Key results
| Metric (France, A/B test) | Uplift with Yuma |
|---|---|
| Add-to-cart rate (primary goal) | +19.2% |
| Checkout completed (secondary goal) | +20.5% |
| Revenue per visit | +23.1% |
| Average order value | +2.2% |
105 days (June 16 to September 30, 2026), 3.49M visits, over 99% statistical confidence on primary and secondary goals. Test run on Kameleoon.
Introduction
Cabaia is a French brand founded in 2015. It first made its name with beanies featuring interchangeable pompoms, then built its reputation on modular backpacks, and today describes itself as France's number one backpack brand, with a wider range of bags and accessories.
Digital accounts for 45% of Cabaia's revenue. The brand runs two Shopify stores (one for France, one for the rest of Europe), sells on marketplaces, and operates a network of around 50 retail points of sale across France, Germany, the Netherlands and Belgium. Germany and the Netherlands are its main international growth markets.
Morgane Collin has led e-commerce at Cabaia for five years. She owns the e-commerce strategy in France and internationally, the online revenue targets, and the on-site experience. Cabaia already used Yuma for customer support, so when Morgane went looking for a way to lift conversion on product pages, testing Yuma Sales AI was a natural next step.
“I was looking for an AI agent solution to increase add-to-cart on our product pages. The idea was to bring the kind of advice you'd get in-store, and remove purchase friction right at the moment it comes up.”
Morgane CollinDirector E-commerce, CabaiaThe problem
Cabaia is a data-driven team. Morgane tracks the full funnel closely: the share of sessions reaching a collection page, then a product page, then adding to cart. Add-to-cart on product pages was the step she wanted to move.
The challenge was that Cabaia could see where shoppers hesitated, but not why. The team pieced together signals from support tickets, customer reviews, in-store feedback, heatmaps and session recordings. Useful clues, but still guesswork.
“That gave us clues about where people looked or clicked, but not the 'why': we could see a customer was hesitating, not what they were trying to find out. A large share of hesitations never turned into a contact: the customer simply left without buying.”
Morgane CollinDirector E-commerce, CabaiaCabaia wanted to bring the kind of advice shoppers get from its in-store teams to its online stores. And Cabaia's products generate a lot of questions: modular backpacks and pouches, different materials, laptop sizes, weather resistance.
“We have products that are sometimes a bit technical, or that come in different materials. Customers always have plenty of questions. So the idea was really to bring more conversation into e-commerce and remove the barriers to purchase.”
Morgane CollinDirector E-commerce, CabaiaSolution
Cabaia had been using Yuma for customer support with good results, which gave Morgane the confidence to test Yuma Sales AI on product pages. Going in, she had three questions.
How does the AI actually learn, and how much upkeep does it need? Cabaia launches products constantly and manages somewhere between 2,000 and 3,000 SKUs. A tool that depended on uploading and maintaining large product files was a non-starter. Yuma Sales AI reads Cabaia's existing product pages and back office directly, so there was nothing extra to maintain.
“We launch a lot of products. We have somewhere between 2,000 and 3,000 SKUs to maintain. We clearly needed something that could read our product pages and our back office.”
Morgane CollinDirector E-commerce, CabaiaWould it sound like Cabaia? Morgane wanted the AI to match the tone of Cabaia's customer service team and the brand's voice overall, not sound generic or robotic. Because Yuma was already handling Cabaia's support, much of that context carried over.
“My question was mostly about tone. Would the tool sound like our customer service team, and more broadly, would it keep the brand's tone of voice? Overall, yes, it does. I haven't had any red flags on that.”
Morgane CollinDirector E-commerce, CabaiaWould the impact on conversion be big enough to justify rolling it out everywhere? That's what the A/B test was for.
A clean A/B test on Kameleoon
Cabaia tests everything before rolling it out, and Sales AI was no exception. The team hard-coded the Yuma widget on product pages across both Shopify stores, then used Kameleoon, its A/B testing tool, to hide the widget for the control group. Morgane found this simpler than implementing the widget inside Kameleoon itself.
The primary goal was add-to-cart rate, with completed checkout as the secondary goal. The aim was to check that the AI genuinely removed purchase friction on the product page, not just that it started conversations. In Morgane's words: test and learn.
Following Cabaia's usual process, traffic was split 50/50 at the start. Once the Yuma variant was clearly outperforming, Morgane switched Kameleoon to automatic allocation, which sends more traffic to the variant most likely to convert. At the end of September, she stopped the test and turned Sales AI on for every visitor.
Outcomes
- 01
+19.2% add-to-cart on product pages
On the French store, add-to-cart rose by 19.2% with Yuma compared with the control group. The test ran for 105 days across 3.49 million visits and reached over 99% statistical confidence.
- 02
+20.5% completed checkouts
The lift held further down the funnel. Completed checkouts rose by 20.5%, also at over 99% confidence. Shoppers weren't just adding more to cart; more of them were actually buying.
- 03
+23.1% revenue per visit
Revenue per visit grew by 23.1%. Average order value stayed almost flat (+2.2%), which tells us the uplift in France came from more shoppers converting, not from bigger baskets.
“Our main goal was to see the impact on add-to-cart. We got a 19.2% uplift with Yuma's solution. The primary objective was more than met.”
Morgane CollinDirector E-commerce, CabaiaBoth markets respond, in different ways
Cabaia ran the same test on its European store, where the brand is less well known and traffic is lower. Results were statistically significant there too, but the shape was different.
| Metric (EU, A/B test) | Uplift | Significant |
|---|---|---|
| Add-to-cart | +10.5% (95% CI: +6% to +15%) | Yes |
| Order rate | +7.0% (95% CI: +4% to +10%) | Yes |
| Average order value | +21.2% | Yes (over 99%) |
| Revenue per visit | +29.7% | Yes (over 99%) |
205,552 visits without Yuma vs. 1,083,038 visits with Yuma, after automatic traffic allocation.
Conversion improved less than in France, but average order value jumped 21.2%, pushing revenue per visit up 29.7%. Morgane is careful with that figure: on a lower-volume store, a handful of very large orders can skew average order value, and she plans to isolate those before drawing conclusions. The conversion gains, on the other hand, are solid on both markets.
A library of product insights
Beyond the numbers, every question shoppers ask Sales AI is a data point Cabaia never had before. The questions build a library of insights on recurring friction points: laptop compatibility (14-inch or 16-inch?), capacity, pouch and bag compatibility, water resistance, durability and the lifetime warranty.
What surprised Morgane is that almost all of this information was already on the product pages.
“It's funny, because all of this is displayed on our product pages, and pretty visibly. But I think it's just easier for everyone to ask Yuma to find the answers to their questions.”
Morgane CollinDirector E-commerce, CabaiaThat's the core of why Sales AI moves conversion. With thousands of SKUs, no team can tailor every product page to surface exactly what each shopper wants to know first. Shoppers skim. Sales AI answers the specific question in the moment, instead of hoping the shopper finds the right bullet point.
The insights also confirmed which reassurance points matter most to Cabaia's customers, and they're feeding directly into a product page redesign that's currently underway. Product launches are another use case.
“When we launch a product, it lets us take the temperature on the questions. Is our product page working? Do we have the right visuals?”
Morgane CollinDirector E-commerce, CabaiaLow-maintenance by design
Sales AI runs without daily oversight. The team checks in around product launches and reviews it on a regular cadence. It also doubles as a quick way to keep product content up to date.
Cabaia's advice for e-commerce teams considering AI on product pages
We asked Morgane what she'd tell another e-commerce team weighing up an AI agent on their product pages.
Run a clean A/B test
“I'd tell them to run the A/B test. It's the only way to prove, or disprove, the real impact before rolling it out everywhere, rather than relying on a gut feeling or deploying blind.”
Morgane CollinDirector E-commerce, CabaiaBalance the test carefully. A poor traffic split, the wrong attribution or the wrong goal can bias results or leave them inconclusive. Cabaia's traffic volume helped it reach reliable results quickly.
Check answers on sensitive topics regularly
Review what the AI says about returns, shipping and warranty, and make sure it stays aligned with the brand's official position. Drift on these topics creates tickets for customer service.
Break the analysis down by market
If you sell in several countries, look at each market separately. Friction points and behaviour can differ between, say, German and French shoppers, as Cabaia's own FR and EU results show.
“Run the test. There's nothing to lose, only things to learn.”
Morgane CollinDirector E-commerce, CabaiaConclusion
Cabaia didn't take Sales AI on faith. It put it through a 105-day A/B test across 3.49 million visits, with a clean control group and a clear primary goal. The result: +19.2% add-to-cart, +20.5% completed checkouts and +23.1% revenue per visit on the French store, plus significant conversion gains on the European store.
Just as valuable is what Cabaia now knows about its shoppers. The questions they ask are feeding a product page redesign and giving the team a fast read on every new launch.
Next up, Morgane is curious to see how shopper questions change over the holiday season, when Cabaia becomes a gift destination.
For brands with a large catalog, technical or modular products, and the ambition to bring in-store advice online, Cabaia's test shows what's possible when the AI answers the question at the exact moment a shopper hesitates. To see how it works, visit Yuma Sales AI, or book a demo and we'll walk you through how to run the same test on your store.
