Beyond Jordan and Nike: What It Actually Takes to Authenticate a New Brand

Ask any marketplace operator what they want from an authentication partner and the second answer, after accuracy, is always coverage. Jordan and Nike are table stakes. The question is how fast you can authenticate New Balance, ASICS, Salomon, HOKA, Saucony, Puma — the brands that are now driving a growing share of resale volume.

The honest answer is that adding a brand to an authentication system is not a checkbox. It is a data problem first, a model problem second, and a business decision throughout. This article explains what the process actually involves, so operators can evaluate vendor roadmaps with the right questions.

The Coverage Gap

Resale demand has broadened. Running and trail silhouettes that were niche five years ago now move at volume. Lifestyle lines from New Balance and ASICS have their own collector markets. Collaborations extend across a dozen brands rather than two.

Authentication coverage has not kept pace, and the reason is structural. Human authentication scales per expert, not per brand. An authenticator who knows Jordan 1s cold does not automatically know Salomon XT-6s. Each additional brand means additional hiring, additional training, and a thinner bench of people who can arbitrate the hard cases. Most human-led operations respond by covering the high-volume brands properly and the rest superficially.

AI authentication does not have the same bottleneck — but it has a different one.

Why a New Brand Is Not a New Checkbox

Every brand has its own set of tells, and they live in different places.

Construction. Stitch density and spacing, the angle at which panels meet, heel tab shape, midsole foam texture. These vary by factory and by production year even within one brand.

Labels and packaging. Box label typography, size tag layout, barcode format, QR placement, the exact font weight on a production code. Counterfeiters get these approximately right; the differences are in the details.

Materials. Suede nap, mesh weave, leather grain, the way a specific colourway reflects light. These are hard to fake consistently and hard to photograph consistently — which matters, because the model only sees what the camera sees.

Our pipeline is built in layers that map onto these tells. Image standardisation — pose estimation and OpenCV-based normalisation — brings every submission to a consistent angle and scale before analysis. Visual feature analysis evaluates colour, texture, and shape ratio across the standardised images. Tag and label authentication runs OCR across barcodes, QR codes, and typography. Each layer produces a signal, and the signals are combined into a confidence score.

Adding a brand means teaching every one of those layers where to look. Pose estimation needs to understand a new silhouette. Feature analysis needs to know which regions carry signal for this brand and which are noise. Label OCR needs the brand's specific tag formats, including the ones that changed two seasons ago.

None of that happens by adding a brand name to a list.

The Data Problem

The constraint that actually gates new-brand coverage is training data — specifically, verified fakes.

Authentic reference images are relatively straightforward. Fakes are not. A reliable model needs confirmed counterfeit examples across the specific models and colourways it will be asked to judge, and the counterfeit market does not supply labelled datasets. The scarce side of the training set is always the fake side.

This is why the honest unit of coverage is the model, not the brand. "We cover New Balance" means very little. "We cover the 550, 990v6, 2002R, and 1906R in their primary colourways, with confirmed counterfeit examples for each" is a claim an operator can actually act on.

It is also why we have invested in generative AI for synthetic counterfeit training data — the focus of our current Cyberport Incubation Programme work. If the bottleneck is the volume of labelled fakes for a new model, the answer is to generate plausible fake variants from known counterfeit patterns and validate them against real-world submissions. This does not replace real fake examples, but it shortens the time between a model entering the market and the system being able to judge it with usable confidence.

The Feedback Loop

A new brand does not launch at full confidence, and any vendor who claims otherwise should be asked how.

The responsible approach is to launch a new model at a conservative threshold. Submissions where the system is highly confident return an automated result. Submissions below the threshold route to human review. Every reviewed case feeds back into training. Over weeks, the threshold moves as the model accumulates evidence, and the share of submissions requiring review falls.

Operators should expect to see this curve, and should expect their vendor to disclose it. Early-stage coverage with transparent confidence reporting is useful. Early-stage coverage presented as mature is a liability — for the operator, because disputed outcomes land on their platform, and for the vendor, because the trust loss is permanent.

Our confidence scoring is designed for exactly this. The operator sees not just a verdict but the verification breakdown and the system's certainty, and decides how to apply it within their own risk tolerance.

How to Evaluate a Vendor's Brand Roadmap

If you are assessing authentication partners on coverage, five questions separate marketing from capability:

Which models, not which brands? Ask for the list at model level, with colourway coverage where it matters.

What is the fake-side data provenance? Confirmed counterfeits from real submissions, purchased reference fakes, synthetic data, or a combination. Each is legitimate; the mix should be stated.

What happens below the confidence threshold? There should be a defined human-review path, not a forced binary verdict.

How long from a new release to usable coverage? The answer depends on how quickly fakes appear in the market, but the vendor should have a process and a track record.

How does the model improve after launch? Reinforcement from live submissions should be standard, and the operator should be able to see coverage maturing over time.

Where We Are

SneakersLite currently authenticates Jordan, Nike, Adidas, YEEZY, and New Balance. Puma, Vans, Converse, Saucony, HOKA, and Salomon are in active development at model level, following the process described above — conservative launch thresholds, human-review fallback, and reinforcement from partner submissions.

We publish coverage at model granularity to our partners, and we tell them where a model is still maturing. That is the standard we think the market should hold every authentication vendor to.

Work With Us

If your platform needs authentication coverage beyond the core brands, and you want a partner who will be specific about what is covered and how well, talk to us about your catalogue.

Contact our team at sneakerslite.com/contact