Good Words Trapped in Chat Logs: Customer Case Studies for Your Vibe Project
Out of 500 paying users, only 96 scored 9 or above, and only 40 of them agreed to talk — 9 published case studies in the end. Pick the right people, run a 20-minute interview, write it in six sections, put it where buyers hesitate. The walk-through project, people, and numbers in this guide are fictional.

Good words stuck in chat logs are the same as no good words at all
A Nan built a small inventory and sales tool for tiny shop owners called Zhanggui Bao, twelve dollars a month. A year and a half after launch it had 500 paying users, renewals looked fine, and the support chat logs were full of praise: "reconciliation is so much faster now," "we dropped a part-timer this month," "finally no more digging through Excel." But when he opened his own website, the case study section was still empty — the homepage had three lines he wrote himself, and there was nothing next to the pricing page. New visitors came in, saw no real person actually using the thing, and nobody answered the question in their head: "does this actually work?" It was not that A Nan never thought about case studies. Every time he thought about it he got stuck: who do I talk to? What do I ask? What does it look like when it is written up? Won't asking for a real name and a photo feel pushy? He put it off for three months and never wrote a word.
Spread the numbers out and it is not that hard. Out of Zhanggui Bao's 500 paying users, 96 scored 9 or above — promoters. A Nan sent each of them an invitation, and 40 agreed to talk. He ran 24 interviews, wrote 14 drafts, and published 9 of them. In other words, out of every 100 paying users, fewer than 2 ended up as a published case study (9 / 500 = 1.8%). But look only at the 96 promoters and about 1 in 10 made it onto the website (9 / 96 = 9.4%). Case studies do not fall from the sky. They get sieved out of 500 people one layer at a time. The sieve itself is not scary. What is scary is not knowing what the sieve looks like, or how many people each step should drop.
One thing up front: the author, the product, and every number in this guide are fictional examples. The method itself is a real, usable approach — treat the numbers as worked examples for doing the math, not as industry data to quote.
First, tell them apart: a case study, a testimonial, and a logo wall are not the same thing
Before you do anything, sort out three kinds of social proof. They are not the same thing, so do not mix them up. The first is a customer case study: a short piece that says "this person used to struggle, and now things are better," with a real name, a company, and before-and-after numbers, about one page long. The second is a one-line testimonial: a user writes "works great, recommended" off the cuff, and you rotate it on the homepage. The third is a logo wall: a row of company logos proving "these companies use it." The costs are very different. A testimonial costs almost nothing — screenshot it and post it. A logo wall means asking each company for permission one by one, tedious but no writing involved. A case study is the most expensive: one piece takes three to four hours from invitation to publication. Expensive for a reason: a testimonial only proves "someone likes it," a logo wall only proves "someone uses it," and only a case study answers the question a new visitor actually has — "someone like me used it, so what happened?"
Put the three side by side and the results differ a lot too. Zhanggui Bao ran a one-month comparison on its website. Per 100 views: the long case study was noticed only 38 times, but 9 of those views clicked "Start free trial." The one-line testimonial was noticed 55 times and got 4 trial clicks. The logo wall was noticed 63 times and got only 2 trial clicks. The logo wall is the most visible, but people glance at it and move on — nobody pays because of a row of logos. The case study is the least noticed, but the people who open it are genuinely hesitating, and the line "reconciliation used to take 6 hours a week, now it takes 40 minutes" lands exactly where it hurts. So do not stack a logo wall on the homepage and call the job done: the logo wall is for people who already believe you, the case study is for people who are still deciding. When budget and time are both tight, do the case study first, add testimonials next, and leave the logo wall for last.
The case study funnel: 500 paying users, 9 stories published
Put the funnel on the table first, then decide which step to fix. Zhanggui Bao's numbers over three months: 500 paying users; 96 promoters who scored 9 or above — this step drops 404 people who were never that happy in the first place, do not mourn them; 40 agreed to talk, an agreement rate of about 42% (40 / 96 = 41.7%), losing 56 people here; 24 interviews actually happened, a completion rate of exactly 60% (24 / 40 = 60.0%), losing 16 people who booked a time and then had something come up; 14 drafts written, a draft rate of about 58% (14 / 24 = 58.3%), losing 10 pieces where the conversation turned out too flat and had no numbers to work with; and finally 9 published, a publication rate of about 64% (9 / 14 = 64.3%), losing 5 where the person did not say yes to going public in the end. Add up the losses at all six steps: 404 + 56 + 16 + 10 + 5 = 491, exactly 500 − 9. The books balance.

The biggest hole in the funnel is not the last step, it is the second: 56 promoters never agreed to talk. Write the invitation wrong and good people drain away for nothing. A Nan's first invitation read like mass mail: "Hello, we are collecting user stories and would love your participation." Fewer than one in five replied. After he rewrote it into three sentences, the agreement rate climbed past four in ten: "Sister Wang, this is A Nan from Zhanggui Bao. I saw you renewed twice and would love 20 minutes to hear how you do reconciliation with it. I will send you the draft first — nothing goes live until you say yes." The first sentence says "I remember who you are," the second says "only 20 minutes," the third says "you have the final say before anything is published." Drop any one of the three and the agreement rate falls. The other way round, the 64% publication rate at the last step is actually not bad: of the people who got a draft, more than six in ten were willing to go public, which means that if the conversation went well and the draft is honest, going public is not the hard part. Fix the funnel in this order: first lift the invitation agreement rate from 42%, then the draft rate, and only then worry about publication. Plenty of people start by fussing over how pretty the case study page looks, which is pushing at the narrowest part of the funnel while the 56 people who never agreed stay lost.
How to pick people: not the ones who love you most, the ones who look like your next customers
Picking people has three filters, and if any one is missing, do not start. The first filter is "got results": the person needs before-and-after numbers to tell — reconciliation used to take 6 hours a week and now takes 40 minutes, say, or a part-timer used to cost 900 dollars a month and now is not needed. A case study with no numbers reads exactly like a one-line testimonial, and you just burned three or four hours for nothing. The second filter is "willing to go public": a real name, a shop name, a photo — at least two of the three. Someone who will only stay anonymous is not unwritable, just unbelievable — "a clothing shop owner says it works great" could have been written by anyone. The third filter is the one people forget: "looks like the next customer." Whoever you most want to sell to next, find people who already look like them. If Zhanggui Bao wants to sell to small supermarket owners next, do not write about a florist, even if that florist adores you. The customers who love you most are not necessarily your target customers. Pick the wrong person and the case study can be beautifully written and still bring in the wrong people.
Where do you find them? Four places, ordered from easiest. First, scores: if you have run a satisfaction survey, pull the list of people who scored 9 or above — that is where Zhanggui Bao's 96 came from. Second, renewals: someone who renewed twice has voted with their wallet twice, which is worth far more than praise in words. Third, support logs: people who messaged you to say something nice are right there in the chat history. Zhanggui Bao's very first case study came this way — a stationery shop owner wrote in support, "reconciliation took only half an hour this month," and A Nan followed that sentence to an invitation. Fourth, usage: the top 10% by usage live in your product every day, so they always have stories. Gather 20 candidates from these four places, run them through the three filters down to 12, then send invitations one by one. Do not list only 9 candidates: the invitation agreement rate is only a bit over four in ten, and if you list too few you will not end up with 9 pieces.
What if someone agrees to an interview but will not go public? Do not push. Step down a level — there are three rungs. The first is an anonymous case study: no real name or shop name, just "a three-person stationery shop," numbers included, story intact. It is about 80% as believable, but better than nothing. The second is dropping to a one-line testimonial: take the most honest sentence from the interview and rotate it on the homepage; no name needed. The third is to file it away: keep the recording and the numbers, and ask again when their business has grown or they have moved to a bigger shop — plenty of people say yes six months later. People you push into going public say only polite things in the interview, and the case study comes out unreadable. Remember: going public is something they give you, not something you are owed.
The interview script: 5 numbers in 20 minutes

The interview does not need to be long. Twenty minutes is enough — past twenty, the other person starts checking the clock. Cut the time into three parts: the first 3 minutes on how bad things used to be, the middle 10 minutes on 5 numbers, the last 5 minutes on one question: "if you recommended this to a peer, what would you say?" Those 5 numbers are the skeleton of the whole case study, so list them before the interview and ask them in order: first, how many hours a week reconciliation used to take; second, how many it takes now; third, how much money it saves a month (a part-timer, overtime, an accountant — anything counts); fourth, how many more orders a month it lets them handle, if that can be counted; fifth, how many days from signing up to the first moment of "this thing actually helps." Ask specific questions, not "how do you like our product" — that only ever gets you "it's fine." Ask "last Tuesday, what time did you start reconciliation and what time did you finish?" Specific questions get specific answers.
Three things to settle before the interview starts. First, ask "may I record this?" and only press record after they say yes. Without a recording you are writing from memory after 20 minutes of talk, and the numbers will drift. Second, tell them the draft goes to them first and nothing is published until they say yes — put it in the invitation, then say it again at the start of the interview, so they dare to tell the truth. Third, do not sell during the interview: do not ask "what do you think of our new feature," do not hand over a business card, do not casually ask about renewing. An interview is an interview. The moment you start selling, the other person knows exactly what the conversation is for, and everything after that changes flavor. Copy the 5 numbers out the same day you record; wait until tomorrow and they start to blur. Send the first draft within a week of the interview — past two weeks, the other person has forgotten what they said, and the back-and-forth on revisions can drag on for a month.
Draft structure: one page that says "how bad it was, how good it is now"
Six sections are enough for one case study — do not turn it into a long essay. The first is the headline, built as "number + outcome": "Reconciliation used to take 6 hours a week, now it takes 40 minutes: one small shop owner's year with Zhanggui Bao." A headline needs a number in it; without one, nobody clicks. The second is a one-sentence intro saying who this person is, what they sell, and how long they have used your product: "Sister Wang runs a stationery shop in a county town, three people on staff, and has used Zhanggui Bao for 11 months." The third is the "before": how reconciliation used to work, how much time it ate, what annoyed them most, with one number attached. The fourth is "how they use it": three steps, not a feature list — what they open first when they sit down, then what they do next. The fifth is the "now": a small before-and-after table — 6 hours a week before, 40 minutes now; 900 dollars a month on a part-timer before, 0 now; reconciliation dragging three days past month-end before, done the same day now. The sixth ends on the other person's own words — that last interview question, "if you recommended this to a peer, what would you say," quoted as-is, without polishing it into something nobody would actually say.
Writing the numbers has three rules. First, only write numbers the other person actually confirmed — never calculate on their behalf. If they said "about 900 dollars a month saved," write "about 900 dollars a month saved," not "10,800 dollars a year" — words they never said are words you made up. Second, write "about" in front of numbers nobody is sure of: how many days it took to get started is something the other person does not remember exactly either, so "about 3 days" is more honest than "3 days." Third, get them to sign off on each item before publishing: real name, shop name, photo, every number — all four ticked before it goes live. Zhanggui Bao sends the draft together with a permission checklist, and one reply of "good to publish" counts, email included. One piece stalled halfway because the person said "do not use the full shop name, just call it Sister Wang's Stationery" — A Nan changed it and published anyway. A shortened name does not hurt the story; a weakened story hurts the case study. The other way round, one person asked to change "saved 900 dollars" to "saved quite a bit." Once the number came out, the whole piece collapsed, and A Nan dropped it outright: a case study with no numbers is not worth publishing.
Do not publish a case study once and forget it — rewrite it into three versions and one piece does the work of three. The first is the homepage version: cut the headline and the two before-and-after numbers into a card in the homepage case study section, linking to the full text. The second is the sales email version: under 150 words, saying how bad it was and how good it is now, sent to people whose trial expired without paying. The third is the social version: three sentences plus a photo of the person's shop, posted on your own account with a link to the full case study at the end. All three come from the same draft — do not rewrite them. Rewrite once and the numbers can drift once. When you adapt, only cut, never add: cut details, keep the numbers and the quotes, and stop cutting only when there is nothing left to cut.
Where to put it, and how to measure it
Where a case study goes matters more than what it says. Three spots work best. The first is next to the pricing page: when someone hesitates at the price, a line beside it — "Sister Wang's stationery shop has used Zhanggui Bao for 11 months and saves about 900 dollars a month" — cuts the hesitation in half. The second is the case study section on the homepage: only the 3 best pieces, never stack all 9 — nobody reads nine in a row. The third is inside sales emails: send people whose trial expired without paying a note that says "here is how a shop owner in a situation like yours uses it," which works far better than another "your trial is expiring" reminder. Do not just post it on a social account or a newsletter: people scroll past and it is gone, nothing stays, nothing gets measured. A case study is a long-term asset. It belongs where people make decisions, not in one-off content.
Measure it with four numbers, once a month. The first is the invitation agreement rate: people who agreed divided by people invited — Zhanggui Bao's example is 40 / 96 = 42%. The second is the draft rate: drafts written divided by interviews completed, 14 / 24 = 58%. The third is the publication rate: pieces published divided by drafts written, 9 / 14 = 64%. The fourth is the case study page's trial click rate: how many of every 100 views clicked "Start free trial" — Zhanggui Bao's case study page gets 9. If the agreement rate falls, rewrite the invitation. If the draft rate falls, rewrite the interview script — if you cannot get numbers out of a conversation, the script is the problem. If the trial click rate falls, swap in a better case study at the top. Do not go by feel and say "case studies seem to help." With four numbers on the table, helpful or not is obvious at a glance.
Case studies also expire, so do not publish and forget them. Check back every six months and ask three questions: have the numbers changed, are they still using it, is there anything new at the shop? Numbers changed, update them. The product changed, rewrite it. They stopped using it, take it down. A piece that still says "saves 900 dollars a month" when the shop has doubled in size looks stale to anyone who knows the business; worse, if they have already cancelled and the case study is still on the homepage, one question from a new visitor exposes the whole thing. An expired case study does more damage than no case study at all. The check-in is also a chance to ask for one fresh quote — update once, and the piece works for another six months.
Three things to do this week
Do not try to finish all 9 pieces at once. Three things this week are enough. First, make a 20-person candidate list from people who scored 9 or above, people who renewed twice, and people who praised you in support chats — then run the three filters down to 12. Second, send invitations in the three-sentence format: "I am A Nan from Zhanggui Bao. I saw you renewed twice and would love 20 minutes to hear how you do reconciliation. I will send you the draft first — nothing goes live until you say yes." Send 12 first, not all of them — keep half back, watch the replies, then adjust the wording. Third, do the first interview you land: 20 minutes, record it, copy out the 5 numbers the same day, and send the first draft to the person within a week. Until the first publishable case study exists, do not think about the second. Once the first one has run through the whole process, the other 8 are just repetition.
One last thing: a case study is not an ad. It is the truth told through your user's mouth. When you say "my product works great," nobody believes you. When someone who looks exactly like your target customer says "reconciliation used to take me 6 hours a week, now it takes 40 minutes," the people who are hesitating believe them. All the praise sitting in your chat logs is worth nothing until it moves onto your website. Make the list of 20 people, and send the first invitation tonight.
Comments (0)
Related articles
One page should stand up on a phone, a tablet, and a computer. This guide follows MDN's Responsive web design, Using media queries, and the viewport meta reference: fluid HTML, the viewport tag, fluid grids and max-width, images that never burst their container, media query syntax, breakpoints set by content, mobile first, responsive type, testing capability instead of width, and a pre-launch checklist.
localStorage only stores strings and has no indexes. IndexedDB is a real database in the browser: structured data, far larger capacity, indexes, and transactions. Following MDN's Using IndexedDB, IDBDatabase, and IDBObjectStore: the five-step routine, opening, upgrades, keys, indexes, transactions, add versus put, get, cursors, ranges, and multi-tab upgrades.
You already call functions every day; this guide names what you have been doing. Following MDN's Functions, Arrow function expressions, and Rest parameters: declarations versus expressions, parameters and arguments, default values, rest parameters, spread syntax, return values, arrow function shorthand, the this trap, callbacks, and scope.