Token-Based Conjoint Analysis: A New Framework For Too Many Product Features

Testing features works well on a small scale. But when your product roadmap expands from 5 features to 40, traditional feature analysis methods collapse under their own weight. Respondent attention wanes, data integrity suffers, and the clarity needed for confident product decisions vanishes.

Token-Based Conjoint Analysis (TBC), a new methodology created by Numerious that won Best Paper at the 2025 Sawtooth Research Conference, lets researchers test up to 40 features at once by having people pick their favorites from a subset and rate their purchase interest based on only those favorites being included in the product offering. TBC is particularly helpful for subscription services, where there are often many possible features to test.

Quality data is critical for business success. When teams work with flawed insights, they build features customers simply don't use. This isn't a rare occurrence — it's the industry standard. According to Pendo, 80% of software product features are rarely or never used, representing millions in wasted development costs.

Token-Based Conjoint helps product teams break this cycle by providing more reliable data. By identifying the most valuable combination of features, teams can focus resources on building what users actually want.

Why old product feature analysis methods fall short

Market researchers typically use two primary methodologies when testing product features: MaxDiff and Choice-Based Conjoint (CBC).

MaxDiff works well for ranking features and measuring reach. Respondents are shown 2-6 features at a time and asked to identify the most important feature as well as the least important feature. But the problem is MaxDiff can't predict how combinations of features work together. Knowing Feature A and Feature B are both important doesn't tell us if they'd create a winning combination. That's where conjoint analysis comes in.

Traditional conjoint analysis measures how consumers value individual components of a product or service. It works by showing respondents 3-5 product concepts at a time with varying feature combinations, asking which they prefer, and repeating this process multiple times with different concept sets. From these choices, researchers can calculate how much unique value each feature adds to the overall product.

But a conjoint experiment can become difficult for the respondent when there are too many attributes (i.e., 15 or more attributes). People experience cognitive overload trying to process so much information at once. And when faced with too many options, participants can resort to heuristics where they just choose based on brand or price which will result in noisy data around the product features we set out to understand with the research.  

What is Token-Based Conjoint?

Token-Based Conjoint offers a new framework that balances feature overload, real-world decision-making, and diminishing returns.

Instead of evaluating an exhaustive list of product concepts, respondents are shown a manageable subset of features (typically 7 per screen). They allocate tokens to the features they value most, signaling their priorities. After making their selections, they indicate how likely they would be to purchase a product with those chosen features. This process repeats across several sets of features.

Allocating tokens mirrors how consumers make decisions in real life (prioritization under constraints). TBC also shows manageable subsets of features, allowing researchers to test up to 40 features without overwhelming participants. This approach improves data quality, yielding more thoughtful, realistic responses.

Using Token-Based Conjoint analysis for many product features: An example

Let's say a meal delivery service called QuickBite wants to know which benefits will make people subscribe. They have 38 potential features to test, far too many for traditional conjoint analysis.

Using Token-Based Conjoint, respondents might see a screen asking them to choose 4 benefits from a list of 7 that would make them most likely to subscribe to QuickBite. The options include things like customizable meal plans, fresh ingredients, discounts, and special offerings.

After picking their top choices, they indicate how likely they would be to subscribe if QuickBite only came with those selected benefits.  The process repeats with a new subset of 7 items, dynamically changing how many tokens can be allocated. 

This enables researchers to identify not only which features individually generate the most interest but also how their combinations influence purchase likelihood.

For example, the TBC analysis might reveal that "Free Shipping" is the most powerful driver overall, with a 41.2% share of preference, followed by "Date Night Special" at 41.0% and "Portion Control Options" at 40.9%. But combining “Portion Control Options” and “Free shipping” is actually more powerful than “Date Night Specials” and “Free Shipping”. 

Considerations when using Token-Based Conjoint analysis

While Token-Based Conjoint solves many problems, there are some limitations. TBC never shows different bundle sizes against each other (like a 2-feature bundle versus a 3-feature bundle). Instead, everything compares to a "none" threshold, meaning no bundle is chosen (I.e., they will not subscribe). As a result, you can learn which features work best together, but not how different-sized bundles perform in a competitive context.

Since the method asks you to allocate tokens to the best features, it does very well at understanding the most preferred features but not as granular for the least preferred features.  If you need to know what features actively drive people away, this approach might not be the right fit. Additionally, further testing is needed to confirm how far TBC can scale beyond 40 attributes. While it handles more features than traditional methods, there may still be an upper limit.

Lastly, if you want to understand  price sensitivity, then this likely isn’t the right approach for your business problem as we suggest locking the price point in the TBC experiment. Should a demand curve be your primary output, a traditional conjoint (with a reduced feature set) is more ideal. Perhaps you’ll use TBC first to reduce the feature list before running a conjoint! 

The future of product feature testing

Product teams no longer need to choose between testing all their features or getting reliable data. Token-Based Conjoint Analysis solves the "too many features" problem by testing up to 40 product features without overwhelming respondents, resulting in more accurate insights about what customers actually want.

This approach gives product teams a practical tool to test complex feature sets while maintaining high data quality. As products continue to pack in more options and customizations, methods like TBC help teams avoid the 80% trap — investing in features nobody uses.

Ready to transform feature prioritization? Numerious has put these powerful tools directly in your hands through open-source templates on GitHub. No expensive software or specialized expertise required — start implementing smarter feature testing today.

Watch our on-demand webinar to see Numerious CEO, Megan Peitz, demystify this award-winning methodology step-by-step, making it easy to understand and implement.