Usually something already broke. Deals stall in the same place, churn ticked up, the messaging everyone liked converts badly, or somebody asked who the buyer actually is and the room went quiet. You have been told to talk to customers. You have weeks, not a quarter.
The mistake is starting from the method. Start from the symptom and every other decision follows from it, in order.
Five decisions, and only the first one is yours to make freely. Name the symptom and the rest of the chain is largely determined.
| If this is happening | You need to learn | Who experienced it | Run | Reach |
|---|---|---|---|---|
| Deals go quiet after the demo and never formally die | Why buyers with the problem still don't move | Prospects who never decided | No-decision | Hardest |
| We win deals and cannot explain why | The trigger, the timeline, the consideration set | Customers who switched to you recently | Switching | Easy |
| They finish onboarding and never come back | Where value stopped being realized | Recent customers inside first value | Adoption | Easiest |
| Health scores were green right up until the cancellation | When the decision to leave was actually made | Churned accounts, ideally within 90 days | Firing | Hard |
| We do not know who we actually lose to | What they hired instead, and what it had to do | Churned accounts and lost deals | Firing | Hard |
| Accounts renew but never grow | What makes a second purchase feel obvious | Accounts that grew, and comparable ones that didn't | Expansion | Easy |
| Messaging tests well internally and converts badly | The words customers actually use for the job | All roles on the buying committee | Core jobs | Varies |
| The team cannot agree on what customers are trying to do | The progress being sought, before any goal is chosen | All roles on the buying committee | Core jobs | Varies |
The sections below explain why each pairing holds, and what the obvious alternative misses. If you already know your symptom, the table is the answer and the rest is the reasoning.
This page is the decision. The six guides are the questions, all 156 of them published in full, free to use and cite. Once you know which interview to run, that is where the actual script lives.
Not stages in a sequence. Goals a team designs for, usually several at once. Each fails for a different reason, and the useful question is which one is the binding constraint right now.
Attracts new customers, users, partners, or opportunities by making adoption more likely.
Switching interviews and no-decision interviews, as a pair.
Why both. Switching gives you the trigger, the timeline, and what else was on the table. No-decision gives you the larger group that never moved at all. Run only switching and you get a flattering, incomplete picture, because you learn what a win looks like without learning what the market's actual default is.
Why not win/loss alone. Win/loss studies people who crossed the decision threshold. In B2B the status quo is usually the largest competitor in the category, and it never files a closed-lost reason. Observed customers are a selected population, and the selection removed the biggest outcome category from the study of outcomes.
Start with one. If you have recent wins worth reconstructing, start with switching. If pipeline dies before anyone decides anything, start with no-decision. And if you have churned accounts, the firing guide feeds this goal too, because what they hired after you is the clearest read you will get on who you actually lose to.
Increases the frequency, depth, or quality of interaction, creating more value with each use.
Why this one. It is the only guide anchored to what happened between the purchase and the first realized value: the friction, the workaround, the moment the account decided this was or was not worth building a habit around.
Why not product analytics. Analytics tells you where people stopped clicking. It cannot tell you what they were trying to do when they stopped, which is the only part you can design against.
Why not wait for churn. This is the same question as retention, asked early enough to still act on the answer. By the time these accounts are churn interviews, the decision has already gone.
Increases the likelihood that customers continue using the product by raising realized value or switching costs.
When the decision was made. The guide works backward from the cancellation to reconstruct the moment the decision actually happened, which is the only moment you could have influenced. Usage data leads the cancellation but lags the decision, often by months.
What they hired instead. Nobody stops having the job. They gave it to someone else, and the same interview reconstructs that side: what the replacement had to do that you weren't doing, who else was considered, and what the switch cost them. A customer leaving is a progress move, not a rejection, and the hire is where the progress is.
Why this beats win/loss for competitive read. Win/loss studies deals you were in. This studies the deal you lost after you had already won it, from someone who used you for a year and can compare in specifics rather than in demo impressions.
Why not an exit survey. An exit survey samples the moment of leaving. The interview samples the moment of deciding, and then follows it forward into the hire.
Switching and firing are the same interview, mirrored. A switch is always a fire and a hire at once. Same structure, same forces, opposite side of the transaction.
Which is why churn evidence does not only serve retention. The hire side tells you what you lost to and what the winner had to be, so it feeds Goal 01 as directly as it feeds Goal 03.
Enables additional growth from existing relationships through more users, use cases, products, markets, or revenue.
Why not reuse the switching guide. A second purchase runs on different forces from the first. The anxieties are about scope, budget ownership, and the internal case the champion has to make, not about whether you are a real company. A first-purchase guide asks questions this buyer answered a year ago.
Why not upsell analytics. You learn who expanded. You do not learn what made deeper investment feel obvious rather than risky, which is the part you can reproduce.
Where it loops. The accounts that expand are usually the same accounts that refer and vouch. This goal feeds Goal 01, which is why expansion evidence is worth more than its revenue line suggests.
The four above are things a team designs for. Core jobs is the progress customers are trying to make, functional, emotional, and social. Not a moment on the timeline. It defines the category the other four operate in, which is why it does not appear as a fifth card.
It also catches the failure the other four cannot. Wrong-time and wrong-sample problems are specific to a goal. Wrong language is everywhere, because the words came from inside the building regardless of which goal you are working on. If your messaging tests well internally and converts badly, this is the instrument, not a positioning workshop.
Core jobs is a diagnostic baseline, not a default. If you have a clear symptom, start there. If the team disagrees about what customers are actually trying to accomplish, or nobody can say which symptom matters most, that disagreement is itself the symptom and core jobs is what resolves it.
It belongs in the same family as the four goals, and teams ask about it in the same breath, so it is worth saying plainly why there is no interview for it.
Creates structural advantages that become increasingly difficult for competitors to replicate or overcome. There is no interview for this one, because it is not a question you can ask a buyer. It is downstream of the other four: what they produce when they compound.
The instruments are different in kind. Where the four goals above are answered by asking customers, this one is answered by examining the structure of the business itself, which is what the loop and moats diagnostics do.
If you were handed a methodology rather than a symptom, this is where it sits. These are different traditions answering different questions, and conflating them is why a lot of founder-run research produces a confident answer to a question nobody asked.
Does a market exist, is there a buyer, and is the problem real enough that someone will pay to solve it. Pre-product-market-fit by design. The population is prospects and non-buyers.
Stops short at: everything after someone becomes a customer. It has nothing to say about why they stop using the thing they bought.
What to build next, and why the last thing did not land. Continuous interviewing with existing users, opportunity solution trees, assumption testing. Assumes you already have customers to talk to weekly.
Stops short at: the people who never bought. A continuous cadence with current users cannot see the market that walked past you.
How to ask without leading. Talk about what already happened rather than what someone would hypothetically do, because past behaviour is reportable and preference is not. The best single correction to founder-run interviews.
Stops short at: who to talk to and which question to be answering. It is a technique for asking well, not a study design.
Structured debriefs after a deal closes, and a form after a cancellation. Cheap, repeatable, and already in your process, which is why they get treated as sufficient.
Stops short at: the largest population in both cases. Win/loss only reaches buyers who crossed the decision threshold. Exit surveys sample the moment of leaving, not the moment of deciding.
How these map onto the five goals. Customer discovery covers Goal 01 and stops. Product discovery covers Goals 02 and 04 and assumes 01 is settled. Retention, Goal 03, sits across both and is usually owned by neither, which is why churn evidence tends to be the thinnest evidence in the building.
None of the four traditions above is wrong. Each one was built for a specific question and then generalised past it.
The most common question is how many interviews are enough, and it is answerable, just not first. Patterns inside one well-defined role stabilise faster than most founders expect. A study spread thin across four buyer roles will not stabilise at any volume, because that is four small studies rather than one large one. If you cannot yet say which role you are studying, no sample size rescues it.
Define the population and the number follows from it. So does the modality.
Behavioural archetypes modelled on your buying committee, each carrying its own personality profile so objections differ by role the way they do in a real room.
Synthetic research produces population-level signal from behavioural archetypes, and it does not produce evidence about your specific customers. Its job is finding where demand probably lives and building the screener and interview guide you use afterward.
Evidence grade inferred, and labelled that way on every page. The right instrument for directional decisions and the wrong one for final validation or compliance-sensitive sign-off.
Real participants at lower cost per conversation, which is what makes hard-to-reach populations practical at all. Non-buyers and churned accounts rarely give you an hour, but they will often give you fifteen minutes on their own schedule.
Breadth across a population. The most complex committee dynamics still reward a human interviewer.
The deepest read on a small number of people. Senior buyers, complex committees, and the conversations where the unscripted follow-up matters more than the guide.
Cost and calendar per conversation. Rarely the way to cover a population you need breadth on.
Three modalities, three different jobs. Synthetic research does not scale into the other two and is not a cheaper version of them, because it answers a different question: it builds the argument and the instrument, and the other two take that instrument to real participants. What decides between them is the evidence the decision requires and who you can reach, not price.
| Goal | Who you need | How hard to reach | Modalities that fit |
|---|---|---|---|
| 01 Acquisition | Prospects who never decided, and recent switchers | Hardest | Synthetic to build the instrument, then AI-moderated |
| 02 Engagement | Recent customers still inside first value | Easiest | AI-moderated or 1:1 |
| 03 Retention | Churned accounts, ideally within 90 days | Hard | AI-moderated, 1:1 for key logos |
| 04 Expansion | Accounts that grew, and comparable ones that didn't | Easy | 1:1 or AI-moderated |
| Core jobs | All roles on the committee | Varies | Any, but run across every role |
What a discovery call actually covers. Which goal is the binding constraint, who you can realistically reach, and which of the three modalities fits that population. Usually fifteen minutes is enough to work out.
"You don't need research yet" is a legitimate outcome and I say it fairly often. If the decision is already made, or the evidence would not change it, the honest answer is to save the money.
All six guides are published in full, free to use and cite. Reaching the right people, including the non-buyers and the churned accounts, and coding what they say into something you can act on, is where most self-run studies stall.
Run the interviews yourself. Switching interviews, quantified JTBD, and the behavioral models underneath them.
Choose the instrumentSynthetic, AI-moderated, and 1:1 expert interviews side by side, with what evidence grade each one produces.
Before you change courseFive questions to settle before you turn any dial, starting with whether anyone is actually struggling.