Run the idea filter before you hire the team

The problem. Every other play on this site assumes you’ve picked the business. If you haven’t (you’re weighing a product idea against the agency you already run, or three ideas against each other), the expensive mistake is finding out slowly that you chose badly. Six months of building is a terrible way to learn what an honest hour of filtering would have told you up front.

The play. Run the idea through four gates and two human checks, in order, and force one of three verdicts: Keep, Bench, or Scrap. The gates test the idea. The checks test you. Claude does the legwork; the structure keeps it from flattering you. This is the one play that belongs to the CEO alone. It runs before the team exists, because staffing Growth, Ops, and Brand around the wrong idea just gets you to the wrong place faster.

The four gates.

1. Proven demand. Competition validates demand; it doesn’t kill the idea. The uncontested idea is the risky kind, because an empty market usually means nobody’s paying. The scan’s job is to find the opening: a weak incumbent, an overpriced one, or an advantage you hold that others don’t. This gate fails only when there’s no opening anywhere. Then add the headroom question: is the market big enough that a small share is a real business? A crude street test helps: could a stranger you pass on the sidewalk plausibly be a customer? The more yeses, the more room you have to execute imperfectly and still win.

2. A real wedge. Take your name and your relationships off the idea. If the advantage disappears with them, it’s a service, not a product: the value is you, and you don’t scale. (A service can be a great business. It’s a different game, priced and staffed differently, and the gate’s job is to name which one you’re in.) Then pressure-test what’s left: is the wedge already table stakes? Could a platform ship it free next quarter? And in a crowded category, is the wedge visible at the moment of choice? At low price points buyers pick from feature lists without research, so a better product loses to lookalikes it can’t out-shout.

3. A hard validation path. Validation counts only when it costs the validator something. Free yeses are fake: friends, communities, and anyone rooting for you will say yes to a pitch that costs them nothing. The gate asks whether a cheap test exists where a real buyer pays real money early: a pre-sale, a deposit, a paid pilot. If the only evidence available is applause, the idea can’t clear this gate yet.

4. Lean on capital, leveraged on time. This gate is two tests, and passing one can hide a failure on the other. A business can be cheap to start and still trap you in per-unit labor with a low ceiling, where every sale costs you hours forever. Check four things: cost of goods near zero (per-use costs compound for years); no per-unit labor trap; a marketing channel you can work yourself from day one, not one that needs an audience you don’t have; and rails you own — if the core of the idea lives on someone else’s platform or API, they can end it.

The two human checks. The gates can all pass and the idea still be wrong, because the idea isn’t the scarce resource. You are.

• Mental space: what do you stop doing? A new idea doesn’t get spare capacity; it takes capacity from something you’re already running. Name the thing that shrinks.
• Endurance: year three. Not “am I excited today.” Will you still want this in year three of slow, linear growth? Most real businesses have a long, unglamorous middle.

One more question rides with these: who builds it? Is the capability held by you, today, or hoped for? An idea whose wedge rests on a skill nobody on hand has is a plan to acquire a skill, not a business. And all else equal, prefer ideas adjacent to what you already run: you know the customers, the vendors, and the failure modes, and half the scan is done before you start.

The scorecard. Copy this into a note and fill it in as you go, or hand it to Claude with the prompt below.

IDEA (one sentence):

Scan done? (what exists, who's funded, who's beatable, how crowded):

Gate 1, Proven demand:        PASS / FAIL  ->
  - competition exists (someone is paying for this)?
  - headroom (the street test)?
Gate 2, Real wedge:           PASS / FAIL  ->
  - survives with your name off it (product, not service)?
  - not commoditized / not about to be shipped free by a platform?
  - visible at the moment of choice?
Gate 3, Hard validation path: PASS / FAIL  ->
  - a cheap test a real buyer pays for (pre-sale / deposit / paid pilot)?
Gate 4, Lean on capital:      PASS / FAIL  ->
Gate 4, Leveraged on time:    PASS / FAIL  ->
  - cost of goods near zero?
  - no per-unit labor trap?
  - a channel you can work from day one?
  - rails you own?

Check 5, Mental space (what you stop doing):
Check 6, Endurance (year three of slow growth):
Who builds it (capability held, not hoped for)?
Adjacent to what you already run?

VERDICT:  KEEP / BENCH / SCRAP
One sentence why:
If BENCH, the one condition that would un-bench it:
If KEEP, the one-month momentum check (date):

Setup (5 min). A fresh Claude conversation is enough; no Project or files required. Bring the idea in one sentence, and be ready to describe what you already run; the filter reads differently for an operator with an existing business than for someone starting from zero. Claude can run the competitive scan itself where web search is available on your plan and surface (check current docs); if it isn’t, do the searching yourself and paste in what you find.

The prompt:

I'm an operator. Here's what I already run: [your business in a sentence or
two, or "nothing yet — this would be the first thing"].

Here's the idea: [one sentence].

Run it through this filter, in order. Be blunt. Help me reach my own
conclusion instead of flattering the idea.

1. Proven demand. Scan the market: what exists, who's funded, who's
   beatable, and how crowded the category is (count them). Competition
   validates demand — the uncontested idea is the risky kind. Where's the
   opening: a weak incumbent, an overpriced one, or an advantage I hold?
   And is there headroom — is the market big enough that a small share is
   a real business?
2. Real wedge. Take my name and relationships off the idea and tell me what
   advantage is left. If it disappears, say plainly: this is a service, not
   a product. Then test what remains: already table stakes? Likely to be
   shipped free by a platform? Visible at the moment a buyer picks from a
   crowded list?
3. Hard validation path. Name the cheapest test where a real buyer pays
   real money (pre-sale, deposit, paid pilot). If the only available
   evidence is free yeses, fail this gate and say so.
4. Lean and leveraged. Cost of goods near zero? Any per-unit labor trap?
   A marketing channel I can work myself from day one? Do I own the rails,
   or does someone else's platform?

Then ask me the human checks — don't answer them for me: what do I stop
doing to make room for this, will I still want it in year three of slow
growth, and who builds it (do I hold the capability today)?

End with one verdict — Keep, Bench, or Scrap — and the reasoning in a few
sentences, not a framework dump. If Bench, name the single condition that
would un-bench it. If Keep, set a momentum check one month out.

What you get. A verdict with reasons attached, and usually one uncomfortable finding you’d been stepping around — the gate the idea really hinges on. Claude isn’t there to decide. A fixed sequence of gates makes you answer the questions in the order that kills bad ideas cheapest, instead of the order that protects your favorite.

The verdicts. Keep means start, and start the clock: within a month of taking the idea to real buyers (conversations, the paid test), something concrete should move. If nothing does, that’s your answer: kill it and be glad the lesson was cheap. Bench means not now, and the rule is strict: a Bench must name the one condition that would un-bench it (a platform opens up, a capability gets hired, an incumbent exits). A Bench with no named condition is a Scrap you’re not ready to admit. Scrap means write one sentence on which gate killed it, and let it go.

What to watch for.

• Expect Scrap. A high kill rate is the filter working, not you failing. Good openings are scarcer than they were a decade ago, and the right response to that is a higher bar, not a lower one.
• Claude wants to be encouraging. The “be blunt, don’t flatter” lines matter. If every gate passes on the first run, run it again and ask for the strongest case against.
• “Service, not a product” is a finding, not an insult. Plenty of operators reading this run services on purpose. The gate names which game you’re playing so you price it and staff it honestly; it doesn’t talk you out of it.
• One beatable incumbent isn’t an opening if forty lookalikes sit between you and the buyer. Make the scan count the category, not just find its weakest member.

Make it repeatable. Keep a running list of verdicts: idea, verdict, the sentence why. Over time it becomes a calibration log: when new information moves an idea between columns, note what the filter missed. Re-run the filter on a real trigger, not a schedule; the natural one is a current business hitting its plateau and you deciding what to build next. And when the verdict is Keep, the next play is hiring the team to run it: the roadmap starts at Week 1.