A verdict before the lawyer.
Check a mark before you file itLogo, name and what it will sell. Three checks, one verdict.
- 32Beers; non-alcoholic beverages; fruit drinks
- 30Coffee, tea, cocoa; pastry
- 29Jams, compotes; preserved fruits
Why people pay for a trademark they will not get
A trademark application costs money and takes months, and most refusals are for one reason: the mark is too close to one that already exists, in its picture, in its name or in how it sounds. The applicant could not have known. The office’s register is public, but searching it properly is a profession of its own: goods classes, image codes, phonetic variants in two languages. So founders file without knowing, wait for months and then get a refusal letter.
The office came to us with a different question than “build a search”. It wanted fewer doomed applications and fewer objections to process, and it wanted to know whether computer vision was ready to do the examiner’s first look. The job was technology scouting with a product attached: prove that a machine can give an honest early answer, and design the product that carries that answer to people who have never read a trademark law.
Having the idea doesn’t make the name yours.
How the machine looks at a logo
The model takes a logo apart into attributes it can name: the object, the colours, the overall shape, whether there is a leaf, whether the fill is flat or a gradient, how many elements there are and where the text sits. Every registered mark in the chosen classes has been described the same way, so a comparison is a list of attributes that agree and disagree, and anyone can check it. The name goes through the same discipline for spelling, and once more for sound, because two marks written differently can still be pronounced alike, and the examiner treats that as a collision.
We decided early to show this decomposition instead of hiding it behind a score. An applicant who sees that the colour and the leaf are the problem can change the colour and the leaf. A percentage on its own can only be argued with.
- lemon
- yellow
- round
- leaf
- dark green
- flat
- curved cut
- bottom tip
- lemon
- yellow
- oval
- leaf
- light green
- flat
- round hole
- diagonal tips
- PTlumo · lúmo
- ENlumo · loomo
- ESlumo
- PTlumio · lúmio
- ENlumio · loomio
- ESlumio
- PTˈlumu
- ENˈluːmoʊ
- ESˈlumo
- PTˈlumju
- ENˈluːmioʊ
- ESˈlumjo
- agrees
- close
- differs
Three verdicts
The office refuses on three grounds separately, so the product judges on three grounds separately. A single similarity score would have been easier to build and impossible to act on: 41% says nothing about what to change. The headline chance is there for the glance, and under it each criterion carries its own risk and its own advice: keep the image, change the text, the pronunciation is the problem. Risk is a level, never a yes or a no. The product recommends, the office decides, and the sentence “the decision is always yours” is on the screen for legal reasons as much as for tone.
The similar marks are the argument
A verdict people will believe needs evidence they can check. Every similar mark shows the part that collides, highlighted in the name, and the three figures that produced it. Status matters as much as similarity: an expired mark is an opening, and a freshly filed one means a race. The filters are the ones attorneys use (goods classes, image codes, territories and offices) because professionals were the second audience. They fold away behind a plus so that a first-time founder is never asked a question they cannot answer. The first results are free; the full list and the report are paid, which is how the office finances the service.
On a phone, in three steps
A founder checks a name while it is still an idea, long before a lawyer is involved. So the mobile flow asks for exactly three things, a photo of the logo, what it will sell and where, and answers with the same three verdicts as the desktop. The classes are suggested from plain words, “lemonade” rather than a class number, because nobody outside the profession knows that drinks are class 32.
When the check becomes a filing
A check that ends on a score is a dead end. If the chance is good, the same data becomes an application, and on the other side of the office the examiners see it in the queue they already work in: new, in examination, objection sent, approved, rejected. The proof of concept covered both sides on purpose. Most of what the office spends goes into examiners’ hours on applications that should never have been filed.
What I did
Our full team built the proof of concept in about four months. It went to the office as the final round of its tender, and the office paid for it, which is the result that counts in a presale: a government client putting money behind a technology it had only read about.