Are Startup Logos Really Black and White? We Measured 499 of Them — and Corrected Ourselves

One match in our verification pass sailed through every mechanical check we had. The label matched: "Coalition." The entity type matched: an organisation. There was no company domain on file to check against, so the domain gate had nothing to reject. It was Coalition Avenir Québec — a Quebec provincial political party. Nobody caught it with a smarter query; somebody caught it by rendering the logo and looking at it. That's the whole methods argument in one sentence: automated checks catch what they're built to catch, and the thing they miss looks fine on paper.

Why isn't "logo colour" just one number?

A company doesn't have one logo — it has a wordmark filed somewhere public and a mark rendered on its homepage today, and they don't always agree. We measured two independently: M1, the canonical Wikidata logo (168 of 499, 33.7%), and M4, the brand mark served from the company's own domain (404, 81.0%). Most published "logo colour" content measures only one — usually whichever is easiest to scrape. Which one you pick changes what you can claim.

The verification gate

Of 499 companies sourced, 196 had any logo recorded on Wikidata at all (230 had none; 73 matched no entity). Of those 196, an entity check passed 174 and rejected 22 — sub-brand suspects, class mismatches, and name collisions like "Snowflake" resolving to Tor's anti-censorship tool instead of the cloud-data company. A domain check then split what remained: 103 matched cleanly, 10 mismatched outright, and 82 had no company domain on file to check against at all — the exact condition that let Coalition through, above. What survived every gate into the analysis: 168.

A separate pass, for the companion piece to this one, caught a different failure entirely: five companies' recorded "brand colour" all extracted as the identical hex code #0000EE — a browser's default colour for an unstyled link, not a brand choice. The scraper had read raw anchor tags on pages where a stylesheet hadn't loaded, and it surfaced only because the same implausible hex kept repeating across unrelated companies.

After the entity gates, we rendered a sample and looked at it by eye — the residual wrong-logo rate that produced is in the FAQ below, printed rather than rounded away.

A rule we wrote before we had data to test it on

The tempting move is to drop the companies whose scale dominates a sample, because they're "obviously" distorting it. But "obviously distorting" is a judgment made after seeing the result, which contaminates the trim with the answer you wanted. So we fixed a mechanical rule first: exclude any company holding more than 5% of total sample capital. That excludes exactly two — OpenAI and Anthropic — with the next-largest, Waymo, at 4.4% and staying in. Applied blind to the earlier, smaller dataset, the same rule selects the same two we'd have flagged by hand. That's what pre-registration buys: a trim you can defend.

The concentration is worth seeing plainly. Across every company carrying a capital figure ($595.8B total), the largest holds 30.2%; the top two hold 52.4%; the top five hold 64.9%; the top ten hold 75.5%. Under the trim, monochrome's share of capital collapses from 72.5% to 42.3% — roughly where its share of company count already sat. The dollar-weighted "monochrome dominates" story was substantially a story about two companies.

What actually held up

Here is the full count-weighted distribution on the canonical-logo measure (M1, n=168), benchmarked against a first-hand-verified Fortune 500 colour count:

Colour bucket	Companies	Share	95% CI
Monochrome (black/white)	66	39.3%	32.2–46.8%
Blue	32	19.0%	13.8–25.7%
Green/teal	28	16.7%	11.8–23.0%
Red/orange/pink	25	14.9%	10.3–21.0%
Multicolour	8	4.8%	2.4–9.1%
Yellow	5	3.0%	1.3–6.8%
Purple	4	2.4%	0.9–6.0%

Bucket	This sample (M1)	Fortune 500	p-value	Verdict
Blue	19.0%	39.6%	p<0.0001	Significant
Green/teal	16.7%	7.0%	p=0.0002	Significant
Monochrome	39.3%	30.6%	p=0.0380	Significant, but see below
Red/orange	14.9%	18.0%	p=0.3542	No difference

Two findings survive on both instruments and the pre-registered trim: blue is markedly scarce in this population (M4 replication: 24.3% vs 39.6%, still p<0.0001), and green/teal is markedly common. Those are the two most trustworthy numbers in the whole study.

The monochrome finding is not one of them. On the brand-mark measure — the higher-coverage instrument, 404 companies instead of 168 — monochrome sits at 30.4% against a Fortune 500 rate of 30.6%: a dead null, z=-0.05, p=0.9600. So the same claim is "significant" on one instrument and indistinguishable from the Fortune 500 on the other, and we don't know which instrument to trust more. We're reporting both, not picking the one with the better headline. Part of why: on the 141 companies where both measures are available, they agree on only 103 — 73%. That 27% disagreement is concentrated exactly where it hurts most.

One finding moved the other direction. A smaller sample last time hedged the founding-era trend — recently founded companies skewing more monochrome — as merely suggestive, at p=0.061. At this size it clears conventional significance: 27.8% monochrome among companies founded 2011 or earlier (15 of 54) versus 59.3% founded 2017 or later (16 of 27), z=+2.75, p=0.0060.

What verification buys you, and what it doesn't

Everything above makes the numbers trustworthy: we know roughly how often we're wrong, we fixed our exclusion rule before we could tune it, and we're showing the result that disagrees with our own prior headline. None of that makes the comparison causal. Every company here already raised enormous sums — there is no group of similar startups that tried a colour and didn't raise. Colour is chosen, not assigned at random, and rebrands typically follow a big raise rather than precede it. The Fortune 500 is a reference point, not a control group.

The weakest leg of this piece

This study cannot tell you whether the canonical-logo or the brand-mark measure is more trustworthy, and they disagree on the one finding — monochrome — that would make the best headline if true. The residual wrong-logo rate is a sample estimate, not a certainty, and roughly two in five passing canonical-logo rows have no company domain to corroborate against — the condition that let the Coalition case through. Coverage isn't random either: companies where a canonical logo resolves skew 35.5% monochrome; companies it doesn't skew 27.8%. That gap is the likeliest reason a wider frame moved our own headline, and whatever bias remains likely overstates monochrome.

Where this goes next

The next two pieces take the findings that did hold up — the blue deficit and the green/teal excess — to founders deciding what a logo needs to do, and investors reading branding convergence as a market signal. A fourth runs the sharpest test this corpus can support — whether colour relates to which funded companies later shut down — on a cohort whose own residual error was worse than this one's, for the reason you'd expect: nobody keeps a dead company's brand assets tidy.

If there's a punchline here, it's this: the two things that nearly slipped through this pipeline were a Quebec political party and five companies whose "brand colour" was an unstyled hyperlink. Most of what passes for data journalism is fact-checking with better typography.

NUVC is a venture capital intelligence platform built on the same discipline: verify before you publish, and say plainly when a number doesn't survive a second look. See how we hold our own scoring to that standard — it's the same evidence discipline we think should sit behind a fundraising or investment decision.

Frequently asked questions

How was startup logo colour measured, and how accurate is it?

Two independent measures — the canonical Wikidata logo (168 of 499 companies) and the brand mark served from the company's own live domain (404 of 499) — each passed a domain check, an entity check, and a human visual audit on a sample. That audit put the residual wrong-logo rate at 1–3% (measured 2.1%, Wilson 95% CI 0.4%–10.9%) — not zero.

Does this study prove logo colour affects funding?

No. Every company already raised substantial capital; there's no comparison group that didn't. Colour is chosen, not randomly assigned, and rebrands typically follow a raise rather than precede it. The study describes colour patterns among already-funded companies — it cannot show colour caused any of them to be funded.