Customer interviews are short conversations with real customers about why they bought, and this is a Skill job that runs across a few sessions. One sitting sets the goals and questions, and you can finish it and send your first interview requests today; the calls happen over a week or two at about twenty minutes each, and one more sitting finds the patterns. The AI runs each step on paper: it turns what you want to learn into goals, writes the questions, files each conversation into a short debrief, and reads the debriefs together for patterns. You do the talking. At the end you have a page of what customers actually said, in their own words, and a short list of what to change because of it.
- Shape
- Skill
- Sitting
- a few sessions
The shape: Skill
This job is a Skill because the method is the hard part, and the method is learnable. Most owners already talk to customers every day. What they rarely do is decide in advance what they are trying to learn, write down their guesses so the calls can prove them wrong, and file each conversation the same way so the fifth call can be compared with the first.
The AI is good at exactly that structure. It can turn a vague worry such as "why don't more people book" into goals you can actually answer. It can write open questions that ask about the past instead of the future, which is the rule that most often separates a useful interview from a polite one. And it can read six debriefs side by side and notice that four customers used the same phrase without being asked.
What it cannot do is sit on the call. No tool needs setting up, no account needs connecting, and no software does the collecting. That is what makes this a Skill and not a Tech job: everything here is a brief you paste and a conversation you have. The three briefs below are yours to keep and reuse, and the second round of interviews goes faster than the first.
What you give the AI
Gather these before the first sitting. None of it needs to be polished.
- The decision you are facing. One sentence, such as "whether to add a weekend class" or "why people who request a quote go quiet." Interviews without a decision behind them produce good stories and no change.
- Your current guesses. Three to six things you believe about why customers buy, what almost stops them, and what they compare you with. Write them as statements you could be wrong about.
- Who you can reach. A short list of the customers you could ask, with a note on each: recent, long-time, or nearly bought. Names stay off the page you paste.
- What you sell, in a paragraph. What you do, who it is for, and roughly what it costs, so the questions fit your business.
- What will not change. Anything off the table, such as your location or your hours, so the AI does not spend questions on it.
- After each call: the transcript or your notes. A transcript from the video tool, or the notes you typed right after the call, with the customer's name replaced by a label.
What comes back
Four things come back, one per step, and each one is a plain document you keep.
1. The goals. A numbered list of what you are trying to learn, such as G1 "what makes a customer decide to book this week instead of next month." Each goal ties back to your decision.
2. Your guesses as hypotheses and the questions. Your guesses rewritten as numbered statements a conversation could disprove, then ten to twelve open questions, each tagged with the guess it tests. Good questions ask about something that already happened: "Tell me about the day you decided to look for someone."
3. A debrief per call. One short reply per conversation, which you copy into its own document, filed against the question numbers: what they said, their key phrases kept word for word, and the questions you did not get to marked as skipped. Anything they said that fits no question goes in a section at the end.
4. The patterns. After five or more debriefs, a read-out of what held up, what was disproved, what came up that you never asked about, and whether to keep interviewing.
Paste this brief at the start of the first sitting.
Act as a customer-interview coach for a small business.
Work through one step at a time and wait for my answer before the next.
The business: [one paragraph on what you sell and who buys it]
The decision these interviews serve: [one sentence]
My current guesses: [three to six statements]
Not changing: [anything off the table]
Who we can interview: [recent buyers, long-time customers, people who nearly bought]
Step 1. Turn the decision into three to five numbered goals (G1, G2...).
Step 2. Rewrite each guess as a numbered hypothesis (H1, H2...) that a conversation could prove wrong, and tag the goal it serves.
Step 3. Write ten to twelve open interview questions (Q1, Q2...), each tagged with the hypothesis it tests.
Every question asks about something that already happened.
No question asks whether they would buy, use or pay for something in future.
No question names the answer we hope to hear.
Put the three most important questions first, in case the call runs short.Paste this after each call, in the same chat so the AI keeps the question list.
Here is the transcript of one interview, with the customer labeled as [Customer A, a recent buyer].
File it as a debrief against the Q numbers above.
One short answer per question actually asked.
Keep the customer's key phrases word for word, in quotes, only if they appear in the transcript.
Mark questions we did not ask as skipped.
Put anything that fits no question in a final section called Addenda.
Do not summarize what they meant.
Report what they said.When you have five or more debriefs, ask for the patterns.
Read all the debriefs above against the hypotheses.
For each hypothesis, say: held up, disproved, or not enough evidence, and cite which customers.
List anything said by two or more customers that no question asked about.
List phrases two or more customers used in nearly the same words.
Report counts, such as four of six, never percentages.
End with one recommendation: keep interviewing, stop, or change who we talk to, and why.How to judge it
A bad result is easy to spot if you know where to look. Reject it, or send it back, when you see any of these.
- Questions about the future. "Would you use an online booking page?" invites a kind yes that predicts nothing. Ask the AI to rewrite it as a question about the last time they booked.
- Leading questions. "How much did you love the follow-up call?" has the answer built in. A fair question could be answered either way.
- Too many questions. More than twelve means you will rush the call and skip the follow-ups, which is where the useful answers are.
- A debrief that sounds smoother than the customer. If the debrief reads like marketing copy, the AI paraphrased. Real customers speak in fragments, and the fragments are the point.
- A quote you cannot find. Search the transcript for a few words of each quoted phrase. If it is not there word for word, the AI wrote it, and the whole debrief needs a second look.
- Patterns from one customer. One vivid story is not a pattern. The read-out should say how many customers said something, and one of six is a note to watch, not a finding.
- Every guess confirmed. If the patterns agree with everything you believed going in, read two transcripts again yourself before acting on it.
What stays with you
The AI handles the paperwork. Four things stay with you, and they decide whether the interviews were worth doing.
The asking is yours. Customers say yes to a call because they know you. Send the request from your own name, keep it short, say it is about twenty minutes, and say plainly that you are not selling anything.
The listening is yours. The best answer in most interviews comes after the prepared question, when you say "tell me more about that" or simply wait. No list of questions can tell you when to stop and follow a thread. Talk less than the customer, and do not explain or defend your business when they describe a problem.
The privacy is yours. Ask before recording, strip names and identifying details before pasting anything, and keep the transcripts somewhere private. The AI does not know what your customer expected when they agreed to talk.
The decision is yours. A read-out tells you what a handful of people said. It cannot tell you whether a change is affordable, whether your team can deliver it, or whether these customers speak for the rest. That judgment is the reason you ran the interviews. When you change something because of what you heard, tell the people who talked to you.
An editorial example
This is an invented business, used only to show the method.
Say you run a small pottery studio that teaches six-week evening courses. The decision: why people ask about the courses and then do not sign up. Your guess: the price is the problem.
The goals come back as three: what made recent students sign up when they did, what nearly stopped them, and what people compared the course with. Your price guess becomes a hypothesis a conversation could disprove: people who did not sign up said the cost was the reason.
You interview four recent students and two people who asked and never booked, each on a twenty-minute video call. After each call you paste the transcript and get a one-page debrief back.
The patterns read-out says the price hypothesis is not supported: neither of the people who did not book mentioned cost. Both said six weeks of fixed Tuesday evenings felt like too big a promise. Three of the four students, without being asked, said they first came to a one-night taster before committing. You check each of those phrases against the transcripts and find them word for word.
The change it points to is not a discount. It is putting the one-night taster on the course page, so people try one evening before anyone asks them to commit to six weeks. That is the job.
Interview or survey
Both are research, and they answer different questions. Choose by what you do not know yet.
| If you need to know | Use | Why |
|---|---|---|
| Why people buy, in their own words | Interviews | Follow-up questions find reasons you could not have listed |
| How common a reason is across all customers | A survey | Many short answers can be counted |
| What to ask in a survey | Interviews first | You learn the choices customers actually think in |
| Whether a fix worked | A short survey | The same question, asked again, shows a change |
If the patterns from your interviews raise a "how many" question, our page on surveys with AI covers the next step. The words customers used on the calls are the raw material for your value proposition, and the customer pain points generator gives you a starting list of problems to test before you write your guesses. When this one is done, the full list of DIY marketing jobs has the next one.
Of every job on this list, social media is the one that comes back every week, and Boomp is software that does that one for you. See ten posts made from your website, free
