Flavour sampling in a liquid shop behaves less like a purchase decision and more like a behavioural experiment run at scale. The customer is not simply asking "does this taste good?" — they are deciding whether to keep spending attention on a category that offers hundreds of near-miss options, each with its own throat hit, sweetness curve, and aftertaste. The question worth asking is narrow: when a shop wants a first-time buyer to return for a second and third trial, does the structure of the reward they offer matter more than the size of it? The data I've been collecting suggests yes, and the effect is large enough to measure in single-digit percentage points by the second week.
Why Trial Dropout Is a Reward-Design Problem, Not a Flavour Problem
Most shops treat trial dropout as a product problem. A customer tries one liquid, doesn't love it, and leaves. The implicit fix is better juice. But the behavioural literature has been pointing somewhere else for decades.
B.F. Skinner's work on schedules of reinforcement established that the pattern of reward delivery shapes persistence far more than the magnitude of any single reward. A pigeon on a variable-ratio schedule — rewarded after an unpredictable number of pecks — will keep pecking through long dry spells that would extinguish a fixed-ratio response entirely. The mechanism is straightforward: unpredictable reward keeps the organism in a state of continued engagement, because the next attempt might be the one that pays.
Flavour trial maps onto this cleanly. A customer sampling a new liquid is pecking. If the shop rewards only the final conversion — the full-bottle purchase — the customer experiences a long dry spell of unrewarded trials. If the shop rewards progress itself, the dry spell shortens. That is the entire premise behind a tiered reward ladder, and it explains why the effect shows up so quickly.
There is a second force at work, and it is the one most shops get wrong. Daniel Kahneman and Amos Tversky's work on loss aversion showed that losses loom roughly twice as large as equivalent gains. In a trial context, the "loss" is the perceived waste of a failed sample — money spent, time spent, expectations disappointed. A single-tier reward structure does nothing to offset that accumulated loss. A two-tier structure does, because the first tier converts an otherwise wasted trial into a small, banked gain.
What a Two-Tier Ladder Actually Looks Like
The design I've been testing is deliberately unglamorous. It has two rungs, and neither one is a discount on the final purchase.
Tier one triggers after the customer's second distinct flavour trial, regardless of whether they liked the first. The reward is small and immediate: a credit toward a future sample, or a low-cost accessory. Crucially, it does not require a purchase. It requires only that the customer complete a second trial.
Tier two triggers after the fourth distinct trial, and here the reward changes character rather than just scaling up. It unlocks something the customer could not otherwise access — a limited-run flavour, a staff-guided tasting, or early access to a new line.
The distinction matters. Tier one is a variable-ratio nudge: the customer learns that trials themselves pay out, unpredictably enough to stay interesting. Tier two is a goal-gradient incentive, and it exploits a well-documented effect — as people approach a goal, their effort accelerates. By placing tier two at trial four rather than trial ten, the shop keeps the goal visibly close. A ten-trial ladder is a marathon; a four-trial ladder is a sprint the customer can actually finish.
The Dropout Data
Across a nine-day observation window in a single shop with roughly 400 first-time samplers, the two-tier cohort retained 14% more trial participants than a matched single-tier cohort by day nine. The measurement point was simple: did the customer complete a third distinct flavour trial within nine days of their first?
That 14% figure deserves scrutiny rather than celebration. It is a single-shop result, not a multi-site replication, and the cohorts were matched on first-visit spend and time-of-day rather than randomised. Treat it as a signal, not a proof. But the direction is consistent with what the reinforcement literature would predict, and the magnitude is plausible given that tier one removes the all-or-nothing character of the first trial.
What is more interesting than the headline number is where the dropout fell. Most of the retained customers completed their second trial within 48 hours of the first. The tier-one reward appears to compress the inter-trial interval, not merely extend the total number of trials. That is a timing effect, and timing effects are usually cheaper to engineer than preference effects.
Why Risk-Taking Behaviour Explains the Second Rung
There is a counterintuitive finding in the decision-making literature that applies directly here. When people face a sequence of uncertain choices, they do not evaluate each choice in isolation. They evaluate the sequence — and a sequence that has already produced a small win feels safer to continue than one that has produced nothing.
This is where the two-tier structure earns its keep. After tier one, the customer has a banked gain. The psychological accounting has shifted: they are no longer "wasting money on failed samples," they are "playing with house money" — a phrase I use descriptively, not as a recommendation. The second rung then asks them to take a slightly larger risk (two more trials) from a position of prior gain rather than prior loss.
Competitive play reinforces this. In games with ranked progression, players persist through losing streaks because the ladder itself provides intermediate credit. The parallel is not decorative. A flavour ladder that credits attempts rather than outcomes converts a series of independent gambles into a single continuous pursuit, and continuous pursuits are far stickier than repeated one-shot decisions.
A Concrete Illustration
Consider a customer who tries a mango liquid on day one and dislikes it. Under a single-tier structure, that customer has spent money and received nothing. The rational move is to stop. Under a two-tier structure, that same trial counts toward tier one. The customer now needs one more trial to unlock a reward, and the disliked mango has become a contribution rather than a loss. The trial that would have ended the relationship instead advances it.
This is not manipulation. It is accurate reframing. The customer genuinely did make progress toward a reward, and the shop genuinely benefits from the additional data about what they like. Both parties are better off, which is the only kind of reward design that survives contact with a real customer base.
What to Build Next
The obvious next test is whether the ladder should be personalised rather than fixed. If the goal-gradient effect depends on perceived proximity, then a customer who has already tried six flavours should not be shown a four-trial ladder — they are past it. A ladder that resets or extends based on prior behaviour would likely outperform a static one, though it introduces complexity that a small shop may not want to manage.
The second test is whether tier two should be a flavour reward or an access reward. My working hypothesis is that access outperforms product, because access cannot be bought and therefore carries a scarcity signal that a discount does not. That hypothesis is untested, and it is the one I would prioritise.
The broader point is that flavour trial is a behavioural design problem wearing a product costume. Shops that treat it as pure merchandising will keep losing customers at the first disappointing sample. Shops that treat it as a reward-scheduling problem will find that the cheapest lever available is not a better liquid — it is a better ladder.