Flavour sampling bundles sold through online liquid flavour shops tend to arrive in one of two configurations: a fixed assortment chosen by the buyer, or a randomized assortment curated by the seller. The randomized version, when it works, produces a distinctive pattern of engagement — a burst of curiosity that decays quickly, sometimes within the first three vials. The question worth asking is whether that decay reflects a failure of the product or a structural feature of how randomized reward is processed. If it is structural, then the conventional wisdom that "more variety equals more excitement" may be miscalibrated, and the optimal bundle size for a randomized flavour assortment may be smaller than sellers assume.
The Reinforcement Schedule Behind the Flavour Bundle
Variable-ratio reinforcement is one of the most robust findings in the operant conditioning literature. B.F. Skinner's work in the 1950s established that when a reward is delivered after an unpredictable number of responses, the resulting behavior is more persistent and more resistant to extinction than behavior maintained by fixed or predictable schedules. The mechanism is not mysterious: unpredictability itself becomes the signal that another reward may be imminent, and the organism learns to keep responding through the gaps.
A randomized flavour bundle is a consumer-facing application of this principle, though rarely described that way. The buyer does not know which flavour arrives next. Each vial is a draw from a distribution. The seller controls the distribution; the buyer supplies the attention. What the seller typically does not control — and often does not model — is the rate at which the novelty of the schedule itself wears off.
The critical variable is not the number of draws but the perceived informativeness of each draw. Once a buyer has sampled three randomized vials, they have typically formed a working model of the seller's curation: the range of flavour families, the quality floor, the likelihood of a genuine surprise. Additional draws add less new information per unit than the first three did. The schedule is still variable, but it is no longer informative in the way that drove the initial engagement.
Why Three Draws Saturate the Curiosity Curve
The information-theoretic account is straightforward. The first vial in a randomized bundle carries high entropy: the buyer has almost no prior on what the seller's distribution looks like. The second vial resolves some of that uncertainty. By the third, the buyer has enough samples to estimate the shape of the distribution — the mean, the spread, the presence or absence of outliers. Further draws refine the estimate but do not change its character.
This maps onto a well-documented pattern in consumer research on assortment. Studies of variety-seeking behavior, including work by Barbara Kahn and colleagues on the "variety drive," show that consumers seek variety up to a point and then experience what has been called "variety fatigue" — a decline in the marginal utility of additional distinct options. The inflection point varies by category, but for hedonic goods with high sensory specificity, it tends to arrive early. Flavour is a sensory category. The palate adapts.
There is also a working-memory constraint. A buyer sampling a randomized bundle is holding a mental model of the seller's curation in active memory while evaluating each new vial against it. Three vials is roughly the span at which that model stabilizes. Beyond that, the buyer is no longer learning; they are confirming. Confirmation is pleasant but not compelling, and it does not sustain the same attention.
Loss Aversion and the Six-Vial Trap
The six-vial bundle is attractive to sellers for a reason that has little to do with buyer psychology and everything to do with unit economics: it raises average order value and reduces per-unit packaging cost. But it may be working against the buyer's actual experience of the product.
Kahneman and Tversky's work on loss aversion established that losses loom larger than equivalent gains — roughly twice as large, in many experimental settings. In a randomized flavour bundle, the relevant "loss" is not monetary. It is the small disappointment of a vial that misses. A buyer who receives three vials and finds two pleasing has a favorable ratio. A buyer who receives six and finds three pleasing has the same ratio but a larger absolute number of misses, and the misses are more salient because they arrived after the buyer had already formed expectations. The sixth vial is evaluated against a model the buyer built from the first five. It is judged more harshly than the first vial was.
This is the six-vial trap: the bundle that maximizes perceived variety also maximizes the surface area for mild disappointment, and the disappointment is weighted more heavily than the equivalent satisfaction. Sellers who extend bundle size without attending to the quality floor of the distribution are trading short-term order value for a longer-term erosion of the buyer's model of the seller's curation.
What a Three-Vial Design Actually Optimizes
A three-vial randomized bundle is not a compromise. It is a design that aligns the product with the shape of the buyer's attention. The first vial establishes the frame; the second tests it; the third either confirms or productively violates it. A well-curated three-vial set can deliver a complete arc — surprise, calibration, resolution — within a single order.
There is a concrete precedent in the tasting-menu literature. Restaurants that serve a short, highly curated sequence of small courses consistently report higher satisfaction than those that serve longer sequences, even when the longer sequences contain more objectively "good" dishes. The finding is usually attributed to palate fatigue and to the diminishing returns of attention. The same logic applies to flavour sampling at home. Three vials is enough to feel like a sequence. Six is enough to feel like a chore.
The practical implication for sellers is not simply "shrink the bundle." It is to treat the three-vial set as the primary unit of curation and to design the distribution behind it with more care than a larger set would require. A three-vial bundle with a high variance in quality is worse than a six-vial bundle with a low variance. The size matters less than the shape of the distribution the buyer is sampling from.
Designing the Next Generation of Randomized Assortments
The forward-looking question is whether randomized flavour bundles can be made adaptive — whether the seller's distribution can update based on what the buyer has already received, so that the fourth draw (if there is one) is genuinely informative rather than confirmatory. This is technically feasible with even modest order-history data, and it would address the saturation problem directly: the bundle would not need to be short if each additional draw carried new information.
Until that exists, the evidence points toward a simpler discipline. Randomize within a tight quality band. Keep the default bundle at three. Let the buyer opt into more only after they have formed a model worth extending. The variable-ratio schedule is a powerful tool, but it rewards the buyer's attention only as long as the schedule remains genuinely uncertain — and uncertainty, in flavour as in most hedonic domains, resolves faster than we tend to assume.