Do Failed Startups Have Different Logos? We Checked 83 That Raised $100M+ and Shut Down

We built a verification pipeline specifically to keep the wrong logo out of this dataset — an entity check, a domain check, a human visual audit on a sample of whatever survived. The two errors that got through anyway, on the hardest cohort in this whole project, were photographs of cars. Both Fiskers, if you're wondering: a failed company stops maintaining its brand assets, and a stale automotive stock photo apparently reads as a "logo" to an automated check more often than you'd like.

That cohort is 83 companies that each raised at least $100 million and later shut down, measured with the identical pipeline as 168 companies that raised as heavily and are still standing. It exists to answer the one question that verification pipeline, and the rest of this colour-and-funding research, stopped short of: does any of this relate to whether a company survives? The comparison was pre-registered before the dead-company data existed — fix what counts as "included," fix the statistical test, then look.

Why this is the sharpest test available — and what it still can't answer

Every other comparison in this series measured heavily funded companies against the Fortune 500 — a different population entirely, not a control group. This one holds "raised a lot of venture money" constant across both arms and only varies the outcome. That's a real improvement. It still cannot tell you whether colour relates to getting funded in the first place, because both arms of this test already cleared that bar. Nothing in this piece closes that gap, and nothing could, without a comparison group of similarly-situated companies that never raised at all.

The result: a clean, well-powered null — with the caveat attached, not buried

The colour distribution of the dead group is statistically indistinguishable from the surviving group. Omnibus chi-square across all buckets: χ²=3.622, df=6, p=0.7277, Cramér's V=0.134 — a negligible effect size even before the p-value. Bucket by bucket: monochrome 34.3% dead vs 39.3% alive (p=0.5801); blue 25.7% vs 19.0% (p=0.3715); green/teal 20.0% vs 16.7% (p=0.6348); red/orange 11.4% vs 14.9% (p=0.5954). None of it clears significance. A second, independent measure — the brand mark rather than the filed logo — is also null (χ²=3.925, p=0.6868), and it disagrees with the first measure on which direction the (non-significant) blue gap even points, which is the signature of noise, not a hidden signal fighting through.

Here is the caveat that has to sit right beside that result, not below it: this test could reliably detect a 25.3-point gap on monochrome at 80% power. The gap actually observed was 5.0 points. Because the surviving arm is fixed at 168 companies, even an infinite dead cohort could only sharpen that detection floor to 10.5 points. This rules out a large colour-survival relationship. It does not rule out a modest one. "No large effect exists, and a modest one can't be excluded" is the honest sentence here — "colour provably doesn't matter" is not, and we're not going to write it just because it reads better.

Two confounds, checked honestly

One expected confound didn't show up: median founding year is 2013 in both the dead and the alive arms, and adjusting for founding era barely moves the result (the monochrome odds ratio shifts from 0.806 to 0.894). A confound we didn't expect did: the dead group raised a median $500M against the survivors' $1,620M — companies that eventually failed had, on average, raised less before they did. Matching both groups on peak valuation instead makes the null stronger, not weaker (monochrome 31.8% vs 38.0%, p=0.5864; omnibus p=0.8424) — which is the reassuring direction: the absence of a relationship isn't an artefact of comparing differently-sized companies.

The blue deficit isn't a survivorship artefact

This is the one place this piece changes how to read the rest of the series. If blue's scarcity among survivors (Piece 3) were somehow a byproduct of which companies happened to still be around, you'd expect the dead group to look more like the Fortune 500 on blue. It doesn't: the dead group runs 20.0% blue on the brand-mark measure — still well under the Fortune 500's 39.6%, and that specific gap is itself significant (p=0.0141). Blue's scarcity looks like a property of venture-backed companies broadly, alive or dead, not something that only shows up in whoever happened to win.

What "branding fashion, not a performance signal" means with this piece added

Put the series together and the picture is consistent: colour converges hard within a funded cohort (Piece 3's era trend), the convergence sits mostly off the wordmark rather than in it (Piece 2), and now — the closest thing to a causal test this data supports — colour shows no detectable relationship with which of the funded companies later failed. All three point the same way: a shared design-trend explanation fits the data better than any story where colour choice does something to outcomes.

The weakest leg, named directly

The dead cohort was harder to verify than the living one, in exactly the direction you'd predict: a failed company stops maintaining its brand assets, so residual error runs 3–8% on the canonical measure and 10–20% on the brand mark, against 1–3% for the funded-and-still-operating sample. The automated verification gates on this cohort scored precision of 1 in 6 and recall of 1 in 3 against a human visual check — including two logos that were, on inspection, photographs of cars. That's a materially higher error bar sitting under a null result than sits under this series' positive findings, and it's disclosed here rather than smoothed over.

What the four pieces jointly establish — and what stays out of reach

Together, this series can describe the colour patterns of a heavily funded population with real statistical care, and it can now say that colour shows no large relationship with which of those companies later failed. What it cannot do, and was never designed to do, is tell you whether colour relates to getting funded in the first place. That would need a fundamentally different study — a matched sample against a full registry of startups regardless of outcome, blind classification, or an experiment showing investors identical material differing only in logo colour. None of that is underway; naming it here is describing a gap, not promising to fill it.

If a null result needs a punchline, it's the one from the top of this piece. The discipline here was never only in the statistics. Some of it was in noticing the logo has wheels.

NUVC is a venture capital intelligence platform, and this series is the kind of evidence discipline we try to hold every number to — including, especially, the ones that come back null. Read how the verification pipeline itself was built if you want the full discipline behind these four pieces, not just this one's result.

Frequently asked questions

Does logo colour predict whether a startup will fail?

Not detectably, based on this test. Among 83 companies that raised at least $100M and later shut down, compared with 168 that raised heavily and are still operating, the colour distributions were statistically indistinguishable (omnibus p=0.7277). The test could reliably detect a large gap and didn't find one; a modest, harder-to-detect gap can't be ruled out.

Is a null result still a useful finding?

Yes, when it's well-powered and pre-registered, which this one is. A clean null here is evidence for the series' broader argument — that logo colour patterns among funded companies reflect shared branding fashion rather than anything that predicts outcomes — and it was reported in full regardless of which way it came out.

Does this mean logo colour has no effect on a startup's chances?

It means a large effect can be ruled out; a modest one cannot. It also still can't speak to whether colour relates to getting funded in the first place, since every company in both groups compared here had already raised substantial capital.