ICP and the trigger window

the idea

Fit and timing are two layers.

Both can be checked from outside. They differ in how fast they go stale, and that decides where each one lives in the system.

The literature disagrees about whether triggers belong in an ICP. Dave Kellogg says no: "Whether they're out shopping right now, whether they have budget, whether they have a buying timeframe and purchasing authority are all important qualification questions, but they are not part of the ICP itself." Loading timing into the ICP makes it a moving target you can't build a list from. Bowery Capital and Winning by Design (the C in SPICED: "an event/deadline that drives customers to make a decision") put buying signals inside it.

They are arguing past each other. Keep both, as separate layers. A candidate must pass the fit filter to enter the list at all, and carries a trigger to decide their place in the queue. Someone who fits but has no live trigger is a real prospect with no urgency: worth nurturing, not a message this week.

The term of art for the timing layer is Craig Elias's Window of Dissatisfaction: "unhappy with Status Quo but not searching for Alternatives yet." It opens when something breaks the person's belief that their current approach works, and closes when they fix it, buy something or give up. Trigger design is an attempt to catch people inside that window.

the idea

What an ICP is, and isn't.

There is no single canonical definition, but the serious sources converge: an ICP describes fit, not readiness, and it must collapse a market into a list you could build by Friday.

• Dave Kellogg: "the ideal person at the ideal company and the problem you're looking to solve for them." Three dimensions, not one.
• Clearbit: "TAM is the total accounts that could buy your product, ICP is the portion of your TAM you should be convincing to buy, and personas are your best type of contacts within ICP accounts." TAM is a number for investors; an ICP is a filter you can run against a database.
• April Dunford says best-fit customers and defines them by behavior first: they "understood your product quickly and bought from you quickly. They became raving fans, referred you to other companies and acted as a reference for you." Pick the behavior, then work back to the attributes.
• Geoffrey Moore adds the test that kills most bad segments: a market is "a set of customers for a product, with common needs and wants, and who have the ability to reference one another." If two prospects couldn't plausibly know each other, you have a list, not a market.

Three tests a real ICP passes

What it is not

• Not a demographic sketch. Jason Cohen: no "Carol has two cats." Invented detail crowds out the attributes that predict fit.
• Not qualification. Budget, authority and timeline are questions for one prospect, not properties of the segment.
• Not your positioning line. "Solo founders who want users on autopilot" is a sentence for a landing page. An ICP is a query, so an agent knows where to look.

Brief Intake · turning fields into a filter

building one

Five ways to build one.

Pick by what you have. The first needs customers; the last two don't.

Regression, not intuition

Kellogg: "Your ICP starts out as an aspiration (about who you want to sell to) and over time becomes a regression." Once there's data, fit it against outcomes: win rate, cycle length, discount depth, retention, referenceability. "Some of your intuition will be proven correct, but some won't." Two cautions: current customers reflect where you happened to hunt, and his failure mode faux focus is an ICP that exists as a document while activity goes everywhere. He writes the result as concentric rings (ring 0 gets full effort) rather than a yes/no gate.

Beachhead

Moore's four gates, any one of which kills a segment: reachable through a channel you have; a compelling reason to buy (acute pain); a whole product deliverable in three to twelve months; nobody already owns the segment. Make it concrete with named scenarios (a specific person, role and day), rank them, then "go hard in the direction chosen, regardless of doubts."

The switch interview

Interview recent buyers and rebuild the timeline backwards to the first thought. The governing equation is Push + Pull > Anxiety + Habit. It gives you the trigger event and the real alternative, the inputs to Dunford's process.

Dunford's process

1. Competitive alternatives: what would they do if you didn't exist? Often a spreadsheet, an intern, or nothing; about 40% of B2B deals are lost to no decision.
2. Unique attributes the alternatives lack.
3. Value: "Customers don't care about your unique features, they care what those features can do for them."
4. Best-fit customers: who is structurally forced to care a lot about that value.
5. Market category: the frame that makes the value obvious.

Step 4 is derived: start from what you're uniquely good at and find who has to care. "It isn't enough to say 'We sell to small businesses'; there are millions of those and they are all different."

Sales Safari

"People go on the internet to complain about things because something hurts." Find the watering holes and read what people say when nobody asks them anything. Take notes in four passes, one per quadrant: pain, jargon, worldview, recommendations. Look for emotion ("struggle, confusion, disagreement, debate"), not feature requests. The rule: "You are not your audience." Its output is doubly useful: the words people use for their pain are both the ICP definition and the copy.

Market Research · four kinds of competition — Brief Intake · audience research step

building one

Starting from zero customers.

With nothing to regress on, write a hypothesis and try to break it. Lenny Rachitsky's founder survey: most founders got their ICP wrong at first; iteration speed won. Persona's Rick Song: "We did the ICP exercise 17 times in the early days."

1. State it as three attributes, chosen from nine categories (company size, job title, pain solved, working style, tech used, type of business, price point, geography, community spaces). If you can't build the list, it isn't a hypothesis.
2. State the predicted consequence. Dunford: they understand fast, buy fast, don't ask for discounts, refer. Vohra: at least 40% say they'd be very disappointed without it.
3. Test with cold outbound, holding the message constant. Warm intros contaminate the data. Vary one attribute at a time; measure replies.
4. Read four signals: conversion jumps, enthusiasm jumps, urgency to act, and "the nod".
5. Kill or narrow. Never widen.

Sales Safari is the strongest zero-customer source: it needs no access to anyone. Rob Walling's order: "Market comes first, marketing second, aesthetic third, and functionality a distant fourth." Own 100% of a 5,000-person niche before 10% of a 50,000-person market. Reachability recurs from three directions (Walling, Moore, and Lenny's "unique community spaces"), so it's a defining property of a good ICP, not a convenience. Dunford's caution pre-fit: hold the segment as a hypothesis you test, not an identity you defend.

building one

How narrow is narrow enough.

Narrower than feels comfortable.

Comically narrow, in practice: Gusto ("only companies that are five or fewer employees in California that offer no benefits"), Gong ("software companies selling in the U.S. in English, via video conferencing, $1,000–$100k ACV"), Snyk ("a developer building with Node.js who was very security-conscious"), and Canva, which only chose social media managers six months in.

Jason Cohen explains why narrow widens reach: aim at Carol (custom-built to love it), and Diana (10× larger) and Eddie (100× larger) still hear you, because the mechanism is message power, not audience size. "Selling to everyone means selling to no one." Nathan Barry adds the reason most templates omit: "The smaller the circle you target the more likely your prospects know each other." Narrowing makes word of mouth mechanically possible.

building one

What real ICPs look like.

ConvertKit, in detail

The targeting unit was much smaller than the category: "Recipe blogs. Paleo recipe blogs. Paleo recipe blogs run by women." The test: "With some googling we could probably list all the paleo recipe blogs run by women." Found bottom-up ("look for a niche where we already have 1-3 customers, then make a list of everyone like them") and widened along community lines: interval training → running → powerlifting, because those people read each other. His prospecting used BuiltWith to find sites already paying MailChimp or AWeber: "already paying a competitor" proves budget, category awareness and dissatisfaction in one check. Result: two years to $2,000 a month, then $15,000 six months after narrowing.

What separates a real ICP from a vague one

• It names something you can list by Friday.
• The best criterion is an observable behavior, not an attribute: "sends 15–40 emails a day", "already pays for MailChimp".
• Every good one has an explicit exclusion, with the reason.
• Narrow is a targeting unit, not a ceiling.
• The purpose is density: a segment where nobody knows anybody can't compound.
• A real ICP changes a budget, a roadmap or a refusal, not just a headline.
• Bottom-up beats top-down. Nobody here reasoned to an ICP from a TAM slide.

building one

Selling to founders.

A common category, and famously hard: founders are broad, low-budget and churn out of subscriptions. Five escapes:

using it

Putting an ICP to work.

A written ICP that changes nothing is faux focus. Six places it should show up:

• Lead sourcing. Count first: the denominator, then aware, interested, considering, chosen. Reachability is a gate.
• Qualification and disqualification. Bob Apollo: fit runs before BANT or MEDDICC; a green, amber, red score; a named list you "either avoid targeting in the first place or politely disqualify." Cohen on turning down non-ICP revenue: "This separates first-time founders from seasoned founders."
• Messaging. The value must land on Carol in three to ten seconds. Sales Safari's jargon and worldview passes are the copy deck. Dunford: "It's always better to be a little boring than completely baffling."
• Channel. Walling treats channel as a property of the niche. Plausible ran Twitter and Mastodon only.
• Roadmap. Vohra's 50/50: half deepening what the very-disappointed love, half removing what blocks the somewhat-disappointed who got the core benefit. Cohen: only ICP churn is signal.
• Pricing. A segment that can't pay your price isn't a segment. ProfitWell's quantified personas attach willingness to pay, LTV and CAC to each segment.

GTM tracking plan · the ICP fit model

using it

When an ICP is wrong.

Dunford's symptoms, in order of diagnostic value: current customers love it while prospects are confused; long cycles with low conversion; high churn; constant discount requests. Her causes: build delta (the product moved, the messaging didn't) and market shift, rooted in "hanging on to a 'default' market position that is rooted in the history of the product idea." Change or persist turns on one question: are your existing customers in the segment thriving? If yes and new ones are confused, it's messaging. If no, the segment is wrong.

vibedraft

The trigger window: day 3 to day 45.

The window opens about three days after launch and closes near day 45, peaking at days 5–21 with a second spike at 28–35. Before day 3 founders are euphoric; after day 60 they're rationalizing an exit, and a message about customers reads as a rebuke.

0–2 — euphoric · send nothing

3–10 — "traffic died, what now" · primary

11–21 — trial purgatory · activation help

28–35 — "my first month numbers" · secondary

36–60 — cooling

60+ — rationalizing an exit · nothing

Bar lengths are the relative priority the research assigned, not measured rates. The shape comes from 26 dated founder posts in a five-week window: directional, not a tracked cohort.

Two forcing events

• The traffic dies, day 2–4. Day zero is roughly 60% of a launch's traffic, days 1–7 about 25%. "Traffic went back to baseline within 48 hours."
• The money doesn't arrive, day 28–35. Trials expire. About 35–45% of signups activate within 7 days and 5–10% of those pay within 30, so a 50-signup launch produces one or two customers, often zero. One founder's #1 Product of the Day with 612 upvotes converted one.

The tone shift is unmistakable: day 3–45 posts ask tactical questions ("where did your first users come from?"); day 60+ posts ask existential ones ("Should I give up or not?").

Buying Moments Map · tried it and it failed, taking stock — Playbook Library · the window after a launch

vibedraft

Detecting launches.

This inverts the acquisition problem. Phrase search returns a noisy stream you filter down. A launch directory returns a complete, dated list, and the date is the trigger, free, with no inference.

The sources, verified Sep 19, 2026

Runners-up: Smol Launch (121 a week, exposes maker handles), Fazier (maker names and upvotes), TinyLaunch (sequential ids). Skip AlternativeTo, F6S, Dev Hunt, Launching Next and Tiny Startups. launchdirectories.com catalogs 100+ directories and which tiers cost money. TrustMRR is enrichment, not detection: its "recently added" means added to TrustMRR.

The chain

launch feeds · → · dedupe on domain + author · → · fit filter (cheap checks) · → · fetch the site: price, team, buyer · → · TrustMRR by domain · → · find the founder's handle · → · read their last 30 days · → · pain stated? tier 1 · else tier 2, expires day 45

Each step is cheap before it's expensive. Finding the handle is the weakest link; fall back through the site's sameAs, footer links, the Show HN submitter, then a site search.

Ranking the queue: likely to reply this week

Detector catalog · launch sites and revenue boards — Detector catalog · test it by hand

vibedraft

What they say: the second trigger source.

Directories tell you when. Their posts tell you whether it hurts yet. The compound that matters: an elapsed-time phrase and a zero in the same post. Ranked by frequency × clarity:

1. "Launched N days/weeks ago" + "0 users" / "no signups" / "$0 MRR": the whole ballgame
2. "crickets" / "silence" / "went quiet" on a first-person launch
3. "nobody knows it exists"
4. "building was the easy part" / "marketing is a completely different ball game"
5. "I'm a builder, not a marketer": reliable on Reddit, not on X
6. "how do I get my first 10 users": frequent, weak alone, strong beside #1
7. Signups but zero revenue, stated together
8. "refreshing analytics": the purest tell
9. "traffic went back to baseline" / "losing momentum"
10. "distribution problem or a positioning problem": they've diagnosed themselves

"Should I give up" is real pain but marks the window's exit.

Nine ways the signal lies

vibedraft

Deriving category, ICP and segments from a brief.

The rule that makes it automatable: none of the three is read off what the founder says. The stated ICP is an input with the weakest evidence. All three are derived from what the product is uniquely good at and who is structurally forced to care.

alternatives · → · unique attributes · → · value · → · who is forced to care: the ICP · → · which triggers this buyer enters · → · segments, scored

Every ICP field carries its source and evidence

Every attribute must pass one test: can you query for it? "Cares about quality" isn't an attribute; "runs Next.js" is. Record the stated and evidenced ICP side by side; evidence overrules the stated version only above roughly five paying customers.

Segments come from triggers

Segments are generated by crossing the ICP with a fixed list of detectable states. The Sep 19 research named seven; this guide has since grown them into 24 triggers, cold and warm.

Score each with Moore's four pass/fail gates (reachable, acute, deliverable now, uncontested), then rank by referenceability, yield (under 20 a month is founder-led sales), speed to value and adjacency. Make the founder choose out loud between revenue first and proof first.

What the brief emits: one category sentence; one ICP with every attribute queryable and a stated-versus-evidenced note; three to five scored segments, each with its detection method, monthly yield and weakest gate; one recommendation with a kill date; and a widening ladder decided in advance.

Buying Moments Map · from a link to triggers — Brief Intake · evidence levels

reference

Vocabulary.

Watch the nesting: Winning by Design puts three to five segments inside the ICP (as this guide does: category, then ICP, then the starting segment); other writers use "segment" for the whole ICP.

reference

Across the guide.

reference

Sources.

• ICP method
• Dave Kellogg — Your ICP Starts as an Aspiration and Becomes a Regression
• April Dunford — A Quickstart Guide to Positioning
• April Dunford on positioning and segmentation — Lenny's Newsletter
• Crossing the Chasm — notes with page cites
• Geoffrey Moore on finding your beachhead — Lenny's Podcast
• Jason Cohen — Selling to Carol
• Lenny Rachitsky — How to identify your ICP
• Alex Hillman on Sales Safari — Lean B2B
• What is Sales Safari? — Stacking the Bricks EP42
• Bob Moesta — The Four Forces of Progress
• Craig Elias — SHiFT! Trigger Event Selling, preview chapter
• Bob Apollo — No Fit, No Deal
• Kyle Poyar — What it means to be focused on your ICP
• Winning by Design — ICP & Segmentation Blueprint
• Bowery Capital — Forging Your First ICP and Buyer Persona
• Unusual Ventures — What is an ideal customer profile?
• Clearbit — TAM, ICP, and Personas
• Rob Walling — Start Small, Stay Small (book report)
• ProfitWell / Paddle — Quantified Buyer Personas
• Named examples
• Nathan Barry — Direct Sales for Bootstrapped SaaS Startups
• ConvertKit's niche focus — Churn.fm
• PostHog — Who we build for
• PostHog — How we found our ICP
• Rahul Vohra — How Superhuman Built an Engine to Find PMF
• Jane Portman — Userlist positioning overhaul
• Tuple — $8,000 in revenue without even a product
• Bootstrapping Postmark — The Startup Foundry
• Clearbit — Why redefine an ICP
• Plausible — $400 to $2,750 MRR in 135 days
• Stripe — First 1000
• Jon Yongfook of Bannerbear — Indie Hackers Podcast #208
• Derrick Reimer on SavvyCal — Tuple Podcast
• Selling to Indie Hackers: Is It a Bad Idea?
• How the biggest consumer apps got their first 1,000 users
• Launch directories and trigger data
• HN Algolia API
• PeerPush API reference
• Product Hunt API rate limits
• TrustMRR API docs
• LaunchDirectories — 100+ launch sites
• Product Hunt traffic data 2026 — Causo Hub
• Product Hunt launch statistics — Shno
• The original research doc
• ICP and Trigger Detection (Sep 19, 2026)

What isn't verified

• The launch-lag shape is 26 dated posts from a five-week window, not a tracked cohort. The day-60 boundary is inferred from tone.
• Reddit endpoints couldn't be tested from the Sep 19 research environment.
• Exact Show HN volume, BetaList's sitemap and TrustMRR's MRR-band query parameters are unconfirmed.
• The ConvertKit quotes came through a fetch layer; re-open the source pages before quoting them publicly.
• Craig Elias's "60–90% close ratio" and ProfitWell's "1% pricing → 11% revenue" are self-published vendor claims.