44,114 parsed. 12,244 graded. Six cleared the top tier. A look at where startup ideas die, and why the most common failure mode isn't the one people warn you about.
44,114 SaaS ideas parsed. 12,244 graded. 1,350 reached A-tier or above. Six cleared the top tier.
Most of what failed didn't fail for the reason people on the internet usually give. That's the headline of this writeup, and the rest is the data behind it.
We pulled raw signal from Reddit startup communities, Hacker News, YouTube transcripts, GitHub issues, and the Y Combinator company database. After deduplication and structuring, 44,114 SaaS ideas were parsed into the pipeline.
Of those, 12,244 made it through grading. The other 31,870 got filtered before reaching a tier: duplicates, low-signal noise, or signal too ambiguous for a heuristic pass to resolve.
The graded ideas distributed like this:
The lone C-tier idea is a rubric artifact, not a meaningful category. Read C and D together: below the cutoff, but not actively bad.
Six stages. Each one filters or annotates.
The competition data was the most surprising thing in the set.
I want to sit on that number for a second. The most common reason people on Reddit dismiss a startup idea is some version of "the market is too crowded." In our data, saturation is by far the rarest competition state. If you're getting that comment on your idea, the base rate says the commenter is probably wrong.
The far more common picture is the opposite: 62% of graded ideas sat in markets with real demand and low competition. A casual read of that number says it's an enormous opportunity pool.
A careful read says it's a trap.
The verdict distribution from the AI deep analysis stage:
So there are two facts in tension. Most graded ideas live in underserved markets. The most common verdict is that the idea is too niche to matter.
That's not a contradiction. It's the failure mode underneath most of the dead bulk in the distribution: real problems people genuinely have, with no competition because there isn't a business at the bottom of the problem. The problem is real. The population with the problem either can't pay enough or won't pay enough to support a company.
The market size dimension confirms it.
If you go to Reddit looking for an unsolved problem, you will find one in five minutes. Finding an unsolved problem with a fundable market underneath it is a different search, and a much harder one.
This is the asymmetry worth internalizing. Demand is cheap. Buyers are expensive.
The 1,350 ideas at A-tier or above don't distribute evenly. Two industries and two categories carry most of the strong matches.
Education shows up because dissatisfaction with the status quo is perennial and willingness to pay is well-established. Developer tools shows up because builders complain about their own toolchain in public, in detail, often with the exact specification of what would fix it. Productivity shows up because the surface area is broad, integration expectations are low, and time to monetization is short. Marketplaces show up because the demand-side signal in our sources is loud, even though execution is structurally hard.
If you're scanning communities yourself, these four areas have the highest signal-to-noise ratio. The D-tier counts in each are also high, which is the same point: lots of activity, more real opportunities than other surfaces, and also more noise.
Six S-tier ideas across 44,114 parsed. 0.014% of parsed. 0.05% of graded.
There's no flourish to add to that. The bar is high, and it's meant to be high. The 1,344 A-tier ideas are where most usable signal lives: credible opportunities that pass every gate without major flaws. S-tier is reserved for ideas that pass every gate cleanly, including the devil's advocate, with no fixable flaws flagged and competitor research showing genuine open territory.
The Y Combinator partner trope says "one in a hundred ideas is real." Our dataset says the ratio is closer to one in nine if you mean "credibly fundable" (A-tier or above) and roughly one in two thousand if you mean "exceptional" (S-tier among graded). The first number is more generous than the conventional wisdom. The second is harsher by orders of magnitude.
A few honest caveats before the takeaway.
The grading reflects the rubric's view at the moment we ran it. Markets shift. An idea verdicted "Too Niche" in 2025 may become viable in 2027 if AI changes the unit economics of serving small populations. A few of those 4,995 too-niche calls will age into Worth Exploring. We have no way to tell you which.
The pipeline doesn't measure execution risk well. Two A-tier ideas can have the same opportunity score and wildly different difficulty curves. The business plan stage tries to surface this through MVP scope and action-plan length, but a tier label flattens it.
Our sources are biased toward English-speaking, software-literate communities. Reddit, Hacker News, YouTube transcripts, GitHub, and YC tilt the inventory toward problems that builders themselves notice. The data is less reliable for ideas that depend on offline workflows, regulated industries, or non-Western markets.
These limits don't change the headline finding. They do mean the headline is about this dataset, on this rubric, at this point in time.
A few takeaways, if you're using this kind of data to decide what to build.
The bottom of the distribution isn't badly-evaluated good ideas. It's real ideas that genuinely shouldn't get built. Roughly 80% of graded ideas landed in D or F because the underlying opportunity wasn't strong enough. The grading isn't being harsh. It's exposing the shape of the field.
The conventional wisdom about saturation is roughly inverted. True saturation applied to 29 of 12,244 ideas. The "Too Niche" verdict applied to 4,995, nearly half. If you ignore Reddit-style "this market is too crowded" comments and pay close attention to market sizing instead, you avoid the most common failure mode.
The head of the distribution is small. 303 ideas got the Strong Opportunity verdict. 1,344 reached A-tier. Six reached S-tier. Anyone serious about finding a real one needs a careful filter or a lot of patience.
---
PerfectIdeas matches builders to ideas that survived the grading pipeline. See your matches.