How to Run Customer Interviews with AI

Building a machine to make your life more uncomfortable

Author

Brian Scordato
September 03, 2025

Idea to Startup: How to Run Customer Interviews with AI

Sponsor: The Let’s Actually Run Customer Interviews Workshop (by Tacklebox) - Starts 9/15

A 10-day program that’ll help you actually, finally, run proper customer interviews (whether you like it or not). This is a momentum machine - it’ll kickstart your business and make it all start to feel real.

This Episode

Today, we build a machine to help you actually run customer interviews. We’ll use AI to tackle the big blockers—accountability, CRM setup, outreach, transcription, and even how to pick your first customer. 

There’s a lot in the episode, so this might be helpful.

Here’s the AI Interview Machine I lay out on the pod. Feel free to build your own version:

1. Accountability

Interviews don’t happen unless you create structure. AI helps by:

  • Calendar blocking: Ask your AI to generate a 30-day plan for 25 interviews (e.g., 15 parents, 5 coaches, 5 consultants). Have it drop tasks straight into your calendar.

  • External commitments: Have AI draft an email to 10 people you admire, promising you’ll finish the 30-day sprint. Then draft the follow-up email you’ll send when you succeed.

  • Daily nudges: Use Zapier or similar to send a morning reminder: “Interviews are the priority. They’ll be uncomfortable. That’s fine.”

2. Logistics

This is where most people give up. AI makes it smoother:

  • CRM: Have your AI spin up a lightweight Notion (or Airtable) CRM template to track outreach and tags.

  • Finding people: Ask it for 50 ways to reach your target audience (booster clubs, Facebook groups, sidelines, LinkedIn, conferences). Use the list to test channels.

  • Outreach: Prompt it to draft 3 outreach templates in your voice. They won’t be perfect, but they’ll save time. Layer in a Calendly link, follow-up sequences, and you’re set.

3. Synthesis

The real gold is in making sense of 25 hours of conversations. AI is perfect here:

  • Recording: Use Zoom + a tool like Granola for transcripts and bullet summaries.

  • Debriefing: After each interview, let AI “interview you about the interview” to capture details while fresh.

  • Database: Upload transcripts + notes into one searchable hub. Then query: What problems came up most often? Who’s spending real money? Where are the patterns?

Pod References

Transcript - feel free to read like a long-form article

Today, we’ve got a pod on running customer interviews using AI.

We almost certainly aren’t going to use AI in the way you’d think we would. But we’ll get there.

And we’ll start by talking about humans.

Humans are good at lots of things, but in my experience the thing we’re the absolute best at is avoiding criticism. We get even better at avoiding criticism when we’re insecure about the thing that might be criticized, which is usually when we most need to be smacked in the face with it.

While this is human, it’s definitely learned. I read that something like 2% of adults score as “creative geniuses,” but 98% of kids do. The fear of criticism eventually beats the interesting-ness out of us. My son is wildly creative and has zero fear of me saying the canoe he made out of our couch cushions isn’t actually floating down the Amazon river.

Anyway, most founders I meet are insecure about their idea because that idea has gotten so twisted and tangled up with their identity and self-worth that you can’t tell where one ends and the other begins. They can’t stomach potential criticism of it. Usually, the longer you’ve had an idea and the more fully formed it is in your head, the tighter it’ll be aligned with your identity and the harder you’ll push against potential criticism. So, watch out for that.

The first things we tell our founders at Tacklebox is to treat your idea like a patient on an operating table. We’re the doctors huddled around the idea, trying to save it - we aren’t the ones lying on the table. The separation will keep you sane.

Over the past 18 or so months the question I’ve gotten the most, by far, is “can I use AI for customer interviews?” Meaning, can I just interview ChatGPT as if it was the owner of a bike shop or an elementary school principle or whoever their customers are to avoid actually speaking with customers.

Founders suggest this will speed up the process, but it’s just a veiled attempt at avoiding criticism.

It’s way easier mentally to just hire a dev shop or start vibe coding and get going on product. It kicks the criticism can down the road. And it can feel more like progress than talking to customers does.

If you’ve ever met someone who spent two years working on a product, quitting their job and draining their savings in the process, only to find out that no one has the problem they build a product to solve - they aren’t dumb or even bad entrepreneurs for not validating that first. They’re just good humans who have mastered our sacred art of avoiding criticism.

So, no, we aren’t going to just run customer interviews on ChatGPT and call it a day, because, as we’ll see, that’s no better than just not running customer interviews and honestly it’s probably worse.

What we will do is use AI to make the interview process 10 times faster and more effective. But we won’t use AI in the way most people are using it - to avoid discomfort. We’ll use it to ramp up that discomfort. We’re gonna build a machine that’ll deliver criticism to you constantly. Which might sound awful, but criticism is an acquired taste. Soon you’ll get used to it and then you’ll seek it out, because you’ll realize the criticism and feedback are fertile ground for the insights you need for a successful business.

So, while everyone else uses AI to make their lives easier, we’ll use it to make ours uncomfortable. In the short term.

But before we get to our AI interview machine, we need to quickly talk about what the point of customer interviews is in the first place. The job we’re hiring them to do.

Which starts with the Worth Your Time to Solve Equation.

—

The Worth Your Time To Solve Equation

The early days of a startup aren’t about product or team or funding - they’re about finding a customer with a problem that’s worth your time to solve.

Digging into that sentence deeper, “worth your time to solve” has two components:

First is the OVERPAY component - the problem you find is so important to this customer that they’ll happily overpay you to solve it. Because with out you, their options are horrible. And simply not solving the problem isn’t an option. They’re at the bottom of the well and they need any sort of rope.

Continuing this analogy, the rope itself isn’t the hard part. It’s noticing the customer in the well.

The Overpay component is critical, because if your first customer won’t or can’t overpay for your first product, your business probably won’t work.

The second component is what we call ASYMMETTRIC KNOWLEDGE.

This customer that’s got this wildly important problem - you need to know more about them and their situation and their process and what success looks like than any competitor.

You need asymmetric information.

If this combination seems like a tight needle to thread - a customer with a problem they’ll overpay to have solved and a customer you know more about than anyone else - it is. That’s why startups are difficult.

But the good news is, you almost certainly won’t start with all of this information and you don’t need to.

This is the job of your customer interviews. It’s what we’re hiring them to do. To help you root around and find customers that might be good potential first customer candidates.

They’re the first tool we’ve got to learn about the problem and the process.

And, honestly, if this sounds good so far and the rest of the pod does as well, you should just join the customer interview workshop we’re running that starts Sept 15th. Link in the show notes and at gettacklebox dot com. Sorry for the random ad but man, writing this it’s just so clear - if you’re prioritizing interviews, just… come to that workshop. We’ll get you set up.

Anyway, as I was saying, interviews are the first tool we’ve got to learn about problem and process. We speak with lots of different types of customers, we learn exactly how they interact with the problem now, and then we try to surface edge cases. Small, cohesive customer segments - what we call SOMs at Tacklebox - that are disproportionately impacted by the problem or have unique scarcity and constraints around a solution. Reasons why the status quo is particularly difficult.

Customer interviews find the edge cases where we can get a foothold.

Back to AI, because I can feel you asking this question through your airpods:

Why don’t I just ask AI for the edge cases?

Well, if you type into ChatGPT “pretend you’re a principle at a high school with 2,000 kids and a yearly budget of X and goals of Y and Z and you’re considering software to track attendance” then run an interview, you’re getting responses based on the same information any other person with an internet connection can get. AI, by definition, doesn’t surface edge cases because it doesn’t know edge cases. It knows the most likely path, which is not what we want.

Interviews are about knowledge separation - you pushing yourself from everyone else. That needs to be done in person.

But, there’s an even better reason this won’t work:

The process of customer interviews give us about 50 other invaluable data points because customer interviews are a dry run of this person being your actual customer:

Let’s say you’re trying to interview that Principal. The first thing you’ve got to do is find them. You’ve got to see if you have any connections to get a few Principals on the phone with you. That tests your network. Once you’ve exhausted your network, you need to speak with some cold Principals. So, maybe you send cold emails or linkedin messages. You need compelling copy in those. And, when they inevitably don’t work, you need another path in. Maybe it’s through parents. Maybe it’s showing up in person. Maybe it’s education conferences. In those conversations you learn about the other places they gather, about how much they speak with other Principles, about how they track and measure they’re school’s success, about who they pit themselves against. Decisions are made on envy, not greed, and envy is irrational. It ain’t showing up on chatgpt.

Getting 25 interviews with potential customers is 15% about what they say and 85% about testing sales channels and marketing copy and organic growth potential and finding design partners. All of your first customers will start as interviewees.

The process of getting and running interviews is what we base our first customer decision on.

It’s just not skippable. And, again, we can make the whole thing easier - and more uncomfortable - with AI.

So… let’s just do it. With the help of a fake startup I’ve been thinking about for a while.

After…. a little smooth jazz.

Another plug. If the interview thing is hitting you right, join the workshop. Details, and pop in your email if those dates dont work but you want to attend a different session.

—

Whisper Ideas

The idea we’ll work through today and in the second half of this episode that’ll come out next week is what we call a whisper idea.

The term Whisper Idea is maybe the oldest one we’ve got at Tacklebox. It started back in 2013, after I’d moved on from Find Your Lobster.

I was running 1 on 1 coffee chats with dozens of founders a week trying to get my footing and decide what I’d do next, and I realized there were two types of entrepreneurs:

The first came in and confidently told me their idea. This person had nearly always had the problem they were solving, tried every other solution on the market, and, out of sheer frustration rather than entrepreneurial spirit, decided to solve the thing themselves. Usually grudgingly.

The second came in and whispered. When I asked what they were up to they’d cautiously look left and right, lean in real close, and in the softest voice possible tell me their idea.

The first person knew it was safe to shout their idea from the rooftops because the hard part wasn’t coming up with the idea itself, it was solving it.

The second thought that the only advantage they had over every other person was that they’d thought of the idea first. The idea itself was the valuable thing. So, they whispered.

The rooftop entrepreneur is, obviously, in a much better spot than the whisper idea entrepreneur. So, for the first 5 years of Tacklebox I had a rule - no whisper ideas.

But, it’s become clear that most founders have, at least at some level, whisper ideas. And customer interviews, done properly, are their ticket off that doomed path of thinking the idea or product is what matters. You can start with a whisper idea and build a successful startup. I’ve seen it dozens or hundreds of times.

So, today’s startup example is a whisper idea and that’s on purpose because most people start with whisper ideas and you’ve probably got a whisper idea and that’s just fine. For now. you just can’t stay that way.

The Idea

I’ve been fascinated by the high school sports world for years.

One of my earliest startup ideas was a linkedin for high school athletes that allowed them to help coaches recruit them.

The forces behind the idea were strong. Here were the big four:

  1. Wild success for my customers, high school athletes, was a college scholarship. Worth hundreds of thousands of dollars.

  2. The status jump, aside from the potential scholarship, was a profile where you put up video of yourself. Making it easier for customers to show off is never a bad business idea, and it’s the best growth strategy on earth.

  3. The economic buyer had extremely deep pockets - the high school kid wasn’t paying, but their parent would in hopes of getting their kid the aforementioned scholarship.

  4. This also made a college coach’s job easier.

So, we knew what wild success looked like, we provided a status level jump, the customer had enormous willingness to pay, and there were great marketplace dynamics as a solution would solve problems for both sides.

It didn’t work, but that’s for another pod.

I’ve kept one foot in high school sports tech ever since - from the video cameras that now record every game and create clips for kids to the advanced analytics that now exist for lots of high school sports.

The first Whisper Idea I had for todays pod was… announcing.

Lots of high school games are now livestreamed so that grandparents and aunts and cousins can watch. The student athletes can study their game film and create clips that they push to their college recruiting profile or tik tok. This is great.

But… the games have no announcers.

My thinking was that announcers would make the whole experience more… fun. Maybe people would watch more, engage more, maybe this would open up to advertisers or maybe we’d figure out better production - creating a youtube channel for schools that got more views and then advertising or… something.

But the idea feels so flimsy. The criteria that made the recruiting idea strong - clear wild success, an obvious status level jump, an economic buyer willing to pay - just didn’t exist for the interview idea without a ton of mental gymnastics.

So, I did what I recommend everyone with an idea do. Get on the phone with someone who has more insight into the industry than I did.

I called my friend who happens to be a high school basketball coach and the dad of a star high school basketball player.

I asked him about his last three months.

I asked him to walk me through any problems that came up repeatedly, any urgent requests or questions from parents or faculty or his boss. I asked him how his boss, the AD, decided whether he’d been doing his job well or not. We talked for 45 minutes and a story emerged.

He said that probably 60% of his players, and he said this was consistent across all the teams at the high school, were interested in playing in college. The reality was that far fewer than that actually would, but a huge percentage of his time was spent interacting with parents who wanted their kids to play in college. They asked about events they should attend, how to contact coaches, if this might get them into a better academic school, what AAU teams to try out for. This was as intricate and time consuming a process as applying to college, with the same amount of upside, and… no system for managing it.

He knew of one person a few towns over who had a consulting business that helped parents manage the process, but he only worked with kids looking to play high division one. What about everyone else?

Wouldn’t some clarity around how the heck you get your kid recruited by a division 3 diving team be useful?

We have a hypothesis.

And now, finally, it’s time to bring in AI.

—

The AI Interview Machine

The time to start interviews is the second you’ve got a hypothesis you’re feeling good about.

It’s this sort of inverse relationship entrepreneurs need to get comfortable with - whenever you feel good, your next step should be to try and poke some holes in that thing you’re confident about.

So, I feel excited about this “manage your kids attempt at playing college sports” idea, which means I need to seek some criticism, or, in other words, better visibility.

Today, we’ll go through the machine itself. Next week, we’ll push the idea through it.

The Machine.

There are three parts to traditional interviews -

Top of funnel - this is customer selection, outreach, scheduling and CRM

The actual interviews - this includes the questions and script, recording and note taking, and follow-up

And finally, Decision Making - taking what you’ve learned and using it to help you choose a path forward

Each of these parts have blockers. The big ones are:

  1. Accountability - actually doing the interview process in the first place

  2. Logistics - everything from finding people to interview to reaching out to them to tracking responses to following up

  3. Synthesis - translating 25 interviews that each might’ve lasted an hour into something coherent

Luckily, AI can help with a bunch of this.

I’ll go through how I approach an interview machine build, but, the beauty of AI tools is personalization. You can tweak these ideas to build something that is a reaction to your blockers and capabilities. Wherever you see yourself slipping, figure out how AI can help create momentum. And that’s the way I think of it. AI exists for when I need a boost.

We’ll start where everyone should start: Accountability.

If you’re going to start running interviews, the first thing you’ve got to do is make appropriate space for them - in your calendar and in your brain. I could do my whole we’re all goldfish rant and talk about how you’ve got to remove before you add, but you’ve hopefully heard that already. So we’ll jump right into asking my AI of choice, Claude, for the practical “making space” help.

For me, accountability works best in two forms - blocks on my calendar and promises to people I admire.

So, I started by asking Claude to create a 30-day interview schedule. I told it that I wanted to speak with 25 people - 15 parents, 5 high school coaches, and 5 people who work in college recruiting or sports consulting over that time. I asked it to build out specific daily goals to get there, using the assumption that we’d have a 3% response rate. I made that up, but it’s usually somewhere between 3-10%, depending on the idea and how good you are at outreach.

It has access to my calendar, so I told it to create blocks of time where I had space over the next 30 days, with tasks for each block of time.

Now, these are never perfect - but the calendar holds with specific tasks are pretty magical to me. Particularly with the personal accountability hook, which was my next prompt.

I asked it to draft an email I’d send to 10 people I look up to with details of my 30 day experiment. I also asked it to draft the second email I’d send in 30 days saying I’d been successful. I sent the first and scheduled the second. I’m on the hook.

For good measure, and because I was messing with some of the Zapier tools, I had it create an email to send me every morning at 8am reminding me why interviews were so important, and adding something inspirational.

This isn’t actually all that inspirational, though it is entertaining to see what a robot thinks will get me going in the morning - it’s just a reminder to my subconscious each morning. This is the priority. This is what I said I’d do. Remember?

Ok - accountability set.

Next up - top of funnel outreach.

The first blocker here is information management. You need a super lightweight CRM that shows the people you’ve interviewed, people you’ve reached out to, and has a place for tags.

I use Notion for everything, so I had Caude create a simple CRM template for me in Notion. I went back and forth a few times until it was built right.

The next blocker is finding people. So, I had the robot brainstorm 50 ways to find parents of athletes, high school coaches, and college coaches or sports consultants. i pulled a few ideas from the list I liked - booster clubs, local sports facebook groups, and showing up in person to speak with parents on the sidelines being a few. I had Claude pump these into my calendar.

I then had it draft 3 outreach templates for interviews using, yep, a prompt about myself. I said “reach out in the way brian scordato from tacklebox and idea to startup would reach out.” It was…. decent. There was problem language and some specificity and a decent CTA. It needed a bunch of tweaking, but it gave me a draft. I used the cleaned up version with some individual customization for most of the early, high potential targets using our Tacklebox content. I can maybe go deeper on this in another pod if it’s interesting.

I also created a calendly link for scheduling ease, and a set of follow-up emails and an easy unsubscribe button.

AI has no clue about edge cases, but it’s incredible at generating comprehensive lists of obvious things and ideas we all would forget.

—

Next is the actual interview execution.

AI isn’t super helpful for the interview questions - we’ve got a bunch of content on that. And, maybe I’ll get into it deeper in a follow-on episode if interested.

BUT, there are a few AI tools that are critical here, and they depend on whether the interview is in person or not.

If the conversation is remote, I highly recommend Zoom over a phone call if possible. And during that call, use something like Granola.ai to record everything. This will allow you to be totally present. Granola will give you a transcript of the call, along with bullets. We’ll use this again later.

If you’re in person and can’t record or take notes, immediately after jump into your AI of choice and turn on voice mode. This is one of my favorite AI use cases.

Tell it that you just had an interview, and ask it to prompt you with a bunch of questions about that interview. Basically, for it to interview you about the interview. This is the best way I’ve found to get your thoughts out and recorded in an accurate way.

I started doing these a year or so ago, and I liked this process so much that I even do it after interviews that are recorded. Using AI as a tool to ask you questions about the interview you just ran and recording your responses becomes a great data point.

Which leads us to AI’s last and, by far, most useful role: synthesis

After each interview, upload the transcripts and your thoughts to the AI you’re using. This will create a massive, searchable, queryable database.

When we started, I said that interviewing chatgpt was a waste of time because there was no specific knowledge.

Well, now you’re creating a database of specific knowledge you can reference.

You can ask about patterns in your interviews, about the problems that came up most often, about the customer that had the biggest problem. You can ask about what you still need to learn, or questions you should go back and ask specific interviewees.

You will know things like 85% of your interviewees mentioned one concern or that 15% of your customers said they’d spent more than $1500 on a solution. When you add information on each interviewee, you might learn that parents with three kids were more likely to try X, or parents who’d played college sports were more likely to do Y. It’s a customer segmentation cheat code.

I give AI a lot of shit, but this… this is useful.

—

The End - criticism

This went on far longer than I’d thought, so I’ll do a dedicated episode on the actual interviews I ran with parents, coaches, and sports consultants and how I built and used the machine. Hopefully this overview gets you started.

And, again, if you made it this far… you should probably just sign up for the interview course. I hate to plug something so much, but it’s just a great way to get all this done.

I started with humans hating criticism and I’l end with it.

We’ll all bend over backwards to avoid the discomfort of someone telling us our idea is bad. Or, even, just telling us it’s not amazing. But critical feedback is the only way you’ll actually build something useful. You need it early and often and interviews, with a boost from AI, are the way to do it.

Give it a shot.

And now it’s time for me to get back to the couch. There’s a canoe waiting to float me down the Amazon.

This was the idea to startup podcast. Have a great week.