By the third product idea, most sellers stop being able to trust their own judgment. The first idea gets a week of careful research. The third gets an evening. The fifth gets a scroll through TikTok and a feeling in the stomach, and by then the feeling is contaminated by everything the first four ideas taught you to hope for.
The fix is not more research. It is comparable research — every candidate pushed through the same questions, in the same order, weighted the same way, so that a 58 scored in March means the same thing as a 58 scored in June. That is all a scorecard is.
What follows is the rubric I would hand a systematic seller: four axes, fixed weights, scoring anchors, and thresholds tied to specific actions. It is the written-down version of the field judgment scattered across the rest of these notes, assembled into something you can actually run an idea backlog through.
Why score at all
Gut feel works fine for one idea. It starts failing around idea number three, because human memory does not store evidence — it stores impressions, and impressions bleed into each other. You end up comparing this week's candidate against a flattering memory of last month's.
A consistent rubric fixes three things at once. It makes candidates comparable across time. It exposes weak evidence, because a blank cell in a scorecard is loud in a way that a vague sense of "the Reddit signal seemed fine" never is. And it speeds up the only decision that matters early: kill or continue.
One caution before the math. A score is a decision aid, not an oracle. A 74 does not promise the product will sell — it says the evidence justifies the next small commitment of money and attention. Nothing about a number changes what the underlying signals can and cannot tell you.
The four axes and their weights
The scorecard grades four things, weighted by how directly each one predicts a new entrant's ability to sell — not how easy each is to measure.
- Demand direction — 30 points. The trend sits under everything else. Selling into rising demand forgives mistakes; selling into decline punishes even good execution.
- Competitive pressure — 25 points. Demand you cannot reach through the wall of incumbents is not your demand.
- Retail proof — 25 points. Evidence that money is actually changing hands, not just that people are curious.
- Community pull — 20 points. The leading indicator of unmet want. Weighted lowest because it is the noisiest of the four.
Adjust these honestly for your model, once, before you score anything. Pure dropshipping lives and dies on paid traffic, so competitive pressure deserves more weight there — 30 points, taken from community. Brand-building is the reverse: community pull is where a brand's early customers come from, so it can carry 30 while competition drops. The one rule is that you set the weights before you look at any specific candidate, write them down, and do not touch them mid-batch.
The rubric, axis by axis
Score each axis 0 to 10, then multiply: demand by 3, competition by 2.5, retail proof by 2.5, community by 2. Maximum 100.
Demand direction — 30 points
Pull the twelve-month picture for the category's plainest search terms at trends.google.com. You are grading direction, not volume.
Rising steadily over twelve months scores 8 to 10. Flat but stable scores 5 to 7 — flat is not failure; stable demand is a perfectly good market for a new entrant. Declining scores 0 to 3, and almost nothing rescues it.
Then apply a stability modifier. Subtract a point or two for wild week-to-week volatility, which usually means the interest is event-driven rather than durable, and for hard seasonality you have no plan for. The full method — including the ways a trend chart lies to you — is in the Google Trends field note.
Competitive pressure — 25 points
Search the category in the Meta ad library and read four things: how many distinct advertisers are running, how old their ads are, how similar their creative is, and whether the copy leads with price cuts.
The point of scoring these together is that healthy incumbency and gold-rush churn look completely different, even at the same competitor count. A handful of advertisers whose ads have run for months, with varied angles and no discount panic, is a proven market that still has room — score it 6 to 8. Dozens of advertisers who all appeared in the last few weeks, running near-identical creative and racing each other to the bottom on price, is a gold rush already eating its own margin — score it 2 to 3.
The field heuristic doing most of the work here: an ad that has run 30 or more days is paying for itself, because nobody funds a loser for a month. Zero advertisers is ambiguous — either demand is unproven or you are genuinely early — so score it 4 to 5 and lean harder on the other axes. The longer treatment of why raw competitor counts mislead is in the saturation note.
Retail proof — 25 points
Amazon's first page tells you whether the demand converts. Two reads: review velocity and page shape.
Velocity first. Are recent reviews still accumulating month over month on the top listings, or did they pile up two years ago and stop? Fresh reviews across multiple listings mean live, current buying.
Then the shape of the page. Several mid-sized listings, each with steady recent reviews and none dominant, scores 7 to 9 — the market is buying and no one owns it. One giant listing with tens of thousands of reviews and nothing behind it scores 3 to 5, because winner-take-most categories are brutal on entrants. No meaningful listings at all scores 2 to 3 until other evidence says otherwise. The mechanics of reading rank and review velocity are their own note.
Community pull — 20 points
On Reddit, TikTok, and X, grade three things about the problem your product solves. Recurrence: does the same complaint show up across months and across unrelated threads, or is it one viral post echoing? Intensity: are people describing workarounds they have rigged up, which is what genuine frustration looks like in the wild? And purchase language: unprompted phrases like "where can I buy this" or "take my money" are the strongest signal this axis produces.
All three present scores 8 to 10. Recurring complaints without purchase language score 5 to 7 — a real problem people may not pay to solve. A topic people discuss but never complain about scores 0 to 3. The listening method is laid out in the Reddit field note.
Thresholds that mean something behaviorally
A score only earns its keep if each band triggers a different action. Otherwise you have produced a number and kept your indecision.
- Under 40: archive with a note. Not delete — archive, with the date and the axis that killed it. Markets change, and a dated note is what lets you re-score in six months without starting over.
- 40 to 70: investigate the weakest axis specifically. Do not average the weakness away. A 55 built on four mediocre axes is a genuinely mediocre market; a 55 hiding a 2-out-of-10 on demand direction is a dying market with nice ads. The follow-up work goes at the lowest number, nowhere else.
- Over 70: proceed to a small paid test. Modest budget, one creative angle, a real page. The score justified the next commitment; it did not replace it.
This is why the per-axis breakdown matters more than the headline number. Two ideas can both score 60 and deserve opposite decisions.
A worked scoring, shown honestly
Here is the math run end to end on a real category shape, anonymized. This is a demonstration of method — not a case study, and not a claimed result.
The candidate: a mid-priced kitchen tool addressing a cleanup annoyance.
- Demand direction: twelve-month trend flat with a slight upward drift and a mild December bump. Stable, unexciting, no volatility penalty. 6 out of 10 × 3 = 18.
- Competitive pressure: eight distinct advertisers in the ad library. Two have run the same creative for over 60 days — healthy incumbency. But four appeared within the last month, converging on the same demo-video format and leading with discounts. That mix reads as early churn forming. 4 out of 10 × 2.5 = 10.
- Retail proof: five or six listings on the first page with recent reviews still accumulating on most, and no single listing dominant. The healthiest axis of the four. 7 out of 10 × 2.5 = 17.5.
- Community pull: the complaint recurs across months in cooking and cleaning threads, with a few improvised workarounds described, but almost no purchase language. 6 out of 10 × 2 = 12.
Total: 57.5. Middle band. The threshold rule says the next move is not "launch" and not "archive" — it is to interrogate the weakest axis, competitive pressure. Concretely: watch the ad library for two weeks and see whether those four new advertisers persist past the 30-day mark or vanish. If they vanish, the churn read was right and the score softens further. If they persist and diversify their creative, the axis re-scores upward and the total may clear 70 honestly.
The automated version
Once you have run this rubric a few times, the scoring becomes mechanical — which is exactly the argument for automating it. This is the shape of what Kestrel computes: it checks live competitor ads in the Meta ad library, Google search demand, Amazon retail proof, and community chatter, then returns a 0–100 market score with the per-signal evidence attached and a Hot / Promising / Weak verdict. A full public example of a scored report lives at /specimen, and the free tier's 20 scans — no card — are enough to grade an entire idea backlog in an afternoon. The manual method above still matters, because it is how you audit any score, including Kestrel's.
Keeping yourself honest
The scorecard has one enemy, and it is not bad data. It is you, after you have fallen for an idea.
So pre-register your threshold before you score anything. Write down "I test anything over 70 and archive anything under 40" while the backlog is still a list of strangers. The characteristic failure of every scoring system is re-weighting the axes until the pet idea wins — and given enough freedom, there is always some weighting under which your favorite clears the bar. If you become convinced mid-batch that a weight is genuinely wrong, finish the batch, change the weight, and re-score everything under the new one. Never just the idea you were hoping to save.
Expect most candidates to score under 40. That is not the scorecard failing; that is the scorecard working, because most product ideas should die on paper before they can die with your money attached. The archive notes compound — six months of dated kill-reasons will teach you more about reading markets than any single launch. The whole apparatus only earns its keep if it is allowed to say no, and that discipline holds whether the math is yours or a scan's.
