Every founder has a version of this story. Months of building, a proud launch, and then silence. Not hatred. Not even criticism. Just silence.
What the data says
CB Insights (opens in a new tab) has studied why venture-backed startups shut down for over a decade. Their latest analysis (opens in a new tab), covering companies that closed since 2023, lists poor product-market fit as a factor in 43% of failures. Running out of money tops the list at 70%, but the report points out that cash is usually the final cause, not the root one. Companies rarely run out of money while customers are pulling the product out of their hands.
Get out of the building, faster
Steve Blank (opens in a new tab) saw this pattern early. In Why the Lean Start-Up Changes Everything (opens in a new tab) for Harvard Business Review (opens in a new tab), he described how traditional startups write a plan and build, "all without getting much feedback from the people they intended to sell to." His alternative: "Lean start-ups, in contrast, begin by searching for a business model." Or, as he puts it on his blog (opens in a new tab): "there are no facts inside the building, so get outside."
Eric Ries (opens in a new tab) turned that into a loop: Build, Measure, Learn (opens in a new tab), and organize everything to "accelerate that feedback loop." And Rob Fitzpatrick (opens in a new tab) wrote the field guide every founder should read, The Mom Test (opens in a new tab): "How to talk to customers & learn if your business is a good idea when everyone is lying to you."
The trouble with talking to customers
Here is the uncomfortable bit. Good customer conversations take time and money. A research firm puts the typical cost of a focus group at "between $4,000 and $12,000" (opens in a new tab), and planning guides cited by Koji (opens in a new tab) note that to seat 10 people you may need to recruit 17. Timelines are measured in weeks. So most teams talk to five friendly people, hear what they want to hear, and build.
Where Cast fits
Cast does not replace getting out of the building. It makes the first lap so cheap that you can run it every week. Describe your customers in the morning, interview twelve of them at once, and by lunch you know which of your three ideas has a pulse, which audience feels the pain most, and what the first objection will be. Then you go talk to real people about the one idea that survived, with better questions and fewer illusions.
“Accelerate that feedback loop.”
The cost of being wrong slowly
Building the wrong thing is expensive in money, but it is more expensive in time. Every month spent on a feature nobody needed is a month a competitor spent learning. This is the core insight of the Lean Startup (opens in a new tab) movement: the unit of progress is validated learning, and the fastest learner wins.
A minimum viable product (opens in a new tab) was meant to be the cheapest way to test a hypothesis. In practice, many teams treat the MVP as a small version of the final product and spend months on it. Often the cheapest test is not a product at all. It is a conversation.
Ask about jobs, not features
Clayton Christensen and colleagues argued in Know Your Customers' "Jobs to Be Done" (opens in a new tab) that people "hire" products to make progress in a specific situation. If you understand the job, the features follow. If you only ask about features, you collect a wish list.
This is where early interviews pay off most. A good customer conversation is about the last time someone tried to solve the problem, what they used, what it cost and what annoyed them. That is precisely the kind of question AI users answer well, because it is grounded in circumstances rather than opinions about the future.
A weekly learning loop with Cast
- Monday: write down the three riskiest assumptions in your plan.
- Monday afternoon: run a Cast study with two or three audiences built around those assumptions.
- Tuesday: kill, change or keep each assumption, and write the questions you still cannot answer.
- Rest of the week: take those sharper questions to five real customers.
- Repeat. The real conversations get better every week because the obvious questions are already answered.
The customer discovery checklist
Steve Blank called this phase customer development (opens in a new tab): finding out who your customer is and what they need before you scale anything. Here is a practical checklist you can use with or without Cast.
- Write your riskiest assumption as a sentence that could be false, such as "clinic managers lose at least an hour a day to rescheduling."
- Name the customer precisely: role, company size, place and the moment the problem shows up.
- List what they use today, including spreadsheets, a person, or doing nothing.
- Write questions about past behavior ("tell me about the last time"), not hypotheticals ("would you use").
- Decide in advance what answer would make you drop or change the idea.
- Run the interviews, then sort what you heard into confirmed, contradicted and still unknown.
- Pick the single next assumption to test, and repeat.
The questions in step four owe a lot to Rob Fitzpatrick's Mom Test, linked above. The idea in step five is the one founders skip most often. If you do not decide beforehand what would change your mind, every result will look like support.
A glossary for founders in a hurry
Riskiest assumption. The belief that, if wrong, sinks the plan. Test it first, even if it is uncomfortable.
Validated learning. Eric Ries's term for progress measured by what you have proven about customers, not by what you have shipped.
Problem-solution fit. Evidence that a real group has the problem and that your approach could solve it. It comes before product-market fit.
Product-market fit. In Marc Andreessen's essay The only thing that matters (opens in a new tab): "Product/market fit means being in a good market with a product that can satisfy that market." One popular check is Sean Ellis's survey, which asks users how they would feel if they could no longer use the product, with 40% "very disappointed" (opens in a new tab) used as a rough benchmark.
Continuous discovery. Teresa Torres defines it as "weekly touch points with customers by the team building the product" (opens in a new tab), with small research activities in pursuit of a desired outcome.
A worked example: two founders, one idea
Imagine two founders with the same idea: software that helps small bakeries predict how much bread to bake each morning.
The first founder builds for four months. At launch, owners like the forecasts but will not type in their sales each evening. The product works. The habit it needs does not exist.
The second founder spends week one on a Cast study with three audiences: owner-bakers, bakery managers in small chains, and café owners who buy in bread. A pattern appears quickly in this hypothetical study: the owner-bakers are at the oven before dawn and have no time for data entry, while managers already keep sales in a point-of-sale system. In week two she talks to eight real bakery managers about their tills. By week four she is testing a version that reads till data automatically, with managers as the first customer.
Nothing in this story needed genius. It needed the uncomfortable question asked early.
Where speed can mislead you
Fast learning has its own traps. Synthetic interviews cannot show urgency, the thing that makes someone pay this month rather than one day. They cannot reveal relationships, such as a supplier who would be offended if a bakery switched. And quick studies tempt you to run so many that you never commit. Paul Graham's advice in Do Things that Don't Scale (opens in a new tab) is a good counterweight: "The most common unscalable thing founders have to do at the start is to recruit users manually." Use Cast to decide who to call. Then call them.
Questions about learning speed
Is Cast a replacement for talking to customers?
No. It is a way to make your real customer conversations count. You arrive with better questions and fewer blind spots.
How many iterations should I run before building?
Until the riskiest assumption has survived both synthetic and real interviews. For many ideas that takes two or three weeks, not two or three quarters.
What should a customer discovery checklist include?
A falsifiable riskiest assumption, a precise customer, their current alternative, questions about past behavior, a kill criterion decided in advance, and a clear next assumption to test.
How do I know when I have product-market fit?
There is no single test. Strong signs include customers pulling the product from you, retention that holds, and a high share of users who would be very disappointed to lose it. Synthetic interviews cannot measure any of these, because they need real usage.
Read next: where synthetic users go wrong, and how we design around it.
