How Coffee Taste Matching Works

The Basics: How Any Coffee Subscription Matches You

Nearly every coffee subscription starts with a taste quiz. You answer 5-10 questions:

• What's your preferred roast level? (Light, medium, dark)
• How do you brew your coffee? (Drip, pour-over, espresso, French press)
• Do you prefer fruity and bright or chocolatey and smooth?
• Do you add milk, sugar, or drink it black?
• How adventurous are you?

Your answers create an initial preference profile. The subscription then recommends coffees from its catalog that match those preferences, light roasts for someone who said "fruity," dark roasts for "chocolatey," and so on.

This works reasonably well for the first few bags. But it has a fundamental limitation: the quiz captures what you think you like, not what you actually like. Many coffee drinkers discover that their real preferences are more nuanced than a quiz can capture.

The Problem with Quiz-Only Matching

When a subscription relies solely on your quiz answers, several things happen:

Your preferences don't update. You said you like medium roast, so you keep getting medium roast. But what if you tried a light-roast Ethiopian and loved it? The system doesn't know.

The matching can't capture complexity. A quiz puts you in broad categories (light/medium/dark, fruity/chocolatey). Real taste preferences are far more dimensional, you might love washed Kenyans but not washed Colombians, prefer high-altitude Guatemalans but not high-altitude Peruvians, enjoy natural-process coffees but only in light roasts.

Some months you get coffees you don't like. Because the matching is based on broad categories, there's significant variance. Some bags are perfect, some are "fine," and some just aren't for you. And there's no mechanism to reduce this variance over time.

Some subscriptions (like Trade Coffee) add a feedback layer, you rate coffees you receive, and "in-house experts use this information to fine-tune future deliveries." This helps, but it's a human-mediated process. A person reads your feedback and adjusts your next pick. The system doesn't mathematically compound what it knows about you.

How Rating-Based Matching Works

Rating-based matching, the approach Siip Coffee uses, adds a continuous learning loop on top of the initial quiz.

Step 1: The Initial Quiz

You start the same way, a taste quiz builds your initial profile. This gives the system a starting point. Your first delivery is based primarily on quiz answers, similar to any other subscription.

Step 2: You Rate Coffees

After drinking a coffee, you tell the app what you thought. Rate it, note specific flavors you picked up, or simply indicate whether you loved it or not. The key is that the system now has a data point: you, this specific coffee, and how much you enjoyed it.

Step 3: The System Learns

Here's where it diverges from quiz-based systems. The system doesn't just record "they liked this coffee." It analyzes what made that coffee different from others you've rated:

• Origin pattern: Do you consistently rate East African coffees higher than Central American?
• Processing preference: Are your highest ratings on natural-process coffees, or washed?
• Altitude correlation: Do you prefer beans grown above 1,800 meters?
• Flavor note affinity: Which flavor notes correlate with your highest ratings?
• Variety preference: Do you rate Gesha and SL-28 varieties higher than Caturra?
• Roast level nuance: You said "medium" in the quiz, but your ratings show you actually prefer the lighter end of medium.

But it goes deeper than these individual data points. The system also finds other users with similar taste profiles and identifies what they love that you haven't tried yet. It spots styles and origins you might never seek out on your own but are likely to enjoy based on your broader pattern. And it continuously evolves, each new rating refines not just what it knows about the coffee you rated, but how it weighs every other dimension of your profile.

These patterns emerge from real behavior, not self-reported preferences. And they're often surprising, many users discover preferences they didn't know they had.

Step 4: Match Scores Update: Instantly

With Siip, the moment you submit a rating, the system recalculates your taste profile and updates match scores across the entire database of 30,000+ coffees. This happens in real time, not on a weekly cycle, not after a human reviews it. Your match scores at 9:01am reflect the rating you gave at 9:00am.

Step 5: Your Next Delivery Is Better Matched

When it's time for your next delivery, the system searches the subscription catalog for the coffees with the highest match scores against your updated profile. Because the profile is more accurate than it was last month, the match quality is higher.

This creates a compounding effect:

• Month 1: Match based primarily on quiz. Solid, but broad.
• Month 2: System has 2 rated coffees. Match improves slightly.
• Month 3-4: With 4-8 ratings, patterns start emerging. Match quality jumps noticeably.
• Month 5+: With 10+ ratings, the system is mapping nuanced preferences. Match scores consistently high.

What Makes This Possible: The Data

A learning system is only as good as the data it can learn from, both about you and about the coffees.

Your Data: Ratings from Everywhere

Siip's system learns from every coffee you rate, not just coffees from the subscription. This is a critical difference. If you:

• Rate a pour-over at a café in your city
• Scan and rate a bag you bought at a grocery store
• Log a coffee you tried while traveling abroad
• Rate a bag a friend gave you

...all of those ratings feed your taste profile. This means the system has much more data to work with than subscriptions that can only learn from the 1 bag per month they ship you. More data = faster learning = better matches sooner.

Coffee Data: Farm-Level Detail

The system needs detailed information about each coffee to find meaningful patterns. If it only knows "Colombian medium roast," it can't distinguish between coffees that are actually very different.

Siip indexes coffees with farm-level detail: the specific producer, farm, region, altitude, variety, processing method, roast level, and flavor profile. With 30,000+ coffees cataloged at this level of detail, the system can identify very specific preference patterns, like your affinity for washed SL-28 from Nyeri, Kenya at 1,900m+ elevation.

The Match Percentage

Siip displays a match percentage on every coffee, a prediction of how well it fits your taste profile. This number is useful in several ways:

• Before a subscription delivery: See the match score before it ships. Swap if you want something higher.
• At a café: Scan bags and see which one is your best match before ordering.
• Browsing online: Sort any roaster's offerings by match score to find your optimal choice.
• Tracking improvement: Watch your average match score increase month over month as the system learns.

The match percentage isn't static, it recalculates every time you rate a coffee, reflecting your updated preferences in real time.

How Fast Does It Get Accurate?

Most users report that Siip's recommendations become noticeably precise after rating 10-15 coffees. Because Siip sends 2 bags per month and you can rate coffees from any source, reaching 10-15 ratings can happen in 2-3 months, or faster if you actively scan and rate coffees from cafés and stores.

The system never stops learning. Even after 50+ ratings, new data points continue to refine edge cases and uncover preferences you might not have encountered yet.

Rating-Based vs. Quiz-Based: A Summary

Sources

• Specialty Coffee Association, sca.coffee. Standards for coffee quality evaluation
• Cup of Excellence, cupofexcellence.org. International coffee competition methodology
• Siip Coffee, siip.coffee. Product documentation