Validating the Value Proposition

Innovation & New Business Proposal - TUHH Institute of Entrepreneurship & Institute of Innovation Marketing, Hamburg · part of my Technology Management MBA · study notes for revision.

---

A value proposition is a promise written down: this group of people has this problem, and our offer relieves it well enough that they will pick us over whatever they do today. On a slide the promise costs nothing. The moment you take it seriously, it stops being one statement and becomes a stack of separate assumptions - about who the customer is, how often the problem bites, what they currently do instead, and whether our answer is worth switching for. Validation is the work of pulling those assumptions apart and putting each one in front of people who have no reason to be kind to us.

This session hands over the toolbox for doing that. There are three families of primary research: surveys, interviews and observations. They are not three flavours of the same thing and they are not interchangeable. Each one fits a different state of knowledge, answers a different kind of question, and fails in a different way. Choosing the wrong one is not a small inefficiency, it produces confident numbers about the wrong thing.

The last part of the session explains why all this discipline is needed at all. The honest answer is not that customers are mysterious. It is that our own heads are unreliable: founders systematically over-plan, over-commit, and hear what they hoped to hear. Market research is, among other things, a set of defences against ourselves.

Tip

The one idea to carry through. Match the method to what you still do not know. Interviews explore when you do not yet know which questions matter. Surveys count once you already know the question and only need the size of the answer. Observations reveal the things customers cannot tell you because they are not aware of them. And run all of it against the gap between what people say and what people do - the survey answer, the interview quote and the observed behaviour of the same person can point in three different directions, and the behaviour is usually the honest one.

1 · The starting question: what would you actually do?

The session opens with a live exercise. A team is launching an innovative pillow and has very little time for market and customer research. What would you do first? The class votes, and the tally is revealing:

Students	Research approach chosen	Concrete examples given
16	Use available data, focused on competition	Product reviews, competitor prices and products, gaps in the market, what the big fishes are doing, sleeping habits, health statistics, using AI to analyse the competition
14	Customer surveys	Satisfaction with existing pillows, DIY home decoration, using AI to analyse customer preferences
5	Interview sales and distribution experts	Finding effective distribution channels, phone calls to stores to test interest, shop-floor sales people, experts who know competitors’ cost of goods sold
4	Interview health experts	Doctors, experts from certification bodies
2	Talk directly to customers	Friends and family, visitors to stores
2	Run an experiment	A landing page, a marketing campaign
2	Observe customers online	Customer and medical forums, Facebook groups, Google Keyword Planner, Wikipedia analytics
1	Test pillow prototypes with customers	Family and friends

Three questions are put to the room: how do you evaluate this result, is there a preference for particular types of research, and is there a sequence of research activities that makes sense? The pattern speaks for itself. The overwhelming majority reach for desk data and surveys - activities you can do from your desk, that produce numbers, and that never require looking a customer in the eye. Only a handful plan to talk to an actual buyer, and exactly one person plans to put the product in someone’s hands.

That is the wrong way round, and the rest of the session explains why. Surveys are a tool for collecting exact information, not for exploring; interviews are the tool for exploring, and the rule for them is to do it early. If you survey before you have explored, you get precise measurements of the questions you happened to guess.

Exploreinterviews and observation - find out which questions matter

→

Measuresurveys and existing data - size the answers you now know to ask about

→

Testprototypes and experiments - see whether people actually act

The sequence that makes sense. Exploration comes first because it tells you what to measure; measurement without it is precision about a guess.

2 · Surveys versus interviews, side by side

Before looking at either method on its own, the session lays them next to each other on ten dimensions. Read this table as one long trade-off: everything a survey is good at, an interview is bad at, and the other way round.

Dimension	Surveys	Interviews
Sample size	Larger	Smaller
Representativity	Higher	Lower
Targeted sampling	More difficult	Easier
Appropriate response behaviour	Difficult to ensure	Easier to ensure
Costs per respondent	Lower	Higher
Convenience for the respondent	Higher	Lower
Number of questions	More	Fewer
Type of questions	Close-ended	Open-ended
Adaptability and flexibility	Lower	Higher
Richness of information	Lower	Higher

The trade-off has a shape: surveys buy breadth and cheapness by giving up depth and control, interviews buy depth and control by giving up breadth. Two rows deserve special attention. Targeted sampling is easier in interviews because you choose each person deliberately, while a survey link goes out to whoever is on the list. And appropriate response behaviour is easier to ensure in an interview because you can see whether the person understood the question, whereas in a survey you never find out that they misread it, skimmed it, or clicked through just to reach the end.

3 · Surveys - for counting, not for exploring

When surveys fitmarket characteristics

Consumer markets (B2C)high number of customers

Heterogeneous demandseveral segments you want to compare

What you must already havepreconditions

A representative customer sample

Clear hypothesesyou already know what you are testing

Issues that are not too complex

What comes outanalysis and applications

Numbersfrequencies, mean values, correlations

Status quo analysis of use · market segmentation · market size (TAM, SAM)

The survey profile. Notice the precondition in the middle tier: you need the hypothesis before you write the questionnaire, which is exactly why a survey cannot be your first move.

The typical questions a survey handles well fall into three groups:

• Obvious, quantifiable user problems. How many employees actively use journaling tools to manage their cognitive workload? How often do people give up on cooking a particular meal because they lack the equipment? How often do expats and long-term travellers need to refill prescription medication abroad?
• Exact measurement of psychological constructs. For example, the attitude towards booking personal services spontaneously or at short notice through a digital platform.
• Simple aspects of current use behaviour. For example, what exactly people who actively search for a firm’s CSR activities want to know.

Everything in that list is countable. Nothing in it asks the respondent to explain, reflect or narrate.

The three survey rules straight from the slides

• Use a survey for collecting exact information, not for exploring
• Use it to understand the big picture: quantities, distributions, means
• Use existing data rather than running your own survey wherever you can

The hard entry condition

• Recruit a large sample - or do not run the survey at all
• Use a representative sample - or do not run the survey at all
• A small, self-selected survey is worse than no survey: it produces numbers, and numbers get believed

How to build the questionnaire

• Tool and look. Standard online survey tools such as LimeSurvey, SurveyMonkey or Google Forms are enough. Apply a clean design and use visual cues.
• Time budget. Five minutes of response time is optimal, never more than ten. Keep the invitation email and the first page of the questionnaire very short and appealing.
• Opening question. Start with something very easy and genuinely interesting. Do not ask difficult questions that cannot be answered right away.
• Close-ended. Only a few open questions, or none at all. Open text is expensive for the respondent and rarely analysed properly.
• Rating scales. Five to seven points with clear verbal anchors: not at all true to very true, very little or never to very much or often, in no case to certainly.
• Reminders. One or two, sent fast. Most people who do not answer right away never answer at all.

4 · The EasyKitchen case - a clean answer to the wrong question

This is the case that makes the whole chapter click, so it is worth remembering in detail.

You are launching an online kitchen platform called EasyKitchen. You buy a list of 10,000 email addresses of people in Vienna who spent the previous month actively searching the internet for kitchens. You send them an advert with an invitation to register. You are delighted: 10 percent register and agree to be called by your call centre. Immediately after registering, those 1,000 people receive a short online questionnaire asking them to rate six concerns they had before signing up, on a scale where 1 means it was not a concern at all and 10 means it was a very big concern.

Concern put to the respondents	Rating (1 to 10)
Number of cooperating kitchen studios (only 8 in Vienna)	9
Time from order to delivery of the kitchen (average 6 months)	8
Variety of kitchen brands and appliance brands (limited)	7
Commission fee charged to the end customer (5 percent)	5
Fear of giving information to an unknown platform provider	3
Look and feel of the website (not yet perfect)	2

The class reads it the obvious way, and the readings are all sensible: the main pain is lack of variety, both in partner studios and in brands; delivery time is the second front, so either speed it up or communicate timelines transparently; and the low scorers - website design, the commission fee, data privacy - are less relevant issues that can wait. One answer groups the six items neatly into variety, benefit of the offer, and trust. The near-unanimous action list: onboard more studios, widen the brand range, fix delivery times. Every one of those conclusions is defensible given the data. The problem is the data.

Caution

Selection bias: you surveyed the wrong 1,000 people. The respondents had already cleared a hurdle - they agreed to register. They are, by construction, the people who were not stopped by mistrust, by the website, or by anything else. The interesting group is the 9,000 who did not register, and they are entirely absent from your dataset. Perhaps they did not trust an unknown company with their data. Perhaps they disliked the website. Perhaps for other reasons, and you have no idea what. It may well pay off more to invest in the website and in building trust than in signing up more kitchen studios - and your survey can never tell you that, because everyone who felt that way is missing from it.

This is the same shape as survivorship bias, the classic aircraft example: you only ever get to inspect the planes that came back, so you end up armouring the parts that were never the problem. In the kitchen case, the low scores for trust and website design do not prove those issues are unimportant. They prove that the people for whom those issues were decisive never became respondents. A low score on a survivor sample is not evidence of a small problem; it is evidence of a filtered sample. The practical lesson: before believing any survey, ask who is missing and which hurdle they failed to clear, then go and find some of them.

5 · Interviews - for exploring, not for testing

The interview profile
Market characteristics	Consumers and business customers, internal and external stakeholders, technology and market experts
Preconditions	Access to the interviewees, good interviewer skills, and issues that sit on the awareness level
Analysis and output	Rough transcripts, content analysis, basic insights supported by quotes
Main applications	Customer profile (pains, gains, jobs), use context analysis, first requirements analysis

The precondition in the middle is the quiet limit of the method: an interview reaches only as deep as the interviewee’s own awareness, which is exactly where observation takes over.

The typical questions interviews answer well are the ones a questionnaire would mangle:

• Complex buying and usage processes. What do transport authorities see as their biggest challenge in delivering real-time, multilingual accessibility? Why have drone coordination platforms been slow to emerge and scale despite a growing number of pilot programmes?
• Reasons behind choices, behaviour, feelings and opinions. For which specific tasks and use cases do people prefer to avoid AI assistants, and what would make them trust such tools more?

Four rules govern the method: use interviews for exploring rather than testing; use them for deep diving and qualitative insight into customer problems, use processes and key requirements; do it early; and do it yourself. That last one matters more than it looks. Outsourcing your interviews hands the surprises, the hesitations and the awkward silences to someone else, and those are the data.

Caution

Do not present your idea right away. The moment you describe your solution, the conversation changes character. People are prompted to be polite, they start evaluating your baby instead of describing their life, and the encouraging noises they make are worthless as evidence. Spend the interview on their world first; your idea can come at the end, if at all.

6 · What to do and what to avoid in a customer interview

Don’ts how interviews get ruined

• Talking most of the time
• Carrying 20 questions on a list, and asking every partner exactly the same ones
• Focusing on your idea or your solution
• Asking what they would do if… and what they plan to change in the future
• Asking about opinions
• Trying to sell your idea
• Hearing what you want to hear
• Trying to impress and to be clever
• Asking leading questions (Wouldn’t you agree that…)
• Interpreting immediately what they say

Dos how interviews produce evidence

• Listening most of the time
• Running a conversation around roughly five fundamental questions, and adapting them as you go
• Focusing on pains and gains
• Asking what they do, and what they have already tried in order to change it
• Asking about facts
• Trying to modify your idea
• Listening for what may invalidate your initial idea
• Trying to relax, and signalling that you need to learn
• Asking neutral, open-ended questions, and asking why
• Taking notes on what they actually said

The two columns are not style preferences. The left column produces agreeable noise; the right column produces facts you can be proven wrong by.

Three traps in that list deserve naming explicitly, because they are the ones that feel most natural:

• The hypothetical question. Would you use a service like this? People are poor forecasters of their own behaviour and generous when nothing is at stake. Replace it with history: what did you do the last three times this happened?
• The future-plans question. What are you planning to change next year? Plans are cheap. Ask instead what they have already tried - money spent, tools downloaded, workarounds built. Effort already sunk is the strongest signal of a real pain.
• The opinion question. What do you think of this? Opinions are free and socially shaped. Facts about behaviour are not.

Twenty identical scripted questions also defeat the purpose: the whole advantage of the interview over the survey was adaptability, and a rigid script throws it away while keeping all of the costs.

7 · Observations - watching what people actually do

Interviews stop where awareness stops. Observation is the method for everything below that line: routines people never think about, workarounds so habitual they have become invisible, and needs the person could not put into words if you asked.

The observation profile
Market characteristics	Usage processes; unaware usage problems and hidden needs
Preconditions	Permission to observe, good observer skills, measuring technology where it is needed
Analysis and output	Observation protocols: field notes, reflective notes, sketches, data from measurement technology
Main applications	Latent need analysis, user process and UX analysis, sensory market research

The output is a protocol, not an opinion, which is why the discipline of separating what you saw from what you concluded matters so much here.

The typical questions come in three levels of ambition:

1. Simple quantitative measures of behaviour. How long does each step - preparation, analysis, cleaning, data recording - take in current lab testing for multiple markers?
2. More complex observable use processes, covering moving, doing, seeing and saying. How do passengers react when they miss an announcement, do they ask the people around them? How do clients respond when shown a 2D, a 3D and an immersive design preview?
3. Physical status activities such as heart rate, skin resistance, facial expression and brain activity. How do knowledge workers physically behave, in gestures for instance, when they hit cognitive overload or decision fatigue, and what are observable proxies for measuring that fatigue?

The four rules for observing: dig underneath the level of awareness; identify the critical events in the use process rather than recording everything evenly; apply technology where it is feasible and reasonable; and observe different customers - customers against non-customers, heavy users against light users. That last rule is the antidote to the EasyKitchen mistake in observational form: the non-customer is the one who shows you why the product fails.

The classic illustration is the design agency IDEO redesigning a shopping cart, where the whole redesign starts from watching how people really move through a store rather than from asking them what they would like in a cart.

8 · Observation is not interpretation

The most common failure in observation is writing down a conclusion and calling it an observation. The session makes this vivid by recording the same scene twice:

Not an observation interpretation smuggled in

The woman is sitting in the food mall and enjoys her greasy, oil-soaked pizza. She is clearly oblivious of the caloric content and the damage it may do to her body.

What went wrong: enjoyment, obliviousness and health consequences all live inside the observer’s head. None of them was seen.

An observation what was actually available to the senses

The woman is sitting in the food mall. As she eats her piece of pizza, which drips oil, she says to a friend: this pizza is good.

Why it works: setting, action, visible detail and a verbatim quote. Someone else can disagree with your interpretation later without destroying the evidence.

The general warning behind the example: illusions and biased perception are strong enough to work even when we know they are happening. Knowing about an optical illusion does not make you see it correctly. The defence therefore has to be procedural rather than mental: record what happened, record it verbatim, and keep interpretation in a separate part of the protocol.

Where the behaviour of interest is physiological rather than visible, there is an instrument for it. The session lists the neural and biometric toolkit for measuring attention and psychological reactions such as emotion:

fMRI - high spatial resolution, locates activity in specific brain areas at all depthsEEG - detects activity between the left and right pre-frontal cortex, read as positive versus negative affectEye-tracking - measures response to visual stimuli, e.g. length of viewing and direction of movementBio-sensors - key physiological reactions taken as indicators of emotion, e.g. sweatFacial imaging - machine learning applied to the theory of facial expressions as emotions

These are the extreme end of sensory market research. The rule from the previous section still applies: use technology if it is feasible and reasonable, not because it is impressive.

9 · Lead users - a need forecasting laboratory

One more sampling idea belongs in this section. When you do early customer research, it can pay to deliberately look for lead users rather than average ones.

Lead users todaythey already face the need, often ahead of the market

→

Mainstream market tomorrowwhat lead users want today is what ordinary customers will want later

→

Need forecasting laboratoryyou observe the future instead of asking people to imagine it

Why lead users are worth the effort of finding. They do not need to imagine the future, because they are already living in it, which sidesteps the hypothetical-question problem from section 6.

This is an elegant escape from a trap we have already met. Asking an average customer what they will want in three years is a hypothetical question, and the answer is worth little. Asking a lead user what they need now is a factual question, and their present is a preview of the mainstream market’s future.

10 · Why we bother: the biases in our own heads

The closing stretch reframes the whole toolbox. Customer research is not only about learning facts about customers; it is a countermeasure against predictable flaws in the founder.

A quick demonstration of overconfidence. The class is asked to give intervals that should contain the true value with 90 percent probability, for three quantities: the size of the bitcoin blockchain in gigabytes, the number of students across all Harvard Business School MBA programmes in the class of 2023, and the share of the world trade ship fleet registered in the United States. In a group of 20 people without overconfidence, roughly 18 intervals should contain the true value and about 2 should miss. In practice far more people miss. The correct values are about 550 GB for the blockchain as of April 2024, 1,013 students in the HBS MBA class of 2023, and 0.62 percent of the world’s total gross registered tonnage under the US flag. The gap between the theoretical 18 and the observed result is overconfidence in our own estimation and forecasting ability.

Why our forecasts fail. We jump too quickly from past experience to conclusions about the future and lean too heavily on what we already know. We are too willing to believe in linear cause-and-effect relations and linear trends. We cling to stories about single examples instead of data covering the whole picture. Unlikely events tend to be hidden or unreported because they get labelled as irrelevant outliers, so we do not even consider black swans - the known unknowns: highly consequential but unlikely events that are practically unpredictable, genuinely impactful, and easy to explain only in retrospect through hindsight bias.

That family of errors sits under the heading of the planning fallacy, the first of four biases the session pins to the innovation and new business development process:

Planning fallacy overoptimism, risk underestimated

• Too focused and too optimistic thinking. The entire picture and the risks are not seen
• Consequence: no planning for potential critical events such as market barriers or negative customer feedback

Hypothesis confirmation bias seeking agreement

• Looking for confirmation of your hypotheses in order to avoid cognitive dissonance
• Produces extreme selective attention and distorted interpretation
• Consequence: disconfirming information is neglected, no reaction to negative signals

Endowment effect my idea is worth more

• We value things more highly once we own them, and the more effort, time and money we invested, the stronger the effect - the IKEA effect, labour leads to love
• Consequence: overcommitment to the initial idea, no alternatives considered

Egocentric empathy gap you are not the user

• Simplistic projection of your own preferences onto the targeted customers
• Consequence: inventors mainly create ideas that inventors like

So the question at the end of the session is a genuine choice: ignore the problem, or improve market research using professional methods. If we know which biases we are prone to, we can pick methods that push against them. Neutral open questions fight confirmation bias. Talking to non-customers fights selection bias. Verbatim observation protocols fight the endowment effect and egocentric projection, because they put evidence on the table that is hard to argue away. This is the point of the whole toolbox

Worked example

Take the pillow venture from section 1 and turn it into something testable.

The value proposition claim: Office workers aged 30 to 50 who wake up with neck pain will pay 89 euro for an adjustable-height pillow, because standard pillows do not match their sleeping position.

That single sentence hides at least six assumptions, and they are not equally risky.

#	Hidden assumption	Why it might be wrong	Risk
1	Enough office workers in this age band wake with neck pain	The segment may be far smaller than assumed	Medium
2	They blame the pillow for the pain	They may blame the desk chair, the mattress or stress	High
3	They have already tried to fix it and spent money on it	If they have never acted, the pain is real but not urgent	High
4	Pillow height is the deciding attribute	It could be firmness, temperature or material	High
5	89 euro is acceptable	Their reference price may be a 25 euro supermarket pillow	Medium
6	They would buy such a product online, unseen and untried	Pillows may be a touch-before-buying category	Medium

Assumptions 2, 3 and 4 would kill the business if false, so they get the research budget. None of them is a counting question yet, which rules out a survey as the first step.

The mini research plan.

Step	Method	Who	What exactly
A	8 to 10 interviews, run by the founders themselves	People who report morning neck pain, deliberately including some who never bought a special pillow (the non-customers)	Roughly five open questions, no product shown
B	Observation at home and in a shop	3 to 4 of the same people, plus store visitors	Watch how they set up the bed, how many pillows they stack, what they do in the shop before choosing
C	Survey, only afterwards	A large, representative panel of office workers	Count the frequency of the specific pain and workaround that the interviews actually revealed

The five interview questions - all about the past, all about facts:

1. Tell me about the last time you woke up with neck pain. What happened that day?
2. What have you already tried in order to make it better? Walk me through each attempt.
3. What did those attempts cost you, in money and in time?
4. How did you choose the pillow you sleep on now? Where were you, and what made you pick that one?
5. Why did the previous attempt not work well enough?

What would falsify the claim. If most interviewees name the mattress, the chair or their workload as the cause and never mention the pillow, assumption 2 fails and the proposition is dead in its current form. If nobody has ever spent money or effort on the problem, assumption 3 fails: the pain exists but is not worth paying to remove. If people describe choosing pillows by firmness or by squeezing them in a shop, and never mention height, assumption 4 fails and the product’s core feature is solving the wrong variable. Any of those three results is a good outcome of the research, because it costs a few weeks instead of a launch.

Apply it to your project

1. Write the claim as one sentence. Name the segment, the problem, the offer and the reason to switch. If it needs two sentences, you are describing two propositions.

2. Split it into assumptions and rank them. List every separate thing that must be true. For each one ask: how confident am I, and what happens to the business if it turns out false? Only the high-risk, low-confidence ones deserve fieldwork.

3. Choose the method by the state of your knowledge. Do not yet know which question matters, or need reasons and processes? Interview. Suspect the answer lies below awareness, in habits and routines? Observe. Already hold a sharp hypothesis and only need its size? Survey - and only if you can recruit a large, representative sample. Otherwise use existing data.

4. Design the instrument against the traps. For interviews: about five fundamental questions, adapted as you go, focused on pains, gains and what people have already done. No hypothetical questions, no future plans, no opinions, no leading phrasing, and do not show your solution first. For surveys: close-ended questions, five minutes, five to seven point scales, an easy opening question, one or two fast reminders.

5. Pick the sample on purpose, and include the people who said no. Ask who is missing from your list and which hurdle they failed to clear. Add non-customers, light users and, if you can find them, lead users.

6. Run the fieldwork yourself and record it raw. Rough transcripts and quotes for interviews; observation protocols that keep what you saw separate from what you concluded.

7. Decide honestly whether the proposition survived. Go back to the ranked assumptions and mark each one supported, refuted or still open. A proposition survives only if the killer assumptions were actually tested and held. If all you collected was agreement, you learned nothing - go back and look deliberately for the evidence that would invalidate the idea.

Key terms

Term	What it means in plain words
Value proposition	The promise that a defined customer group has a defined problem which your offer solves better than their current alternative
Validation	Breaking that promise into separate assumptions and testing the riskiest ones against real people
Survey	A large, close-ended, standardised questionnaire used to count things you already know to ask about
Interview	An open-ended conversation with a small, deliberately chosen sample, used to explore reasons, processes and pains
Observation	Watching what people actually do in their real context, including things they are not aware of
Selection or survivorship bias	Your sample contains only the people who already cleared some hurdle, so those stopped by the real problem are invisible and you end up fixing the wrong thing
Leading question	A question whose wording pushes the answer, such as asking whether someone would not agree that something is true
Hypothetical question	Asking what someone would do, or plans to do, instead of what they have actually done
Latent need	A need the customer has but cannot state, because it sits below the level of awareness
Lead user	A user who already experiences tomorrow’s mainstream need today, usable as a need forecasting laboratory
Planning fallacy	Overoptimism and underestimated risk, so critical events are never planned for
Hypothesis confirmation bias	Seeking evidence that agrees with you and disregarding the signals that do not
Endowment effect	Valuing your own idea more because it is yours, made stronger by the effort you put in (the IKEA effect)
Egocentric empathy gap	Assuming customers share your preferences, so you build what you personally like

Test yourself

1. On which dimensions do surveys beat interviews, and on which do interviews beat surveys? Name at least three of each.
2. In the EasyKitchen case the website concern scored 2 out of 10 and the number of studios scored 9 out of 10. Why is it still risky to deprioritise the website?
3. Give three things you should never do in a customer interview, and say what you should do instead in each case.
4. Why does the session say interviews only reach issues on the awareness level, and which method covers what lies below that line?
5. Rewrite as a proper observation: the user was frustrated by the confusing checkout page and gave up.
6. Which cognitive bias is at work when a founder who has spent nine months building a prototype refuses to consider alternative concepts, and which research habit pushes against it?

Answers

1. Surveys win on sample size, representativity, cost per respondent, convenience for the respondent and the number of questions you can ask. Interviews win on targeted sampling, ensuring appropriate response behaviour, adaptability and flexibility, richness of information, and the ability to use open-ended questions.
2. Because of selection bias. Only the 1,000 people who agreed to register answered; anyone put off by the website or by mistrust never registered and is therefore absent from the sample. The low score merely shows that the website did not stop the people it did not stop, which is circular. The missing 9,000 may have been lost exactly there.
3. Any three of: do not talk most of the time (listen instead); do not present or sell your idea early (people turn polite and evaluate instead of describing); do not ask hypothetical or future-plan questions (ask what they did and what they have already tried); do not ask leading questions (ask neutral open questions and ask why); do not interpret on the spot (take notes on what they actually said).
4. Because an interview can only surface what the interviewee consciously notices and can put into words. Habits, workarounds and hidden needs sit below that line, and observation is the method that reaches them, because it digs underneath the level of awareness.
5. Something like: the user scrolled to the bottom of the checkout page, moved the cursor back and forth across the payment section for about 20 seconds, said this is not what I expected, and closed the tab. Setting, visible action, timing and a verbatim quote. The frustration is your interpretation and belongs in a separate note.
6. The endowment effect, strengthened by the effort invested (the IKEA effect), usually together with hypothesis confirmation bias. The countermeasure is to design research that actively hunts for evidence which would invalidate the idea: neutral open questions, non-customers in the sample, and verbatim protocols instead of remembered impressions.

Revision summary

Tip

The one-minute recap

• A value proposition is a stack of assumptions. Validation means ranking them by risk and testing the killers, not collecting agreement.
• Sequence matters: explore with interviews and observation, then measure with surveys and existing data, then test with prototypes and experiments. Most teams instinctively do this backwards, as the pillow exercise showed.
• Surveys buy breadth cheaply and lose depth: larger samples, higher representativity, lower cost per respondent, close-ended questions, lower richness. Use them for exact information, not for exploring. Large and representative sample, or do not run one at all. Design rules: five minutes, never more than ten, close-ended, easy opening question, five to seven point scales, one or two fast reminders.
• EasyKitchen is the warning: the 1,000 registered respondents had already passed a hurdle, so the 9,000 who refused are invisible. That is selection bias, the same shape as survivorship bias. Always ask who is missing from the sample.
• Interviews are for exploring, not testing: about five fundamental questions, adapted live, focused on pains and gains, asking what people did and what they have already tried. Do it early, do it yourself, and do not present your idea right away. Avoid leading, hypothetical, future-plan and opinion questions.
• Observations dig underneath awareness, target critical events in the use process, apply technology where reasonable, and compare different customers including non-customers and light users. The output is a protocol: separate what you saw from what you concluded, as the pizza example shows.
• Lead users already live in the mainstream market’s future, so asking them about their present replaces asking ordinary customers to forecast.
• We do all of this because of our own biases: planning fallacy, hypothesis confirmation bias, endowment effect and egocentric empathy gap. Knowing them only helps if you then choose methods that fight them.

Next: The Customer Journey & Service Blueprinting → - mapping the experience end to end.