behavior, pay money, or use your product. They're a social reflex -- the same way you say "I'm good" when someone asks how you are. People compliment you because:

• They want to be supportive
• Saying "bad idea" feels cruel
• They haven't thought about it deeply enough to have a real opinion
• They want the conversation to end pleasantly

How to deflect compliments:

Compliment	Deflection
"That's a great idea!"	"Thanks! But tell me -- how are you dealing with this problem right now?"
"I'd definitely use that"	"That's encouraging. What are you currently using? What's frustrating about it?"
"My team would love this"	"What's your team struggling with specifically? Walk me through a recent example."
"You're going to crush it"	"I appreciate that. To make sure we build the right thing -- what's the biggest pain point in your workflow today?"
"This is exactly what we need"	"What have you tried before? Why didn't those solutions work?"

The golden rule: Every time you receive a compliment, convert it into a question about their life and past behavior.

Type 2: Fluff

Fluff is vague, generic, or hypothetical talk that sounds informative but carries no real-world weight. It's the verbal equivalent of empty calories.

Three sub-types of fluff:

Generic Claims

• "I usually..." (without a specific instance)
• "I always..." (really? always?)
• "I never..." (provably false in most cases)
• "We generally..." (who, specifically? when?)

Deflection: "Can you give me a specific example of the last time that happened?"

Hypothetical Promises

• "I would definitely..." (but you haven't)
• "If you built that, I'd..." (future tense = fiction)
• "I think I'd probably..." (double hedging)
• "That would be worth at least $X to me" (imaginary money)

Deflection: "You mentioned you'd pay $X. What's the most you've actually paid for a tool like this?"

Future-Tense Predictions

• "Next quarter, we're planning to..."
• "We'll probably need something like this soon"
• "I'm going to start looking for a solution"
• "That's on our roadmap for later this year"

Deflection: "When did you first realize this was a problem? What have you done about it so far?"

The fluff test: If you can't assign a date, a dollar amount, or a specific event to what they said, it's fluff. Dig until you hit concrete ground.

Type 3: Ideas (Unsolicited Feature Requests)

When people understand what you're building, they start designing your product for you. This feels collaborative and exciting, but it's usually misleading because they're solving an imagined version of the problem.

What idea-giving sounds like:

• "You should totally add [feature]"
• "It would be amazing if it could also [function]"
• "Have you thought about integrating with [tool]?"
• "What if it also did [tangentially related thing]?"
• "The real killer feature would be [thing they thought of in the last 30 seconds]"

Why ideas are dangerous: Feature requests are solutions, not problems. When someone says "you should add a chat feature," you don't know:

• What problem they're trying to solve
• Whether they've actually experienced that problem
• Whether they'd use the feature if you built it
• Whether this is a deal-breaker or a nice-to-have

How to deflect ideas back to problems:

Idea	Deflection
"You should add Slack integration"	"That's interesting. How does Slack fit into your current workflow for this?"
"It needs a mobile app"	"Tell me about the last time you needed to do this on mobile. What happened?"
"Add AI to automate the reports"	"Walk me through how you create reports today. Where's the most time spent?"
"You need a dashboard"	"What metrics are you currently tracking? How? What do you do when they change?"

The idea deflection formula:

1. Acknowledge: "That's an interesting idea."
2. Record: Write it down (shows respect).
3. Dig: "What's driving that? Tell me about the last time you needed something like that."

The Approval-Seeking Trap

The most subtle form of bad data collection is when you unconsciously seek validation instead of information. This happens when:

You're Fishing for Compliments Without Realizing It

Signs you're fishing:

• You describe the idea in glowing terms before asking questions
• You lead with the solution and then ask "what do you think?"
• You show excitement and energy that makes disagreement feel rude
• You ask leading questions: "Don't you think it would be better if...?"
• You share your vision for 10 minutes, then ask a token question

Signs you're genuinely learning:

• You haven't mentioned your idea yet and they're doing most of the talking
• You're asking about their problems, not your solution
• You're comfortable with silence after a question
• You're genuinely curious about answers that might kill your idea
• You're taking notes on what they say, not on how they react to your pitch

The Pitch-Creep Problem

Pitch-creep happens when a learning conversation gradually becomes a sales pitch. It usually follows this pattern:

1. You start with good questions about their problems (learning mode)
2. They describe a problem you can solve (excitement builds)
3. You say "Actually, that's exactly what we're building!" (pitch mode activated)
4. The rest of the conversation is about your solution (learning stops)
5. They say "That sounds great!" (compliment received)
6. You leave feeling validated (but you learned nothing new after step 2)

How to prevent pitch-creep:

• Set a rule: no pitching until the last 5 minutes
• Bring a partner who kicks you under the table when you start pitching
• Write "SHUT UP" on the top of your notepad
• Practice the phrase: "That's really helpful context. Can you tell me more about [their problem]?"

The "Would You Buy" Trap

"Would you buy this?" is the single most popular and most useless question in customer development. Here's why:

The scenario: You describe your product. You ask "would you buy this?" They say "yes." You feel great.

The reality: Of course they said yes. Saying "no" to your face would be uncomfortable. They're not lying -- in this hypothetical moment, they genuinely believe they would. But this belief has no predictive power because:

• They haven't felt the pain of actually parting with money
• They haven't compared your solution to alternatives
• They haven't considered whether this is a top priority
• They haven't thought about the switching cost from their current workflow
• They're answering in a context of social pressure (you're sitting right there)

What to ask instead:

Instead of...	Ask...
"Would you buy this?"	"What are you currently spending on this problem?"
"Would you pay $50/month?"	"What's your budget for tools in this category?"
"Is this worth paying for?"	"What have you tried before? What did you pay for it?"
"Would your company buy this?"	"How does your company typically buy new tools? Who decides? What's the process?"

Real-Time Bad Data Detection

During a conversation, monitor for these warning signs:

Conversation Quality Scorecard

Signal	Score	Meaning
They're describing specific past events	+3	You're getting real data
They're using exact numbers (dates, dollars, hours)	+3	High-quality factual data
They're showing you their current workflow	+2	Observable behavior
They're volunteering problems you didn't ask about	+2	Genuine pain points
They're offering to connect you to someone	+2	Reputation commitment
They're nodding and saying "great idea" a lot	-2	Compliment mode
They're using "I would" or "I usually" without specifics	-2	Fluff mode
They're suggesting features	-1	Idea mode (dig for the problem)
You've been talking for more than 2 minutes straight	-3	You're pitching, not learning
They seem eager to end the conversation	-3	They're being polite

Running score interpretation:

• Positive score: You're learning. Keep going.
• Zero or negative: You've drifted into bad data territory. Reset with a behavior question.

Recovery Phrases

When you realize you're in bad data territory, use these phrases to reset:

• "I appreciate the encouragement. To make sure we get this right -- can you walk me through how you dealt with this last week?"
• "That's helpful. Let me back up -- tell me about the last time this problem actually cost you time or money."
• "I want to make sure I'm not just hearing what I want to hear. What would make this NOT work for you?"
• "Let's put my idea aside for a second. What's the biggest headache in your [relevant area] right now?"
• "I hear you saying you'd use this. Help me understand -- what would you stop doing if you started using something like this?"

The Post-Conversation Gut Check

After every conversation, ask yourself these five questions:

1. Did I learn any new facts I didn't know before? If no, you were probably collecting compliments.

2. Can I summarize what I learned without mentioning my product? If no, the conversation was about your idea, not their life.

3. Did anything surprise me or challenge my assumptions? If no, you were probably asking leading questions.

4. Did they give a concrete commitment? If no, you may have a zombie lead.

5. Would my co-founder learn something new from these notes? If no, the notes contain opinions rather than facts.

If you answer "no" to three or more of these questions, the conversation produced bad data. Don't count it as validation. Learn from the mistake and adjust your approach for the next conversation.

The Emotional Cost of Good Conversations

Good conversations are uncomfortable. You will hear things like:

• "I don't think I'd actually pay for that."
• "That's not really a problem for me."
• "I already have a solution that works fine."
• "I don't think this is a priority."

These responses feel bad in the moment but are enormously valuable. A painful truth heard early saves you months of building the wrong thing. The emotional cost of a hard conversation is tiny compared to the cost of building a product nobody wants.

Reframe: A conversation that kills a bad idea is the most valuable conversation you'll ever have. It just saved you a year of your life.