Start with clarity: define the situation. Start by defining your target ideal customer profile (ICP), their functional, social, and emotional needs (JTBD), and the key objectives for what you're trying to learn. Do not skip this step – ambiguity here contaminates everything downstream.
Collect data: different sources = different functions. Use Research, Simulations, Interviews, and Observations (RSIO) to collect data that reveal different insights into your customers. Each method catches what the others miss.
Research tells you what the world already knows. You should use AI to synthesize public data (market reports, competitor analysis, Reddit, forums, and customer reviews) to get a top-down, outsider view of your market.
Simulations tell you the rational behavior of your customers. You should use AI to simulate your ICP, pressure-test assumptions, and learn customer-specific jargon before you enter the field.
Interviews tell you the human customer story. This is the counter to rational behavior — idiosyncrasies, irrational workflows, and unexpected emotions live here. You should conduct intentional interviews with your ICP using unbiased, open-ended questions.
Observations tell you how people actually act. You should use AI to build lo-fi tests, MVPs, and functional workflows that let you observe your customer in action. As the cost of building approaches $0, "launch" is no longer a significant milestone – it's a standard part of the discovery process.
Analyze your outcomes: find "Informational Alpha." This is the valuable difference between your predicted and actual outcomes. You should use AI to review your RSIO data for "aha!" moments – common pain points, behaviors, and language that will differentiate your solution.
Integrate your results: put insights to work. Take your informational alpha and apply it directly to your business – update sales narratives, marketing copy, product roadmaps, and positioning. Trust your RSIO results and directly address questions defined in Step 1.
Repeat the process: discovery never ends. Take your newfound knowledge and redefine your situation – update your ICP, update your objectives, or shift directions entirely and start again. Customer discovery is continuous because customer needs never stop moving – and your competitive edge depends on your ability to keep up.
Customer Discovery is the iterative process of learning about a specific customer in order to answer two questions: does a problem exist, and does my solution solve it? While the Customer Discovery Process continues to be the most effective way to address both of those questions, AI has fundamentally changed how that process actually works.
Customer discovery is a continuous, iterative process. It is ongoing and does not end. Its purpose is to answer two questions: (1) does a problem exist for a target customer, and (2) does my solution solve it?
The process starts with clear definitions of your target customer (ICP, persona) and objectives (goals, unknowns, outcomes).
Then, collect RSIO (Research, Simulations, Interviews, and Observations) from your target customer.
Sift through your data for "informational alpha" – the insights and "aha moments" that will differentiate your solution.
Integrate outcomes with your business and repeat.
BEFORE YOU START
Define Your Customer
Segment before you begin. Every time. Choose a specific person with a specific context. Without a clearly defined customer, you are collecting noise. Develop an ICP — an Ideal Customer Profile — that describes who you are trying to learn about before you ask a single question.SECTION: QUESTION 1
Does the Problem Exist?
Your hypothesis: this problem is real, frequent, and painful for this specific person. Use all three methods to collect evidence:
Simulations
Before talking to real customers, simulate the conversation with AI. Feed it your persona and hypothesis. Ask it to push back. This gives you a baseline — the obvious answers, the expected objections, the language of the space. Use it to sharpen your questions before you spend social capital on real conversations.
Interviews
Talk to real people. Ask about the past, not the future. "Tell me about the last time you dealt with this" is gold. "Would you buy this?" is worthless. Listen 80% of the time. Use the 5 Whys. Look for emotion — frustration, embarrassment, workarounds. That is where the real signal lives.
Observations
Watch what people do, not just what they say. People describe idealized behavior in interviews. Observations catch real behavior. This includes watching someone use a lo-fi prototype, an AI-built demo, or an existing product. What confuses them? Where do they hesitate? What do they ignore entirely?SECTION: QUESTION 2
Does My Solution Solve It?
Your hypothesis: my solution changes how this specific person experiences this problem.
Use the same three methods — Simulations, Interviews, Observations — pointed at your solution instead of the problem.
Show something real. With AI tools, the same hour that used to produce a wireframe now produces a functional prototype. There is no longer an excuse to ask customers to react to a description. Build something. Put it in front of them. Watch what happens.SECTION: OUTPUT
Iterative Learning
After every cycle, reflect. Not just on the answers — but on the context. Where does this problem live in the customer's workflow? Where does the solution fit? What did you learn that changes what you need to find out next?
Discovery doesn't end at launch. The best founders are still learning when their product has 10,000 users. Retention isn't a retention problem — it's a discovery problem.SECTION: THE ALPHA
Informational Alpha
The most important thing AI cannot do is surprise you.
AI simulations are trained on existing patterns. They will give you the expected answer. Real customers will not.
The gap between what an AI simulation predicts and what a real customer actually says — that gap is your alpha. It shows up as irrational behavior, unexpected emotion, a workaround nobody should have to use, a frustration that doesn't fit the model.
Your job in every interview and every observation is to find that gap. That is where the real insight lives.SECTION: THE TENSION
The Discovery/Sales Tension
In practice — especially with a small customer base — discovery and sales are not separate. You are asking questions and you are pitching. The customer knows it. You know it. The mistake is pretending otherwise.
· Name it internally. Start in discovery mode. Earn the right to pitch.
· Separate conversations when possible. First = discovery. Second = pitch.
· Sequence carefully when you can't separate them. Ask all discovery questions before mentioning your solution.
· Watch for false positives. "That sounds interesting" is not signal.
· Track what you're learning, not just what you're hearing.SECTION: QUICK REFERENCE
Always
· Define your customer before starting
· Ask about the past, not the future
· Listen 80%, talk 20%
· Simulate with AI before real interviews
· Build something real before your 5th conversation
· Treat discovery as permanent
Never
· Ask "would you buy this?"
· Fill silence — let it breathe
· Treat simulation as a substitute for real conversations
· Trust your memory — record and analyze transcripts
· Show slides when you could show a demo
· Declare discovery doneFOOTER
Prepared by Max Stern · Duke University, Innovation & Entrepreneurship · 2026 · AI-assisted research and drafting · Living document.
Questions or feedback: [email protected] · linkedin.com/in/m-sternCustomer Discovery is the iterative process of learning about a specific customer in order to answer two questions: does a problem exist, and does my solution solve said problem? While the Customer Discovery Process continues to be the most effective way to address both of those questions, AI has fundamentally changed how that process actually works.
BEFORE YOU START
Define Your Customer
Segment before you begin. Every time. Choose a specific person with a specific context. Without a clearly defined customer, you are collecting noise. Develop an ICP — an Ideal Customer Profile — that describes who you are trying to learn about before you ask a single question.SECTION: QUESTION 1
Does the Problem Exist?
Your hypothesis: this problem is real, frequent, and painful for this specific person. Use all three methods to collect evidence:
Simulations
Before talking to real customers, simulate the conversation with AI. Feed it your persona and hypothesis. Ask it to push back. This gives you a baseline — the obvious answers, the expected objections, the language of the space. Use it to sharpen your questions before you spend social capital on real conversations.
Interviews
Talk to real people. Ask about the past, not the future. "Tell me about the last time you dealt with this" is gold. "Would you buy this?" is worthless. Listen 80% of the time. Use the 5 Whys. Look for emotion — frustration, embarrassment, workarounds. That is where the real signal lives.
Observations
Watch what people do, not just what they say. People describe idealized behavior in interviews. Observations catch real behavior. This includes watching someone use a lo-fi prototype, an AI-built demo, or an existing product. What confuses them? Where do they hesitate? What do they ignore entirely?SECTION: QUESTION 2
Does My Solution Solve It?
Your hypothesis: my solution changes how this specific person experiences this problem.
Use the same three methods — Simulations, Interviews, Observations — pointed at your solution instead of the problem.
Show something real. With AI tools, the same hour that used to produce a wireframe now produces a functional prototype. There is no longer an excuse to ask customers to react to a description. Build something. Put it in front of them. Watch what happens.SECTION: OUTPUT
Iterative Learning
After every cycle, reflect. Not just on the answers — but on the context. Where does this problem live in the customer's workflow? Where does the solution fit? What did you learn that changes what you need to find out next?
Discovery doesn't end at launch. The best founders are still learning when their product has 10,000 users. Retention isn't a retention problem — it's a discovery problem.SECTION: THE ALPHA
Informational Alpha
The most important thing AI cannot do is surprise you.
AI simulations are trained on existing patterns. They will give you the expected answer. Real customers will not.
The gap between what an AI simulation predicts and what a real customer actually says — that gap is your alpha. It shows up as irrational behavior, unexpected emotion, a workaround nobody should have to use, a frustration that doesn't fit the model.
Your job in every interview and every observation is to find that gap. That is where the real insight lives.SECTION: THE TENSION
The Discovery/Sales Tension
In practice — especially with a small customer base — discovery and sales are not separate. You are asking questions and you are pitching. The customer knows it. You know it. The mistake is pretending otherwise.
· Name it internally. Start in discovery mode. Earn the right to pitch.
· Separate conversations when possible. First = discovery. Second = pitch.
· Sequence carefully when you can't separate them. Ask all discovery questions before mentioning your solution.
· Watch for false positives. "That sounds interesting" is not signal.
· Track what you're learning, not just what you're hearing.SECTION: QUICK REFERENCE
Always
· Define your customer before starting
· Ask about the past, not the future
· Listen 80%, talk 20%
· Simulate with AI before real interviews
· Build something real before your 5th conversation
· Treat discovery as permanent
Never
· Ask "would you buy this?"
· Fill silence — let it breathe
· Treat simulation as a substitute for real conversations
· Trust your memory — record and analyze transcripts
· Show slides when you could show a demo
· Declare discovery doneFOOTER
Prepared by Max Stern · Duke University, Innovation & Entrepreneurship · 2026 · AI-assisted research and drafting · Living document.
Questions or feedback: [email protected] · linkedin.com/in/m-sternCustomer Discovery is the iterative process of learning about a specific customer in order to answer two questions: does a problem exist, and does my solution solve said problem? While the Customer Discovery Process continues to be the most effective way to address both of those questions, AI has fundamentally changed how that process actually works.
Formats typically include some combination of 1:1 sessions, team workshops, and asynchronous audits.
Deliverables typically include breakthrough solutions to specific business challenges, the direct implementation of technical systems, and an aligned team prepared to tackle similar problems in the future.
BEFORE YOU START
Define Your Customer
Segment before you begin. Every time. Choose a specific person with a specific context. Without a clearly defined customer, you are collecting noise. Develop an ICP — an Ideal Customer Profile — that describes who you are trying to learn about before you ask a single question.SECTION: QUESTION 1
Does the Problem Exist?
Your hypothesis: this problem is real, frequent, and painful for this specific person. Use all three methods to collect evidence:
Simulations
Before talking to real customers, simulate the conversation with AI. Feed it your persona and hypothesis. Ask it to push back. This gives you a baseline — the obvious answers, the expected objections, the language of the space. Use it to sharpen your questions before you spend social capital on real conversations.
Interviews
Talk to real people. Ask about the past, not the future. "Tell me about the last time you dealt with this" is gold. "Would you buy this?" is worthless. Listen 80% of the time. Use the 5 Whys. Look for emotion — frustration, embarrassment, workarounds. That is where the real signal lives.
Observations
Watch what people do, not just what they say. People describe idealized behavior in interviews. Observations catch real behavior. This includes watching someone use a lo-fi prototype, an AI-built demo, or an existing product. What confuses them? Where do they hesitate? What do they ignore entirely?SECTION: QUESTION 2
Does My Solution Solve It?
Your hypothesis: my solution changes how this specific person experiences this problem.
Use the same three methods — Simulations, Interviews, Observations — pointed at your solution instead of the problem.
Show something real. With AI tools, the same hour that used to produce a wireframe now produces a functional prototype. There is no longer an excuse to ask customers to react to a description. Build something. Put it in front of them. Watch what happens.SECTION: OUTPUT
Iterative Learning
After every cycle, reflect. Not just on the answers — but on the context. Where does this problem live in the customer's workflow? Where does the solution fit? What did you learn that changes what you need to find out next?
Discovery doesn't end at launch. The best founders are still learning when their product has 10,000 users. Retention isn't a retention problem — it's a discovery problem.SECTION: THE ALPHA
Informational Alpha
The most important thing AI cannot do is surprise you.
AI simulations are trained on existing patterns. They will give you the expected answer. Real customers will not.
The gap between what an AI simulation predicts and what a real customer actually says — that gap is your alpha. It shows up as irrational behavior, unexpected emotion, a workaround nobody should have to use, a frustration that doesn't fit the model.
Your job in every interview and every observation is to find that gap. That is where the real insight lives.SECTION: THE TENSION
The Discovery/Sales Tension
In practice — especially with a small customer base — discovery and sales are not separate. You are asking questions and you are pitching. The customer knows it. You know it. The mistake is pretending otherwise.
· Name it internally. Start in discovery mode. Earn the right to pitch.
· Separate conversations when possible. First = discovery. Second = pitch.
· Sequence carefully when you can't separate them. Ask all discovery questions before mentioning your solution.
· Watch for false positives. "That sounds interesting" is not signal.
· Track what you're learning, not just what you're hearing.SECTION: QUICK REFERENCE
Always
· Define your customer before starting
· Ask about the past, not the future
· Listen 80%, talk 20%
· Simulate with AI before real interviews
· Build something real before your 5th conversation
· Treat discovery as permanent
Never
· Ask "would you buy this?"
· Fill silence — let it breathe
· Treat simulation as a substitute for real conversations
· Trust your memory — record and analyze transcripts
· Show slides when you could show a demo
· Declare discovery doneFOOTER
Prepared by Max Stern · Duke University, Innovation & Entrepreneurship · 2026 · AI-assisted research and drafting · Living document.
Questions or feedback: [email protected] · linkedin.com/in/m-sternCustomer Discovery is the iterative process of learning about a specific customer in order to answer two questions: does a problem exist, and does my solution solve said problem? While the Customer Discovery Process continues to be the most effective way to address both of those questions, AI has fundamentally changed how that process actually works.
100
200
300
400
500
600
700
800
900