ICP Definition
A guided worksheet. Answer in order. You come out with your ICP locked and the raw material for every ad angle.
How to use this
Fill each blank for your company, in order. Under every block, four reference companies are filled in exactly as their own founder would have filled them, confidently, in good faith. Two of these companies won and two lost. You often won't be able to tell which from the answers alone.
That's the point: the answers look fine on the surface. The takeaway under each block is where the crack gets exposed: the subtle thing that separated the winner from the one that had more funding, more pedigree, or more hype and still lost.
Scope
For paid-or-free, single-user B2C software. Not marketplaces, hardware, or B2B. Tests whether your ICP is real, specific, and targetable, not whether the business survives.
Your four reference companies
Two matched pairs, same niche each.
Pair A: identical product (AI photo → calorie count)
ICP + marketing decided it
Cal AI ✅
What it is
AI calorie tracker. Snap a photo of your meal, get a fast estimate. Built for speed and low friction, not lab-grade precision.
Stats
- 15M+ downloads at acquisition.
- $30M+ annual revenue at acquisition (MyFitnessPal-quoted to press).
- ~$30/yr subscription. Bootstrapped; no VC raised pre-deal.
Where they are now
Acquired by MyFitnessPal (deal closed Dec 2025, announced Mar 2026). Still a standalone app, now tied into MyFitnessPal's food database.
SnapCalorie ❌
What it is
AI calorie tracker from ex-Google Lens researchers. Emphasizes accuracy, portion measurement, and detailed nutrition data.
Stats
- $2M seed round, Jun 2023 (Accel, Index Ventures, YC, Eric Roza).
- Total raised ~$2.3M–$4.3M depending on source.
- At launch (2023), ~1,000 new users/month reported. No public scale figures since.
SnapCalorie does not publish downloads, revenue, or MAU. Funding totals vary by database.
Where they are now
Still operating as an independent app (updated on app stores through 2026). Niche scale vs Cal AI. No acquisition announced.
Sources
- TechCrunch (Jun 26: 2023), $2M seed, launch traction
- CB Insights: funding history
- PitchBook: total raised (varies by database)
Pair B: near-identical product (paid/premium vs. free feeds)
Identity decided it
Deepstash ✅
What it is
Microlearning app. Bite-sized ideas from books, articles, podcasts, and video. Freemium: browse free, save unlimited with Pro.
Stats
- 10M+ users claimed on company site (self-reported).
- Pro ~$30–90/yr depending on plan and promo.
- ~$1M/month revenue (~$12M ARR run-rate).
Where they are now
Still operating independently (Brainstash, Inc.). Active on iOS and Android. No acquisition announced.
Quibi ❌
What it is
Premium mobile short-form video. Hollywood-produced episodes, ~10 minutes each, built for on-the-go viewing.
Stats
- ~$1.75B raised before launch.
- $4.99/mo with ads; $7.99/mo ad-free at launch (Apr 2020).
- Shut down Oct 2020, ~6 months after launch. Roku bought the content library for ~$100M.
Where they are now
Defunct. Service shut down Oct 2020. Content assets sold to Roku. Company wound down.
Read the fills first and see if you can spot the two that lost. Outcomes are revealed in the takeaways.
Magic Sentence: rough draft
“We are a ________________ that helps ________________ get ________________ by selling ________________ for ________________.”
Draft it rough, it's the clay the next questions sharpen. You rewrite it at the end.
Filled
Cal AI
SnapCalorie
Deepstash
Quibi
Takeaway
Who exactly has this problem?
“My person is a ________________ who ________________, and feels this problem when ________________.”
Paul Graham, YC: choose a small number of people who want it a lot, not a large number who sort of want it. Startup = Growth
Life-stage, situation, identity. Specificity gate: if your answer is “everyone,” you're either ChatGPT (you're not) or you haven't found your ICP. Narrow until it stings.
→ Ad angle: who you target.
Filled
Cal AI
SnapCalorie
Deepstash
Quibi
Public messaging (sourced)
The fills above mirror how each company positioned itself in the wild. These links are the proof we are not guessing at their messaging.
Cal AI
Effortless tracking / no more guessing (photo → calories, low friction).
SnapCalorie
More accurate than a nutritionist (precision, depth sensors, expert-backed data).
Deepstash
Bite-sized insights to grow / self-improvement (microlearning, smarter every day).
Quibi
Quick bites / on-the-go short-form (premium mobile episodes for spare moments).
Takeaway
Cal AI vs. SnapCalorie: same product, different framing. Cal AI sold the outcome; SnapCalorie sold the instrument. Everything else is downstream of that. Nobody wants calorie counts, they want to lose weight, and the thing that gets them there is a log they actually keep. Precision improves the instrument, effortlessness improves the outcome. SnapCalorie perfected the measurement and lost the result.
The reason this is deeper than market size: it's not that the effortless crowd is bigger, it's that “accurate” was never the mechanism. You can be the most accurate tracker in the world and still deliver nothing, because the failure mode in this category is abandonment, not error. Cal AI's angle attacked the actual point of failure. That's why the same product, aimed differently, produces a different business.
Deepstash vs. Quibi: same instinct, opposite outcome. Both picked a “bite-sized content for busy people” avatar, so on the surface they look like the same bet. The split is how the person is defined underneath. Deepstash chose a person with an ongoing identity: “self-improvement-minded, wants to grow, catches themselves doom-scrolling.” That's a person who has the problem every day and wants to become something, so the product has a reason to exist tomorrow. Quibi chose a person defined by a context: “a commuter with a spare ten minutes.” That's not who someone is, it's a situation they're occasionally in, and when the situation changes (lockdown erased the commute) the avatar evaporated. Deepstash picked a durable identity; Quibi picked a fragile circumstance.
The pattern across both pairs: the same product can be aimed at more than one person, and the choice decides the ceiling. Cal AI and SnapCalorie show it on mechanism: sell the fix for the real point of failure (abandonment), not a spec that sounds impressive but doesn't move the outcome. Deepstash and Quibi show it on durability: point at who the person is (an identity they carry daily), not a situation they're briefly in. When you write block 2, you're not guessing at a customer, you're choosing one from several valid framings. Find the real reason people fail at the outcome, and sell the fix for that. If your differentiator improves something other than the true point of failure, it's a spec, not a switch reason.
Why now?: the trigger
“They act right now because ________________ just happened.”
“Yeah, maybe I could see using something like that” summed across a population is zero users. This block screens for urgency, not politeness. How to Get Startup Ideas
The event that flips the problem from background annoyance to “I need this today.” A recurring trigger also proves the problem ranks high enough to pay for.
→ Ad angle: hook + timing.
Filled
Cal AI
SnapCalorie
Deepstash
Quibi
Takeaway
How are you solving it?
“We solve it by ________________, which kills ________________.”
One answer is enough. It must be something the product actually does.
Eugene Schwartz, Breakthrough Advertising: when a market has heard the claim before, the claim stops working, a new mechanism is what makes an old promise believable again. Your prospect has been promised weight loss a hundred times. What they haven't heard is how.
Answer from the problem in blocks 2–3, not from your competitors. This is not “why we're better”, that's the next block. This is: what does the product actually do that makes the outcome possible?
You do not need a laundry list of features. Lock the one mechanism you would lead with if you only got one shot, the clearest “how” that attacks the failure in block 2. Extra features can wait for later creative tests; they are not required to finish this block.
The pointing rule: you must be able to point at the feature, screen, or behaviour in the app. If you can't, it's a claim, not a mechanism. “It's smarter” is a claim. “It reads the label from a photo” is a mechanism.
The failure-point rule: the mechanism should attack the real reason people fail at this outcome (block 2). A mechanism that improves something other than the true point of failure is a spec: impressive, and it moves nobody.
→ Ad angle: the proof. Persona and angle say what you're claiming; the mechanism is why it's believable. One locked mechanism is your lead creative concept. If you later test another “how,” that is a second concept, not homework for this block.
Mechanism positioning
Desire × operational complexity
Products win when the mechanism creates strong desire without asking the user to absorb unnecessary complexity.
Cal AI
One-photo logging
Deepstash
Card-based learning
SnapCalorie
Sensor + human review
Quibi
Premium short episodes
Operational complexity →
Filled
Cal AI
SnapCalorie
Deepstash
Quibi
Takeaway
Run three tests on the line you wrote.
Can you point at it? All four companies pass. This is the low bar: it screens out slogans, nothing more.
Does it attack the real failure point? Here Pair A splits, and it's the same split block 2 flagged, now visible in the product itself. The failure mode in calorie tracking is abandonment, not error. Cal AI's mechanism removes a step. SnapCalorie's adds steps (depth scanning in service of accuracy) because they were solving error. Same technology category, opposite direction, and the direction was set by which failure they believed in. SnapCalorie's mechanism was real, demonstrable, and technically superior. It made the true problem worse.
Does the problem exist outside your product? This is where Quibi dies. “Short videos aren't cinematic enough” was not a problem anyone had. Nobody left TikTok because the production values were low. Turnstyle is a genuine feat of engineering answering a question nobody asked. When your mechanism solves a problem your market never reported, you've built the answer first and gone looking for the question.
Deepstash is the instructive pass. Its mechanism is weak against free on pure function (free feeds also give you short content), which is why block 7 (identity) has to carry the switch. But look at what the mechanism actually does: it replaces the scroll instead of competing with books. A mechanism that slots into an existing habit is what lets identity do its job. Identity alone, with no mechanism under it, is Quibi.
What to do with this answer: this is your lead ad concept's reason to believe. Hold persona and angle fixed and lead with this mechanism. That answer is portable: it tells you what to lead with on the landing page, in onboarding, and in your store listing. You can test alternate mechanisms later as separate concepts; you do not need them to lock the ICP.
One warning. A mechanism ad makes a promise the product has to keep. A mechanism that tests well and only half-delivers is the cleanest possible route to a great CAC in month one and an LTV that never arrives. Before you scale on a winning mechanism, check it against retention, not against CPA.
What are they using today: and why would they switch?
“Right now they use ________________. The exact moment this alternative fails them is when ________________, and they'd switch to us because we ________________.
And the reason they still might not: ________________.”
Christensen's Jobs to Be Done framing: what job are they firing their current solution for, and what pushes them to switch now?
List every alternative, including free ones and “nothing.” Be brutal: if the alternative is free and good enough and your only edge is “higher quality,” that's a feature, not a switch reason.
The last sentence is the purchase objection, and it is not the same as the problem. The problem is why their life is worse. The objection is what makes them close the app store page anyway. Write the one they'd actually say out loud, “it's another subscription,” “I've downloaded three of these already,” “I don't trust it to be right,” “I can get this free,” not the polite version. This is the only field on the sheet that produces mid-funnel material; everything else generates ads that assume the person already wants to buy.
→ Ad angle: “unlike X” positioning + mid-funnel objection.
Filled
Cal AI
“Right now they use MyFitnessPal, a paper log, or nothing. The exact moment this alternative fails them is meal three of day two, when a restaurant plate isn't in the database, they don't build the entry, and once one meal is missing the whole day's data is worthless, so they stop. They'd switch to us because we turn ten taps into one photo.”
“And the reason they still might not: they've already paid for a tracking app they stopped using, so the objection isn't price, it's ‘I'll quit this one too and waste another thirty dollars.’”
SnapCalorie
“Right now they use MyFitnessPal or other photo trackers. The exact moment this alternative fails them is when they catch a portion estimate that's obviously wrong, or two entries for the same food that disagree. They'd switch to us because depth sensors and human reviewers cut the error rate.”
“And the reason they still might not: accuracy is invisible before purchase, they can't verify the claim without using it, and the free tracker they already have gives them a number that looks fine.”
Deepstash
“Right now they use free social feeds, saved-article apps, and a stack of unfinished books. The exact moment this alternative fails them is closing Instagram after 40 minutes feeling worse than when they opened it. They'd switch to us because the same 40 minutes is curated for growth instead of algorithmic junk.” Weak vs. free.
“And the reason they still might not: the feed is free and already installed, and nothing about closing Instagram at midnight feels like a moment that requires a credit card.”
Quibi
“Right now they use TikTok and YouTube (free) and Netflix (paid). The exact moment this alternative fails them is hard to name. TikTok never runs out, Netflix is already downloaded, and neither leaves them stuck. They'd switch to us because Quibi is premium and finishable in one sitting.” Weak vs. free.
“And the reason they still might not: they have three streaming subscriptions already and no unmet need this would fill.”
Takeaway
Pair A: Cal AI switches you on effortlessness; SnapCalorie on accuracy: a spec you have to care about. “Less friction” moves millions; “lower error rate” moves a niche. Pair B: both Deepstash and Quibi are paid versions of something free feeds already do. The entire switch has to be carried by block 7 (identity).
Read the four objections in order and the pairs sort themselves again. Cal AI's objection is about the person's past failure, which their mechanism directly answers, so the ad can handle it. SnapCalorie's is about an unverifiable claim, which no creative can resolve before purchase. Deepstash's and Quibi's are the same objection, “free already works,” and only one of them has an identity strong enough to override it. If your objection is one your mechanism can't answer, you don't have a creative problem, you have a positioning problem.
What outcome do they want?: functional result
“In their words: 'I want to ________________ so that ________________.'”
In JTBD terms, the “so that” clause is the progress they're hiring you to make, not the feature they're buying. Christensen, HBR
→ Ad angle: benefit.
Filled
Cal AI
SnapCalorie
Deepstash
Quibi
Takeaway
Who do they become?: identity & emotion
“Once it works, they feel ________________ and become the kind of person who ________________.”
Emotional and social jobs, same JTBD lineage. Know Your Customer's Jobs to Be Done
When block 5 is weak against free, this is the only block that can carry the switch.
→ Ad angle: aspirational / status.
Filled
Cal AI
SnapCalorie
Deepstash
Quibi
Takeaway
ICP strength map
Switch reason × identity
Deepstash
WonIdentity carries the switch
Cal AI
WonStrong switch + strong identity
Quibi
LostNo switch reason to carry
SnapCalorie
LostFunctional switch, weak identity
Block 5 · switch strength →
Rewrite the Magic Sentence: locked
“We are a ________________ for ________________ who ________________, triggered by ________________, who want to become ________________, and who show it by ________________.”
The final clause is the habit signal: what does a keeper actually do in week 1 that a ghost doesn't? Answer from real users, or your own usage if pre-launch. This behaviour becomes your retention metric (Sean Ellis's PMF question, “how would you feel if you could no longer use this?”, is the parent of this idea). PMF survey
Using every answer above, rewrite the one-liner. This sentence is your ICP: every clause maps to an angle.
Filled
Cal AI
SnapCalorie
Deepstash
Quibi
Takeaway
Timing: the macro “why now”
“What changed in the world that makes this possible or necessary now, not five years ago: ________________ ? If nothing changed, why hasn't this already been won?”
Your personal trigger (block 3) can't see this; this block can.
Filled
Cal AI
SnapCalorie
Deepstash
Quibi
Takeaway
Verdict: how the four sorted
Hard pass
Specific person, emotional trigger, functional outcome, and identity all hold up. Block 5 is strong on function, or block 7 identity carries the switch.
- Cal AI ✅, won: effortless ICP + emotional trigger + distribution
- Deepstash ✅, won: weak block 5 but strong learner-identity carries it
The instructive losses
- SnapCalorie ❌, built avatar, trigger, outcome, and identity around accuracy instead of effortlessness. Lost to two teenagers.
- Quibi ❌, vague identity, assumed habit that never formed, fatal macro-timing.
The block 5 × block 7 map, which block your ads must carry:
- Strong 5 + strong 7 → dominant; lead with either. (Cal AI)
- Weak 5 + strong 7 → viable; ads must sell identity, not features. (Deepstash)
- Weak 5 + weak 7 → dead on arrival. (Quibi)
- Strong 5 + weak 7 → fine on function, but fragile, and if your "strength" is a spec the mass market doesn't feel (accuracy), it's weaker than it looks. (SnapCalorie)
Final verdict
Winners and instructive losses
Identity carries a weaker functional switch.
Deepstash
The clearest person, trigger, switch, and identity.
Cal AI
Accuracy is a specification, not the desired outcome.
SnapCalorie
No durable need, identity, or switching pressure.
Quibi
Note on sources: Company stats are linked on each reference card above. Block 2 messaging is linked directly under the fills (store listings, homepages, launch press). Deepstash revenue is estimated from Sensor Tower charts. Other block-level ICP answers are reconstructed to illustrate the framework: plausible founder-perspective fills, not quoted statements.
Filled example
What a completed worksheet looks like when every block is answered. This one is Cal AI, filled as a reference copy.
ICP worksheet · filled
Cal AI.html · saved locally
ICP worksheet · filled
Cal AI
AI calorie tracking · consumer health / weight loss · ~$30/year subscription
Filled this in as a reference example so you can see what a completed ICP looks like.
1. Magic sentence (rough)
2. Who exactly has this problem?
specific person, not everyone
3. Why now? (the trigger)
4. How are you solving it?
one is enough · must be a real feature you can show in three seconds
5. What are they using now, and why switch?
be honest about free options
6. What outcome do they want?
7. Who do they become?
the identity / feeling
8. Locked sentence (this is the ICP)
Revenue: ~$30/year subscription for AI-powered food logging and nutrition tracking.
The week-one keeper signal is simple: they actually use the camera to log meals repeatedly. A ghost might download the app, take one photo, look at the result, and disappear. A keeper photographs another meal the next day because the thing that previously made calorie tracking annoying has been removed.
Verdict: hard pass. Person, trigger, mechanism, outcome, and identity all hold up. Block 5 is strong because the mechanism directly answers the failure point of the alternative: manual logging friction. Block 7 is strong because the desired identity is durable and tied to a meaningful self-image: becoming someone who is disciplined and in control of their health. That puts Cal AI in the strong 5 + strong 7 quadrant: ads can lead with either the switch reason or the identity, with the photo-to-calories mechanism providing the proof.
Completed example · Cal AI. For reference, don't edit this one.
Grade your own sheet
You've now got a filled worksheet and a finished example to compare against. Before you build anything on top of it, pressure-test it. Copy the prompt below, paste your completed sheet underneath it, and send it to whichever LLM you use. It will score each block, name the weakest one, and then ask you questions to push it deeper. Don't let it rewrite your answers for you. The point is that you end up with a sheet you can defend, not one an AI wrote.
Grading prompt
You are grading an ICP worksheet for a consumer software product. You are a demanding
paid-acquisition strategist, not an encouraging coach. Your job is to find the weakest
answers and expose them, because a vague ICP produces expensive ads.
Do not compliment the work. Do not soften findings. If an answer is generic, say so
plainly and explain what makes it generic.
Below is a completed worksheet. Grade each block against the rules for that block.
BLOCK RULES
Block 1: Magic sentence (rough)
Low bar. Passes if it names a product type, a person, an outcome, and how it makes money.
Do not praise a clean answer here; it means almost nothing.
Block 2: Who exactly has this problem?
Must be a specific person, not a demographic and not "everyone." Age ranges and income
brackets are not answers. The strongest version names a person defined by an ongoing
identity or a repeated failure, not by a temporary situation. Ask: if this person's
circumstances changed next month, would they still be this person? If not, the ICP is
built on something fragile.
Block 3: Why now? (the trigger)
Must be an event that already happened, not a state of wanting. Score it on three things:
is it emotional or merely rational; does it recur or fire once; and where is the person
when it fires, mid-scroll, or in a browser comparing options? A rational, one-time,
search-intent trigger is weak for paid social no matter how real it is.
Block 4: How are you solving it (the mechanism)
Two tests. The pointing test: can you point at a feature, screen, or behaviour in the
product? "It's smarter" fails; "it reads the label from a photo" passes. The failure-point
test: does this mechanism attack the actual reason people fail at this outcome, as named
in block 2? A mechanism that improves something else is a spec, it may be technically
impressive and still move nobody. Also check whether the mechanism adds steps or removes
them, and whether it can be shown in about three seconds.
Block 5: What are they using now, why switch, and why they might not
Three parts. The alternatives must include free options and "nothing." The failure moment
must be a specific moment, not a general dissatisfaction, if it can't be named, there is
no switch to win. The objection must be the real purchase objection stated bluntly, not
the polite version, and you should check whether the mechanism in block 4 is capable of
answering it.
Block 6: Functional outcome
The "so that" clause must land outside the product, in the person's life. If it ends on
using the product, sticking with it, or having better data, it has mistaken the means for
the end. This is the most commonly failed block.
Block 7: Identity and emotion
Must be an identity someone would want to be and would pay to keep being. Test for
durability: does it depend on something the company must continuously supply, or on
circumstances that could disappear? A description of a state ("someone with accurate data")
is not an identity.
Block 8: Locked sentence
Must be consistent with blocks 2 through 7, flag any clause that contradicts an earlier
answer. The final clause is a week-1 habit signal and must describe something the user
does, not something their schedule or circumstances do.
Timing: macro why now
Check for two unlocks, not one: did something change that makes the product possible, and
did something change in how you reach people? Most answers only have the first.
OUTPUT, IN THIS ORDER
1. A score out of 5 for each block, each with one sentence naming the specific weakness.
No block gets a 5 for being well-written; it gets a 5 for being specific and defensible.
2. The verdict grid: is block 5 strong or weak, is block 7 strong or weak, and which
quadrant does this product land in:
strong 5 + strong 7 = lead with either
weak 5 + strong 7 = ads must sell identity, function won't carry it
strong 5 + weak 7 = works on function but fragile
weak 5 + weak 7 = the ICP is not viable as written
3. The single weakest block, and why fixing it changes the most.
4. The contradiction check: any place where two blocks disagree with each other, or where
a later block quietly widens a person the earlier block narrowed.
5. Then stop and ask me ONE question aimed at the weakest block. Wait for my answer before
asking the next. Keep going one question at a time until that block would score a 4 or 5,
then move to the next weakest.
Do not rewrite my answers for me. Ask questions that make me write a better one myself.
If I give you a vague answer, say it's vague and ask again.
Here is my completed worksheet:
[PASTE YOUR SHEET HERE]