How treating CRO, persona research and creative production as one connected system produced +265% YoY revenue and 3× a brand’s annual revenue target in under twelve months.
The scoreboard, before the how
In eCommerce growth, no single variable ever explains a +265% year-over-year increase in sales. The JNPR case is a record of what happens when data research, conversion optimization and creative production stop operating as separate functions and are instead run as one system — observed, measured and iterated on continuously.
+265% revenue YoY. +49% conversion rate YoY. −4.1% customer acquisition cost. 3× the annual revenue target, reached in under twelve months.
Already validated: the brand before the system
JNPR is a French alcohol-free spirits brand built around a specific promise: making the aperitif moment possible without giving up on flavor, regardless of the reason someone chooses an alcohol-free, sugar-free alternative. The range covers alcohol-free alternatives to the classic spirits used in an aperitif — gin, spritz, bitter and beyond — alongside gift sets and cocktail-recipe collateral designed to make the ritual easy to recreate at home. The positioning — "L’apéritif, réinventé" — speaks to people seeking moderation without giving up on pleasure.
Since launching in 2020, that trajectory has built into a track record visible well beyond the brand’s own channels: its founder received the Choiseul 2025 award and won the Croissance category of Madame Figaro’s Business with Attitude prize in 2026, alongside publishing a book on mindful drinking, while the sugar-free range was named best soft-drink innovation of 2025. Coverage in Le Monde, Elle, Le Figaro and Vogue followed the same trajectory. Six years in, this was a company already visible and already being watched — which meant every decision on creative and conversion carried more weight than usual.
It’s against this backdrop that a recurring trait of the work stands out: at every stage, a willingness to widen the scope of testing beyond what had been planned at the outset, whenever the data pointed toward a direction different from the original hypothesis. That openness — to test personas, angles and formats that departed from the brand’s initial assumptions about its own audience — is a large part of why the results below were possible at all.
One system, not two parallel projects
The work never separated on-site optimization from creative production and media buying. Both areas ran on the same loop — research, hypothesis, test, measurement, iteration — fed by the same data and tracked through a proprietary technology layer that functioned as a single point of reference across every stage, from the initial diagnostic work to the decision on which concept to scale. Every hypothesis, every test and every outcome moved through that same layer, which is what made it possible to always have an up-to-date read on where things stood and what the next step should be, instead of piecing together separate initiatives after the fact.
CRO: from data to test
The work started with a structured Conversion Rate Optimization process, focused on the entire path from visitor to customer and built around three phases.
Research and analysis. Before any hypothesis was formed, a technical audit verified that analytics tracking was set up correctly and measured conversion rates by browser and device alongside site-speed performance, using dedicated tools to flag bottlenecks. This first pass surfaced and resolved issues that — left unaddressed — would have made every subsequent step harder to read correctly, risking interventions on what were ultimately secondary problems.
That technical layer was followed by a page-by-page heuristic audit, evaluating each page against five criteria: relevance (does the page meet user expectations, in content and design?), clarity and immediacy (is the offer obvious?), value (is it communicating enough motivation to act?), friction (what causes hesitation or doubt, and how can it be simplified?) and distraction (what pulls attention away from the intended action?). This was paired with on-site behavioral analysis — heatmaps, scroll maps, click maps and session recordings — to see exactly where users lingered and where they dropped off, with a minimum sample threshold of roughly 2,000–3,000 pageviews before treating any heatmap pattern as reliable.
The third layer was qualitative: customer interviews and post-purchase surveys surfaced information no quantitative dataset could have produced on its own. This is where the research uncovered product usage the team hadn’t anticipated — specific cocktail recipes customers had improvised on their own — a recurring need for fresh inspiration for alternative aperitifs, which translated into a measurable increase in returning customers, and a secondary segment which soon became a primary one: partners buying the product for a pregnant partner, sitting alongside the already-known "mums-to-be" audience.
Hypothesis and prioritization. Every insight from the research phase was converted into a testable hypothesis and placed on a checklist weighted by expected impact and cost, following a fixed structure: we believe that doing [A] for people [B] will make outcome [C] happen; we’ll know this when we see data [D] and feedback [E]. This is what determined which work cycles got tackled first, rather than working through changes in the order they were noticed.
Implementation, measurement, iteration. Every test was tracked with dedicated tools built for mapping results, so that the outcome of one cycle directly shaped the next. The backlog was never a static document: it was continuously updated against the performance of the most recent weeks, so that priorities always reflected the latest state of the data rather than a plan fixed months earlier.
That first cycle of work produced the initial results: +265% revenue YoY, +49% conversion rate, −4.1% acquisition cost. From there, the work expanded along two parallel tracks: increasing average order value, and systematizing creative production.
From personas to creative to dedicated landing pages
The quantitative and qualitative data gathered during the CRO phase fed directly into building testable personas — work that started with six personas and was subsequently expanded as early test results pointed to segments with more potential than the initial hypotheses had assumed. Prioritization across personas followed four axes — share of revenue generated, acquisition cost, lifetime value, and strategic priority — to decide where to concentrate creative volume rather than spreading it evenly.
One axis was dedicated specifically to people buying "for someone else" rather than for themselves. In line with what the qualitative research had surfaced — the partner buying for a pregnant partner — creative was built from the point of view of the person making the purchase as a gesture toward someone else, rather than addressing the end user directly. This is a territory typically under-covered in DTC advertising, but particularly relevant for a product built around shared, convivial moments.
Each persona did not simply receive a dedicated ad: it received an end-to-end experience. Every creative was paired with a customized landing page, matched in language and offer to the message that had brought the visitor there in the first place. The goal was to close the gap between the promise made in the ad and the page the user actually landed on — a specific experience for each persona, instead of one generic page serving everyone regardless of what had prompted the click.
Partnership ads anchored to personas
Alongside standard campaigns, partnership ads were tested — creative built in a native, organic-looking format, styled closer to an Instagram Story than a traditional ad, while keeping the same discipline of mapping each execution to the pain points and needs already identified per persona. This format’s more natural feel translated into lower cost per thousand impressions and solid conversion on audience segments already warmed up by other touchpoints — while never dropping the persona-level targeting logic used across the rest of the system.
Creative mix and calculated volume
Not every creative variation produces genuine diversity in the eyes of distribution algorithms. The work drew a hard line between weak axes of variation — format, hook rotation on an otherwise identical body, cosmetic changes — and strong axes — persona, problem framing, visual world, tone of voice, and the source of authority behind the message. Only variation along the strong axes was treated as a genuinely new concept, tested on its own; variation along the weak axes was used to remix concepts already validated, isolating what needed to stay fixed in a winning creative (its angle, its promise, its argument structure) from what could be re-expressed across different formats — static, UGC and talking-head, high-production video, AI-generated animation.
Creative volume itself wasn’t estimated, it was calculated. Starting from the spend allocated to new-customer acquisition, the historical average spend required to produce one winning concept, and the hit rate observed across previous cycles, it was possible to work out exactly how many assets needed to be produced in a given period to support both the spend target and the expected results — with a built-in buffer to absorb the normal uncertainty of the process.
The same discipline extended to format testing itself: how-to explainers, "3 reasons why" comparisons, unboxing content, short-form UGC, product reviews addressing objections up front, multi-SKU grid formats, narrated slideshows, before/after framing and reaction-style content were all treated as distinct territories to validate, not interchangeable styling choices on the same message.
The scientific loop: data, creative, iteration
Every new concept moved through the same three-stage path. Exploration identified genuinely distinct territories — differences in persona, problem, awareness level, message and format, not minor hook tweaks. Validation used a controlled, single-variable test built on a hypothesis written down before launch. Scaling diversified the execution of a validated concept while preserving the exact signal that had made it win, rather than simply duplicating the winning asset.
Results from every test were read along a sequential signal chain — attention, interest, commercial intent, business outcome — rather than judged on single-day performance of a single asset in isolation; the read was always at the portfolio level. Every test cycle, won or lost, fed back into the backlog: winning concepts updated the coverage map by persona and territory, while concepts that failed to validate narrowed the areas still worth exploring. No cycle started from zero — every new batch of testing was built on what the previous batch had already established.
This is where the proprietary technology layer mattered most: by keeping every hypothesis, every test and every outcome tracked across the full process, it made it possible to maintain one continuous, coherent read of how the system was performing — instead of a series of disconnected initiatives that each had to be manually reconciled after the fact.
Growing average order value
After the first cycle of improvement on conversion rate and acquisition cost, attention shifted to increasing average order value — a lever that supports higher acquisition costs during periods of heavier competition and improves cash flow while holding or improving margin.
The starting point was correcting a common mistake: treating AOV as a single average figure without looking at the actual distribution of order values. Rather than one blended average, order volumes clustered around two distinct price points — a bimodal distribution that a single average would have completely hidden, and that pointed to two different buying behaviors sitting inside the same number.
Based on that finding, targeted levers were developed: in-cart incentives calibrated to a free-shipping threshold, new cross-sell offers with pricing points calculated specifically to push orders toward the AOV target, and bundles built for specific customer clusters — distinguishing, for example, whether customers in the lower price cluster were first-time buyers or returning customers, since that distinction changed which offer made sense for which group. This work produced a 96% increase in AOV in its first phase.
What the system delivered
Taken together — data research, CRO, personas, creative, dedicated landing pages, partnership ads, calculated creative mix, and a continuous testing loop — the system produced:
+265% revenue YoY. +49% conversion rate YoY. −4.1% customer acquisition cost. +96% average order value. 3× the annual revenue target, reached in under twelve months.
Not a single intervention, but a system that updates itself: every test feeds the next one, and every iteration starts from a higher level of knowledge than the one before it.
Ready to apply it to your brand?
Bring the context: our senior team will help turn it into operating priorities.