Venture Studio · Guided Apprenticeships · Agentic Marketing Agency

The Log · Essay · 11 min read

Essay · Operations

I Killed a Survey App Over 50 Cents. A Company in My Category Sold for $1.07 Billion.

By Max Palmer · 2026-10-29 · 11 min · Essay
MaxShip Log card, The Log · Kill Switch. A bronze airship captain at a riveted terminal grips a brass lever with a red knob. Blank speech bubbles, spilled copper coins, and a small silver airship sit against sunset clouds. Masthead reads MAXSHIP and THE LOG · KILL SWITCH.

In 2016 I co-founded a survey app in San Francisco called Chati.Chat. About a year later I shut it down, partly because I couldn't figure out how to pay people 20 to 50 cents a question. Four years after that, Lucid, a company that matched researchers with survey respondents, sold to Cint for $1.07 billion.

Lucid wasn't my idea, and nobody stole anything. Lucid was in my category, though, and so was Swagbucks. Swagbucks' parent company did more than $300 million in revenue in 2021.

Data chart titled I quit in 2017. The category didn't. Timeline from 2008, when Swagbucks starts paying rewards in points, through May 2016 to June 2017 marked Chati.Chat: 14 months, then I kill it, to Dec 29, 2021, Cint buys Lucid for $1.07B, with Swagbucks' parent at $300M+ revenue in 2021. A zoom says Facebook's rules changed 5 times: Apr 12, 2016 Messenger opens to bots; May 26, 2016 Platform Policy updated; Aug 15, 2016 Messenger rules rewritten, 24-hour reply window; Apr 18, 2017 and May 15, 2017 Platform Policy updated.

This post covers that call, the opposite mistake, and the system I use now to avoid both.

The survey app I killed too early

Brands and researchers wanted answers. We used Facebook's ad targeting to find the right people, then asked them questions over Messenger and paid them per answer. It was bootstrapped, with a small team, and I paid developers out of my own pocket.

Three things killed it. I'll give you all three, because the clean one-reason version would be a lie.

Card titled Three things killed it. Eyebrow Chati.Chat · San Francisco, 2016. 1 The payout math: the fees could eat most or all of a 20 to 50 cent answer, chip Adjust problem. 2 Facebook's rules: paying strangers per answer inside Messenger was never what the rules were built for. 3 My focus: I had a growing Amazon business, and I wanted to put my hours there. Kicker: The one that stings is #1. Nobody solved micro-payments. They batched them.

1. The payout math. We paid through PayPal, and the obvious rails were brutal for pocket change. The standard rate was 2.9% plus 30 cents, and PayPal's newer Payouts API charged a flat 25 cents per US payment, so the fee could eat most or all of a 20 to 50 cent answer. The cheap option, Mass Pay at 2%, needed PayPal's permission, and PayPal could pull it any time. I couldn't crack the formula, and I overthought the economics until I talked myself out of it.

Data chart, PayPal 2016, fee as a share of one payout. Headline: The fee was bigger than the answer. Standard rate 2.9% + 30¢ is 153% of a 20¢ answer (31¢ fee) and 63% of a 50¢ answer. Payouts API, 25¢ flat per US payment, is 125% of 20¢ and 50% of 50¢, and also needed PayPal's permission. Mass Pay at 2% capped at $1 is 2% of both, 0.4¢ and 1¢, and PayPal could pull permission any time. A line marks fee equals the whole answer. Footer: The cheap rail existed. It was gated.

2. Facebook's rules. Facebook only opened Messenger to bots in April 2016, then rewrote the rules that August. From then on you could only message people within 24 hours of their last reply. Opt-in subscription messages were limited to news, productivity and personal trackers. On top of that, the platform policy only let you reward people for logging in, entering a promotion or checking in. Paying strangers per answer inside Messenger was never what those rules were built for. It was a great use of existing tech that the terms didn't make room for.

3. My focus. I had a growing Amazon business at the same time, and I wanted to put my hours there.

All three are real. The one that stings is the payout problem, because the market had already solved it before I started. Swagbucks had been running on points since 2008. You pile up small earnings and cash out once you pass a threshold. Today that's $5 to PayPal, or a $1 Amazon gift card. Survey Junkie cashes out at $5. Google Opinion Rewards pays Android users in Play credit and only sends PayPal cash once you reach $2. Nobody solved micro-payments. They batched them.

Data chart titled Nobody solved micro-payments. They batched them. The same 25 cent PayPal fee is 125% of a single small payout and about 5% once those payouts are batched to a cash-out threshold. Stamped data, sourced.

The fix for my "impossible" problem was points and gift cards, and they were sitting right there in the market while I was building. I didn't look sideways.

Prompt card KS-17, Look sideways before I hit kill. Use it when your finger is on the kill switch over one problem. The card shows a copyable prompt: I'm about to kill [IDEA] because of [THE PROBLEM]. Before I do, look sideways: who already solved it, is it a kill or an adjust, and what's the cheapest 2-week test. From The survey app I killed too early.

KS-17 · Look sideways before I hit kill

I'm about to kill [IDEA] because of [THE PROBLEM].

Before I do, look sideways:

1. Who in this category has already solved this problem or worked around it? Name real companies and say how they did it.
2. Is mine a kill problem or an adjust problem? Explain why.
3. If it's an adjust problem, what's the cheapest 2-week test of the workaround?

Only name companies you're confident exist. If you're not sure, say so.

So I killed a whole business over one choke point. That's an adjust problem, and I treated it like a kill problem.

The opposite mistake: an agency I kept alive too long

Killing too early is loud. Running too long is quiet, and it costs more.

At an agency I ran, we got stuck in experiment loops. We'd try something, read the tea leaves, then try something else. The problem was that we tried each direction once and made big decisions off that single data point.

One attempt is an anecdote, not a test.

Both failures come from the same place. No gate was set before launch. With Chati.Chat I quit on a feeling about the math. With the agency I kept going on a feeling that the next experiment would be the one. Neither time had I decided upfront what "working" looked like, or how many reps a direction got before I judged it.

Diagram titled Two ways to get it wrong. One missing gate. Stamped diagram, not measured data. Killing too early and running too long both come from no gate set before launch.

You can't score apples like oranges

Every business is shaped differently, so every business needs its own scorecard. Take two I've worked with.

A game publisher lives and dies on fandom. One trusted expert reviewer can move thousands of buyers, and the fans pay attention to what reviewers say. Reviews and fan sentiment are leading numbers.

A hospitality technology optimization business is B2B back-office tech. It squeezes small margin and operations gains out of hospitality businesses. That makes it pure numbers. Customer opinion barely matters as long as the tech delivers what it promised. What counts is whether the margin moved, whether the system stayed up, and whether the client renewed.

Score those two on the same field and you'll kill the wrong one.

Diagram, niche numbers versus the universal line. A matrix of business shapes with their own metrics, and a shared line for P&L, margin, and cash at risk. Stamped diagram, not measured data. KS-07 stands in for the apples-and-oranges card.

A reorder rate is a great metric for a consumable and a useless one for a SaaS tool. Ask every opportunity the same questions so you can compare them, but let each niche set its own metrics, thresholds and timelines.

Sometimes the deciding gate isn't the P&L at all. It's a rule, a license or a platform policy. Chati.Chat is proof.

What AI speeds up, and what it can't

AI changed the front half of this. I can build a site, a shop, an app, ad creative and a launch in a fraction of the old time. Six months of "is there anything here?" turns into a couple of weeks of real signal. (Here's how I build a site with AI.)

There's plenty AI can't rush: long-term loyalty, reorders, a real organic following and trust from reviewers. A reorder only shows up after a full reorder cycle, and no tool shortens that.

So you get fast signal: can we buy attention, and do people convert? You also get slow signal: do they come back, and is a customer worth more than they cost? Don't kill on a slow question using fast data.

Gates, not formulas

A gate is a scheduled check-in with yourself, and your partners if you have them, to decide where to go next.

Here's one example clock, for a consumer product people reorder fast:

  • Day 15: ad and creative signal, cost to acquire a customer
  • Day 30: early retention
  • Day 45: conversion and first reorders
  • Day 60: reorder pace
  • Day 90: lifetime value and the big verdict
Diagram, set your gates before day 0. An example clock on a day scale: Day 0 launch, Day 15 ad signal, Day 30 retention, Day 45 conversion, Day 60 reorders, Day 90 verdict. Each gate has continue, adjust, and kill. Footnote: one niche's clock, a consumer product people reorder fast. Stamped example clock, not measured data.

That's one niche's clock. A venue, a SaaS product or a service line runs a different one. Set yours before day 0, and decide what day 0 even means (site live, first ad or first sale).

Prompt card KS-14, Write my kill criteria before I start. Prompt 1 of 5. Use it before day 0, before you're attached to it. The card shows the copyable prompt that asks for three numbers, gate dates, a kills / revives / next-test line, and a stop rule and a scale rule. From Gates, not formulas.

KS-14 · Write my kill criteria before I start

I'm about to launch [IDEA]. Before I start, help me write my kill criteria.

Business shape: [fandom / pure numbers / B2B tech / other]
Day 0 means: [site live / first ad / first sale]
Time and money I can risk: [AMOUNT]

1. Pick the 3 numbers that matter most for this shape, and say why.
2. Set gates for my niche (for example day 15, 30 and 90) and what each one checks.
3. For each gate, write what kills it, what revives it, and the cheapest next test.
4. Write my stop rule and my scale rule in one sentence each.

"I'm tired of this" is not a kill condition. Don't let me use it.

Every gate has three possible verdicts: continue, adjust or kill. Every verdict has to name three things:

  • What kills it. Something specific you can observe, like losing money on every order after a set number of orders, or a platform saying no in writing.
  • What revives it. Evidence, not enthusiasm. A cohort that reorders, a real price from a real buyer, or a gatekeeper's yes.
  • The cheapest next test. Run a free check before a paid one, and have a conversation before you spend money.

Run Chati.Chat back through that. Kill: the platform rules block delivery. Revive: a payout method where fees stop eating the reward. Cheapest next test: a points wallet with a gift-card threshold for a small beta group, two weeks, near-zero cost. I never ran that test. I just hit kill.

Card titled Continue, adjust or kill. Then write down: 1 What kills it, something specific you can observe. 2 What revives it, evidence not enthusiasm. 3 The cheapest next test, a free check before a paid one. A strip reruns the survey app: kill if platform rules block delivery, revive with a payout where fees stop eating the reward, test a points wallet with a gift-card threshold.

Adjust means fix the choke point

Take a hypothetical. You launch an e-commerce product and your acquisition cost comes in lower than expected, which is great, but returns are high.

Don't throw it out. Find the single step with the most friction and fix that first. That might mean better instructions, better packaging or a small product tweak.

That's the bottleneck idea: a system only moves as fast as its tightest point. It comes from Eli Goldratt's Theory of Constraints.

When you adjust, change one variable, whether that's the offer, the audience, the channel, the price or the product. Then give it enough reps to mean something. That's the agency lesson again.

Prompt card KS-16, Find the choke point before I kill it. Use it when a gate comes back mixed. The card shows a prompt that names the single step with the most friction, three fixes cheapest first, and how many reps to give one change. From Adjust means fix the choke point.

KS-16 · Find the choke point before I kill it

My launch has a problem. Help me find the choke point before I kill it.

What's working: [e.g. acquisition cost is lower than expected]
What's not: [e.g. returns are high]
My funnel: [e.g. ad > page > checkout > delivery > reorder]

1. Name the single step with the most friction, and show your reasoning.
2. Give me 3 fixes for that step only, cheapest first.
3. Pick one variable to change (offer, audience, channel, price or product) and tell me how many reps it needs before I judge it.

Don't pitch me a new idea. Fix this one first.

The captain picks the rule

There's no single right way to make the call. It depends on your time, budget, effort and reason for doing it. Three setups work:

  1. Pure numbers, ruthless. Set thresholds in advance and let the numbers decide.
  2. An automated KPI kill. For example, a subscription business shuts itself off if acquisition cost or lifetime value isn't better than average by the gate. (I wire tripwires like this in PostHog.)
  3. Gate questions. What did this launch prove? What did it generate? What value did it bring? What lessons did it give us? Answer them at the gate and decide on the spot.
Card titled At the gate, ask four questions. Eyebrow The captain picks the rule · setup 3. 1 What did this launch prove? 2 What did it generate? 3 What value did it bring? 4 What lessons did it give us? Answer them at the gate and decide on the spot.

AI can watch the dashboards. The captain still has to stay awake at the wheel.

My gut vs. the spreadsheet

My gut wins on executive, brand, customer-experience, design and creative calls. There's a line attributed to Henry Ford: "If I had asked people what they wanted, they would have said faster horses." Sometimes the captain sees past what customers can put into words.

The spreadsheet wins on the verdict itself. Gut decides what to try. The numbers decide whether it worked.

My gut has also lied to me, in two patterns I've learned to override.

Card titled My gut lies to me in two patterns. Pattern one, the honeymoon: excited to start isn't excited to keep going. Only the second survives the long, boring middle. Pattern two, quitting right before the diamonds: I'm tired of this isn't a kill condition. Footer: Gut decides what to try. The numbers decide whether it worked.

The honeymoon. When I get excited about something, I take it as a sign I should do it. A lot of that early excitement is the honeymoon phase of a romance that won't last. It's a new toy. Being excited to do something is different from being excited to do it for a very long time, and only the second one survives the long, boring middle.

Quitting right before the diamonds. You've probably seen the meme: a miner tunneling through rock turns around and walks away, and he's inches from a wall of diamonds. My gut tells me to stop a bit too early. Gates are how I catch it, because "I'm tired of this" isn't one of the kill conditions.

Prompt card KS-18, Honeymoon or a 5-year interest? Use it when you're fired up about a new idea. The card shows a prompt that interviews you, seven questions max, about a boring Tuesday in year 3, the weekly work, and the lifestyle the business drags behind it, then says whether this is honeymoon excitement or a 5-year interest. From My gut vs. the spreadsheet.

KS-18 · Honeymoon or a 5-year interest?

Am I excited about [IDEA], or excited about a new toy?

Interview me. Ask one question at a time, 7 questions max, about:

- what I'd be doing on a boring Tuesday in year 3
- the parts of the work I'd have to do every single week
- the lifestyle this business drags behind it if it works

Then tell me straight: is this honeymoon excitement or a 5-year interest? Quote my own answers back as evidence.

Before the kill switch comes the pick

Everything above is for after you launch. Picking the idea comes first, and I covered it in my 10-sign checklist video. Sign five is that competition is proof the dogs are eating the dog food. With Chati.Chat I saw Swagbucks as a threat when it was actually a playbook.

Rubric chart titled Score the idea before you build it. Ten signs with score bands, from the video Pick the Right Business Idea Before You Waste Years. Quotes on the card include I just saved you a lot of money and a year of work, and Go for it. You're in the right zone.

Sidebar: The Gate Sheet

Fill this in before day 0.

  1. Business shape: fandom / pure numbers / B2B tech / other: ______
  2. Day 0 is: ______ Gate dates: ______
  3. My 3 niche numbers: ______ Universal line (P&L, margin, cash at risk): ______
  4. Stop rule: ______ Scale rule: ______
  5. Who decides, and how (pure numbers / automated KPI / gate questions): ______
  6. At each gate: verdict ______ kills it ______ revives it ______ cheapest next test ______
  7. Premortem: "It's day 90 and this failed. Why?" ______
Night card, Sidebar · fill this in before day 0. The Gate Sheet. 1 Business shape. 2 Day 0 and the gate dates. 3 My 3 niche numbers, plus P&L, margin, cash at risk. 4 Stop rule and scale rule. 5 Who decides, and how. 6 At each gate: verdict, kills it, revives it, cheapest next test. 7 Premortem: It's day 90 and this failed. Why? Kicker: Set the gates before you launch. I'm tired of this isn't a kill condition.
Prompt card KS-15, Run a premortem on this idea. Use it the week before launch. The card shows a prompt: It's day 90 and this failed. Write the post-mortem. Five ranked reasons, the earliest signal for each, any reason that is a rule or a platform policy, and the cheapest test this week. From Sidebar: The Gate Sheet.

KS-15 · Run a premortem on this idea

Run a premortem on [IDEA].

It's day 90 and this failed. Write the post-mortem.

- Give me the 5 most likely reasons, ranked by how likely they are.
- For each one, name the earliest signal I'd see and the day I'd see it.
- Flag any reason that's a rule, a license or a platform policy, not a P&L number.
- Tell me the cheapest test I could run this week to rule out the top reason.

Be blunt. I'd rather hear it now than on day 90.

The short version

Set your gates before you launch, read your own niche's numbers, and make every verdict name what kills it, what revives it, and the cheapest next test. Before you kill something, look sideways. Someone in your category may have already paid for the answer.

If you're picking your next idea, start with the 10-sign checklist. If you're already running one, steal the prompts above and set your gates this week. I post what I'm building, and what I kill, on X @PalafoxPalmer.

The idea, in brief

  • Lesson 1 of 10, The idea in brief. Look sideways before you kill. I killed a survey app over payouts Swagbucks had already solved with points. Someone in your category may have already paid for the answer. From The survey app I killed too early.
  • Lesson 2 of 10, The idea in brief. Write the kill criteria before you start. Decide what working looks like, and how many reps each direction gets, before day 0. After launch, your mood writes the rules. From The opposite mistake.
  • Lesson 3 of 10, The idea in brief. One formula can't score every business. A game publisher lives on reviews. A back-office tech business lives on renewals. Score them the same way and you'll kill the wrong one. From You can't score apples like oranges.
  • Lesson 4 of 10, The idea in brief. Universal numbers still get a vote. Let each niche pick its own metrics. P&L, margin and cash at risk still sit on every scorecard. From You can't score apples like oranges.
  • Lesson 5 of 10, The idea in brief. AI speeds up the signal, not the loyalty. Six months of "is there anything here?" is now a couple of weeks. Reorders, loyalty and a real following take as long as they take. Don't kill on a slow question with fast data. From What AI speeds up, and what it can't.
  • Lesson 6 of 10, The idea in brief. "Kill" alone isn't a verdict. Every verdict names three things: what kills it, what revives it, and the cheapest next test. From Gates, not formulas.
  • Lesson 7 of 10, The idea in brief. Adjust means fix the choke point. Find the one step with the most friction and fix it first. Change one variable, then give it real reps. One attempt is an anecdote, not a test. From Adjust means fix the choke point.
  • Lesson 8 of 10, The idea in brief. The kill call is the captain's. Pure numbers, an automated KPI kill or gate questions. Pick your rule before launch. AI can watch the dashboards, but you stay at the wheel. From The captain picks the rule.
  • Lesson 9 of 10, The idea in brief. Gut picks what to try. Numbers say if it worked. My gut wins on brand, design, customer experience and creative. The spreadsheet gets the final verdict. From My gut vs. the spreadsheet.
  • Lesson 10 of 10, The idea in brief. Your gut lies twice. Once at the start, when a new toy feels like a calling. Once near the end, when you're about to quit inches from the diamonds. From My gut vs. the spreadsheet.

Sources