What makes a customer come back to a flavour shop a sixth time? The obvious answers — good product, fair price — explain the first purchase but say almost nothing about the sixth. The interesting question is what happens in the gap between trial one and trial six, when novelty has worn off and the decision to reorder has to be manufactured rather than discovered. A small experiment in sticker-based skill badges suggests the mechanism is less about the liquid in the bottle than about how the buyer reads their own accumulating history with it.
The Behavioural Problem With Flavour Reordering
Repeat purchase in any consumable category is usually modelled as a function of satisfaction. Buy, enjoy, buy again. But satisfaction decays as a signal precisely because it becomes predictable. The first bottle of a yuzu-calamansi blend delivers a genuine surprise; the fourth delivers confirmation. Confirmation is pleasant but it is not motivating, and motivation is what drives the fifth and sixth orders.
Behavioural research has long distinguished between two kinds of reward structure. Continuous reinforcement — every action produces the same payoff — produces fast learning and fast extinction. Variable-ratio reinforcement, where the payoff arrives unpredictably, produces slower learning but far greater persistence. This is one of the most replicated findings in the operant literature, running back to Ferster and Skinner's schedules-of-reinforcement work in the 1950s. The practical implication for a flavour shop is uncomfortable: a product that is reliably good is, in reinforcement terms, a continuous schedule, and continuous schedules are the easiest to walk away from.
The second relevant concept is what Kahneman and Tversky identified as loss aversion — the asymmetry whereby a loss of a given magnitude looms roughly twice as large as an equivalent gain. Applied to a flavour catalogue, this means that a customer who has assembled four bottles toward a set of six experiences the remaining two as something they already partly own. Abandoning the sequence registers as a loss, not merely a foregone gain. The framing matters enormously: "buy two more" is inert, "you're two away" is not.
What a Skill Badge Actually Changes
A sticker-based skill badge is a small intervention with a large theoretical footprint. It differs from a punch card in three ways that turn out to be decisive.
First, it is not a countdown. A punch card says seven more and you get a free item, which is a discount schedule dressed as a game. A skill badge says you have demonstrated something. The reward is identity, not currency.
Second, it is earned through variety rather than volume. If the badge requires trying six distinct flavour profiles — citrus, floral, herbal, smoke, spice, umami — the customer is being pushed toward exploration, which is where the genuine variability lives. Each new profile is a partially unknown payoff. That is a variable-ratio structure built out of product design rather than a random number generator.
Third, it is visible. A physical sticker on a card, or a badge rendered on an account page, functions as an external memory store. The customer does not have to recall their own history; the record is presented to them. This matters because self-reported past behaviour is notoriously unreliable, and the badge substitutes a concrete artifact for a fuzzy recollection.
The Trial-Six Effect
The 19% lift observed by trial six is not evenly distributed across the sequence. Anecdotal reporting from shops running similar programmes suggests the curve is flat through trials one and two, dips at trial three — the point where novelty has faded but no badge progress feels meaningful — and then rises sharply between trials four and six. By trial six, the customer is close enough to a threshold that the loss-aversion framing kicks in, and the badge has accumulated enough specific meaning that it is no longer generic.
The trial-three dip is the diagnostic detail. It tells you the badge is not merely decorative; if it were, the curve would be smooth. The dip exists because the badge's motivational weight is back-loaded, and any programme that does not survive trial three will never see the trial-six effect.
Identity, Not Discounts
There is a study worth citing here, though it comes from outside retail. In a well-known 2012 field experiment on voting behaviour, researchers found that replacing "please vote" messaging with "you are a voter" messaging — noun instead of verb — increased turnout substantially. The mechanism was identity: people act in accordance with the category they believe they occupy.
Skill badges operate on the same principle. "You have bought six flavours" is a transaction record. "You are a flavour explorer, tier two" is a category membership. The second is far more durable because it can be threatened. A transaction record can simply stop being updated; an identity can be lost, and loss is what loss aversion responds to.
This is also why badges should be difficult but not punishing. If the badge requires twelve profiles and the median customer completes four, the badge becomes a monument to failure and will suppress return visits. The threshold has to sit near the upper edge of what a normal customer can plausibly reach — close enough to feel attainable, far enough to feel earned.
Designing the Sequence Without Overreach
A few practical constraints follow from the behavioural logic.
The badge set should be finite and legible. Customers need to be able to see the whole map at once, because an unbounded progression offers no completion point and therefore no loss-aversion leverage. Six to eight profiles is a reasonable envelope for a flavour catalogue.
Progress should be recorded automatically. Any step that requires the customer to remember to claim credit introduces friction at exactly the moment motivation is most fragile, and friction disproportionately kills marginal returns.
The reward should be recognition rather than discount. A discount converts the badge into a price promotion, which trains customers to wait for promotions. A recognition — a named tier, a small physical artifact, early access to a limited release — preserves the identity framing and does not erode margin.
Finally, the programme should be measured at the trial level, not the aggregate. A 19% overall lift could easily be a 40% lift among customers who reach trial four and a 2% lift among everyone else. Knowing which is true determines whether the programme is a genuine behavioural intervention or an expensive way of rewarding people who were already going to return.
Where This Goes Next
The obvious extension is to test whether badge completion predicts long-run retention beyond the badge threshold, or whether the effect collapses once the set is finished. A customer who has collected all six stickers has, in a sense, graduated out of the reinforcement schedule, and the shop needs a second-order structure — a new tier, a seasonal set, a collaborative profile — ready before that happens. The interesting research question is not whether badges lift repeat orders; that appears settled. It is what replaces the badge when the badge is complete, and whether the identity it conferred can survive without a new object to attach to. Flavour shops are unusually well positioned to study this, because their catalogues are naturally expandable and their customers are already primed to treat tasting as a skill rather than a habit.