BUILDING TASTE PROFILES THAT ACTUALLY RECOMMEND THE RIGHT CIGAR

Generic recommendation engines work on a simple principle: customers who bought X also bought Y. For books, shoes, and streaming content, this works well enough. For cigars, it's nearly useless — because it ignores everything that actually determines whether a person will enjoy a cigar.

A member who bought a Montecristo White Series and a My Father Le Bijou 1922 isn't expressing a consistent preference. One is a mild, Connecticut-wrapped Dominican with a creamy profile. The other is a full-bodied, box-pressed Nicaraguan with espresso and dark chocolate notes. A "customers also bought" algorithm would happily recommend both to the same person, learning nothing about what they actually like.

WHAT A REAL TASTE PROFILE CAPTURES

A meaningful cigar taste profile tracks preferences across multiple dimensions, weighted by recency and frequency:

• Wrapper preference. Does this member gravitate toward Connecticut, Habano, Maduro, Oscuro? What's the ratio over their last twenty purchases?
• Strength and body. Mild, medium, full? And is there a trend — are they migrating from medium toward full over time?
• Origin preference. Nicaraguan puros, Dominican blends, Honduran-Nicaraguan hybrids? Some members discover they love a specific country's tobacco and stay there.
• Size preference. Robusto loyalists, Toro fans, or Churchill-for-special-occasions? Ring gauge and length preferences are surprisingly consistent within a member's purchasing pattern.
• Brand affinity. Some members are brand loyal — all Padron, all Fuente. Others are explorers who rarely buy the same brand twice. Both patterns are useful for recommendations.

HOW THE AI USES IT

Once a taste profile has enough data — typically ten to fifteen purchases — the AI can start making recommendations that feel genuinely informed rather than random. The process isn't matching keywords. It's understanding that a member who consistently buys medium-full Nicaraguan maduros in Robusto format and recently tried (and reordered) a Plasencia Alma Fuerte would likely enjoy an Aging Room Quattro Maduro — because the flavor profile, body, and wrapper characteristics align, even though the brand and blend are different.

The AI can also detect when a member is exploring. If someone who usually buys medium maduros suddenly purchases a mild Connecticut, the system doesn't overwrite their profile. It notes the exploration and might suggest: "Trying something lighter? If you enjoyed that Ashton Classic, the Davidoff Signature 2000 has a similar profile with a touch more complexity."

THE STAFF ANGLE

Taste profiles aren't just for automated recommendations. They're a staff tool. When a regular walks in and a newer staff member is working, they can pull up the member's profile and see: "Prefers medium-full maduros. Bought Liga Privada T52 three times. Tried Oliva Serie V last visit — first time. Connecticut: never."

That's thirty seconds of context that turns a generic "what can I get you?" into a knowledgeable recommendation. The experienced staff already has this in their head for regulars. The taste profile captures it in data so it scales beyond what any one employee can remember.