Practice
How much of an identity do you actually need to be recognised?
Words by
Wall
December 18, 2026
Take away the logo, then the colour, then the typeface, then the graphic device. At which step does the brand stop being itself? The answer usually arrives earlier than anyone wants, and it tells you what is really carrying recognition.
A bottle designed to be recognised in the dark
In 1915, Coca-Cola ran a competition for a new bottle. As the company has long told it, the requirement was an unusual one for a packaging brief: the bottle had to be recognisable by touch in the dark, and recognisable even when lying broken in pieces on the ground. The winning entry was the ribbed, swollen-waisted bottle submitted by the Root Glass Company in Indiana.
The interesting thing is the brief itself. It is a subtraction test, written down more than a hundred years ago. Remove the light: is it still recognisable. Remove the label, the lettering, even the intact shape of the object: what is left that is enough to say whose this is.
Most identity systems today are not built on that logic. They are built by addition: a logo, a palette, a typeface, a grid, a pattern library, an art direction for photography, an icon set, a way of using motion. Each piece is correct and well made, and the question of who carries recognition never gets asked, because everything is present at once so nobody needs to know.
Out in the world, though, everything is rarely present at once. People see a brand in the corner of a screen, in a two-second scroll, in peripheral vision, on a van going past, printed badly, shrunk, cropped, or reduced to a fragment. Real conditions of reception are much closer to the 1915 brief than to the pages of a brand book.
Removing one layer at a time
The test is simple and anyone can run it in an afternoon. Take six to ten things your system has produced: a post, a web page, a product box, an outdoor board, a deck. Then strip away one layer at a time, and after each pass ask the same question of people who do not work at the company: whose is this.
Its limits are worth stating before you run it. This is a diagnostic heuristic, not a complete study of brand recognition. Results depend on who you ask, how much exposure they have had, whether they already know the category, and the context in which they see the test material. The same system can survive the third pass with existing customers and fail the first pass with people who have never bought. What the test produces is an indication of where to look more closely, not a number to report.
Pass one: remove the logo and the name. Cover the primary mark and every occurrence of the brand name. This is the most brutal pass and the one where many systems stop immediately. If nobody can answer, that does not mean the system is bad. It means the entire burden of recognition is sitting on one small element, usually less than three per cent of the surface area, and one that is skipped in most of the situations where people actually see you.
Pass two: remove the colour. Convert everything to greyscale. This pass separates two cases that are easily confused: a system recognised through colour, and a system recognised through structure. If greyscale makes everything collapse, colour is carrying almost all of it, and you want to know that before somebody proposes widening the palette.
Pass three: remove the typeface. Set everything in a neutral face. For systems running on off-the-shelf commercial fonts, this pass often changes the answers very little, and that is the finding: typeface does enormous work on feel without necessarily carrying the naming. The reverse holds too. For wordmark-led identities, custom lettering, or a commissioned face used exclusively over many years, typography can be the strongest recognition asset in the system, and this pass will bring everything down. When you run it, separate two things that usually get merged: the letterforms of the wordmark, and the face used for body copy. They rarely carry the same weight.
Pass four: remove the graphic device and the layout. Keep the content, remove the pattern, the frame, the cut, the familiar proportion, the way images are placed. This pass tends to produce the most surprising results, in both directions: some systems die here having survived the previous three, and some turn out to need only one layout proportion to stay recognisable.
One detail matters while running it: do not only count correct answers. Count the wrong ones and note who they named. If three people in ten look at your de-logoed work and name the biggest competitor in your category, that is far more useful information than the seven who got it right.
Four subtraction passes on one design. The question stays the same at every pass: whose is this.
Two axes, not one list
The results of the test cannot be arranged into a ranked list from strongest to weakest, because every asset has two separate properties and they do not move together.
The first is fame: how many people see it and think of some brand. The second is uniqueness: of those people, how many think of you rather than someone else. Jenni Romaniuk and the Ehrenberg-Bass group set this out under the name distinctive assets, and its value lies in separating two things that daily practice tends to collapse into one.
An asset with high fame and low uniqueness is the dangerous kind, because it looks like it is working. Blue in banking and insurance. Green attached to sustainable goods. Product shots at a three-quarter angle on white. A coined word that sounds like the future when you say it. Everyone recognises that language as belonging to a category; almost nobody can attach it to a name.
The wave of identity changes around the late 2010s is the clearest collective example. A great many brands moved to the same kind of geometric sans, dropped effects together, spaced letters the same way. Each individual decision was reasonable for screens. The cumulative result was that typeface nearly stopped doing the work of differentiation, and the burden of recognition shifted entirely onto other layers.
An asset everyone recognises but nobody attributes correctly is the category's asset, not yours.
What carries it is usually not the logo
When the test is run on brands that survive several passes, whatever holds up tends to fall into one of four groups, and the logo is rarely among them.
Colour, when it is held long enough and narrowly enough. Cadbury filed to register the purple Pantone 2685C and spent roughly eighteen years in proceedings, refused by the English Court of Appeal in 2013 over how the scope was described, and only in July 2022 granted two of three applications by the High Court. The notable detail is not the legal outcome but the gap: the public had attached that purple to the company long before the legal system managed to describe it as a mark. Recognition runs first, paperwork follows.
Shape and silhouette. The 1915 bottle. The triangular prism of a chocolate bar. The proportion and cut line of a car body. This is the strongest group under bad conditions, because it still reads when reduced to a black silhouette on a light ground, which is to say under the conditions of most real sightings.
Placement, rather than the element itself. Louboutin's red sole is not protected because it is red; it is protected because the red is on the sole. The same red placed elsewhere says nothing at all. A great many strong assets are of this kind: a white space always in the same position, a band of colour always along the bottom edge, a split proportion repeated over years.
The things you cannot see. Two notes before a film. The sound of a lid coming off. A phrase repeated long enough to become a way of speaking. The subtraction test usually misses this group because it does not live on the page, but in practice these often outlast everything sitting next to them.
What the four groups have in common: they survive poor reproduction. Bad colour printing still leaves shape. Shrunk to two centimetres still leaves layout. Half a second of peripheral glance still leaves colour. The logo, meanwhile, is the first thing to stop working as size drops, and also the thing placed in the smallest corner of most designs.
Why most systems do not pass
Not for want of talent. Most systems fail this test for four structural reasons, and most of those sit outside the designer's decision.
The system is written as a set of rules, not a set of assets. A thick guideline answers "how do I use this correctly" in great detail and barely answers "which things must be present in every single appearance". Those are different questions, and only the second one produces recognition.
Assets get replaced too quickly. A new pattern each campaign, a new art direction each year, a new treatment with each change of owner. None of it is wrong on its own. But recognition is a function of repetition, so an asset replaced after eighteen months never has time to become anyone's asset.
Too many things claim the lead role. A system with twelve signature elements in practice tends to have none strong enough, because the audience's attention budget does not divide into twelve. At the same level of exposure, a few elements repeated densely usually have better odds than many elements splitting the same number of repetitions between them.
Every rebrand resets the clock to zero. This is a real cost and it appears on no invoice. An asset only becomes distinctive through time; replacing it discards every year accumulated so far in exchange for something starting again from nothing. Sometimes that is the right decision. But it is rarely counted as a loss, because the thing lost sits on nobody's books.
A strong asset is one that still reads when only the silhouette is left.
So how much is enough
Far less than what is currently in your system, and repeated far more than anyone on the team finds comfortable.
A workable starting range is two to three assets carrying the load, each of which has to stand alone: show it by itself, with nothing else attached, and people still name you correctly. This is a working range rather than a rule — a system spread across multiple product lines, markets or audiences may well need more, and what matters is not the number but whether anyone can answer the question of which assets are carrying. Everything else in the system still needs to exist, still does its work on feel and aesthetics, but it should not be counted as recognition and should not be defended as though it were carrying.
The distinction is fairly blunt in practice. For those two or three, the default answer to any proposed change is no, and whoever wants the change has to make the case. For everything else, the default answer is yes, and whoever wants to keep it unchanged has to explain why. Most organisations run this exactly backwards: fiercely protecting details nobody remembers, and replacing the thing that carries with relative ease.
Teams tend to tire of an asset well before outsiders have had a chance to notice it.
There is a mismatch worth naming here, and it explains a good share of bad identity decisions. The person deciding to replace an asset has looked at it several thousand times over three years. The buyer may have seen it four times. The feeling that "this is getting old" is a real feeling, but it belongs to the person who has looked too much, not to the person the asset is meant to serve.
The subtraction test does not tell you whether your system is beautiful, and it does not replace proper measurement either. It answers exactly one question: if you take things away, what remains. And it is worth running early, because the market will run it on your behalf whether you want it to or not, every time somebody passes you at a speed where no layer gets to speak except the strongest one.
So the question is probably not how many things an identity needs. It is: how many of them can you live with never changing.
Real conditions of reception are closer to the 1915 brief than to the pages of a brand book.
Resources
The Coca-Cola contour bottle, Root Glass Company, 1915
Jenni Romaniuk, Building Distinctive Brand Assets, 2018
Byron Sharp, How Brands Grow, 2010
The move to geometric sans-serifs, roughly 2015–2020
