How accurate is RecoMelon AI matching?

Accuracy depends on what 'good match' means

Two products can look broadly similar and still be poor alternatives for the same shopper. RecoMelon therefore treats recommendation quality as contextual rather than reducing it to one universal similarity measure.

VisualDNA™ establishes the visual relationship

RecoMelon can analyse characteristics such as silhouette, colour relationships, pattern, cut and visible structure. IntentWeave™ and ChromaSense™ can provide additional attribute and colour context.

Useful similarity preserves intent

A strong alternative should retain enough of the original product to explain why the shopper was interested while allowing the characteristic causing hesitation to change.

Dealbreaker™ Filters add explicit shopper intent

If neckline, sleeve, colour or another attribute is non-negotiable, the shopper can refine the recommendation set instead of depending on the model to infer every preference.

Behaviour can sharpen relevance

In-session signals can add context as the shopper explores, while VisualDNA™ remains the product-level foundation.

How merchants should evaluate quality

Review representative products across major categories, difficult edge cases, new arrivals and long-tail inventory. Combine qualitative match review with engagement, discovery depth and influenced commercial outcomes rather than judging accuracy from a handful of products.