How a controlled product data feed experiment changed Sweaty Betty’s product discovery performance

Summary

Sweaty Betty syndicates its product feed to six channels across three markets, and until this year the feed described its products in less depth than its own website did. The brand could not generate accurate, product-specific highlights at catalog scale without writing and maintaining rules product by product, which limited how well Google could match its products to the questions shoppers actually ask.

In the second quarter of 2026, Sweaty Betty ran the Athos Commerce Channel Assistant against two priority segments, running and yoga, using the AI-generated product highlights capability. The work was structured as a controlled A/B split: a test group received AI-generated highlights, and a matched control group was held back.

The yoga test group returned 62% more impressions, 72% more clicks, and 80% more revenue than its control. The running test group returned 79% more clicks and 43% more revenue.

Background

Sweaty Betty is a premium British women’s activewear brand founded in Notting Hill, London. What began as a single boutique now sells globally, with technical, style-led performance wear built for yoga, running, and training. The product range runs from leggings and tops through outerwear and accessories.

The brand went live with Athos Commerce in January 2026, migrating from GoDataFeed. Athos manages Sweaty Betty’s product feed syndication to Google, Meta, TikTok, AWin, Braze, and StoryStream across the UK, Irish, and US markets. Behind those feeds sits a layer of custom logic: rules that handle stock fragmentation, exclude archive sale items, map product categories, and apply the custom labels that segment campaigns. Athos also runs the paid search experiments that test changes to the feed before they reach the whole catalog.

By the time Channel Assistant entered the picture, the two teams had been working on the same product data together for months. The experiment described here is an expansion of a running relationship, not a first deployment.

The challenge

Sweaty Betty’s product feed carried less detail than its own website product pages. The team could not generate product highlights and enriched product data that matched what shoppers saw on site, which left Google Shopping describing products in less depth than the website did.

That shortfall lands hardest on one specific attribute. On Google Shopping, the Product Highlights attribute feeds both product quality scoring and the way a shopper evaluates a product before clicking. As shoppers move between traditional search results, Google AI Mode, and Gemini, product data that states what a product is for in plain language decides whether an engine can match it to a query at all. Well-structured product highlights give the ranking system and the shopper the same information.

The search queries themselves have changed shape, especially with the rise of conversational search.. A shopper looking for yoga shorts asks what will stay comfortable in a UK heatwave, or what will keep them cool through a class. Material and color alone answer neither question.

Building highlights of that kind by hand does not scale. Someone has to review each product and work out which selling points and use cases apply to it. Then that judgment has to be encoded as rules, and the rules have to be maintained as the range turns over.

Objectives

Sweaty Betty set three objectives for the test.

• Generate accurate product highlights at catalog scale: produce product-specific highlight text across a large, fast-turning range without writing per-product rules.
• Improve the quality of paid search traffic: reach shoppers whose search queries match what a product is actually built for.
• Prove the effect before expanding: measure against a held-back control group, and widen the rollout only on the result.

The solution

Sweaty Betty adopted the Athos Commerce Channel Assistant, an AI agent that audits and optimizes product feed content. One capability carried this test: AI-generated product highlights, which write the Google Shopping Product Highlights attribute for each product from the product data already flowing through the feed.

The Experimentation capability applied the generated highlights to a defined product set and held a matched set back untouched, which is what makes the comparison attributable to the highlights rather than to seasonality or spend. The Performance capability then tracked what each set did across impressions, clicks, and revenue.

Sweaty Betty chose two priority segments. Running came first, because the brand had already seen results from Athos product highlight experiments on its ‘Run’ persona products earlier in the year, timed to the London Marathon. Yoga followed for a different reason. After the Wimbledon Tennis Championships and a brand sponsored padel tournament, Sweaty Betty wanted to bring attention back to its core line, and yoga products are less exposed to weather and event cycles than seasonal product categories and activities are.

Implementation

The experiment ran as a controlled A/B split across matched product sets in each segment.

• Group A (control): products with no AI-generated changes applied. Some still carried manually created, rules-based highlights from earlier work.
• Group B (test): products with AI-generated product highlights applied through Channel Assistant.

Athos tracked impressions, clicks, and revenue for both groups and compared them two ways: the straight difference between groups after launch, and a before-and-after comparison that accounts for each group’s baseline in the period before the test began.

The results

Yoga produced the strongest signal. The test group of 161 products with AI-generated product highlights gradually pulled away from its control after launch.

• 62% more impressions than the control group.
• 72% more clicks than the control group.
• 80% more revenue than the control group.
• 52% higher impression share, with return on ad spend (ROAS) up 40%.

Running returned a consistent result.

• 79% more clicks than the control group.
• 43% more revenue than the control group.
• 25% higher impression share.

Looking ahead

The two segments were chosen to isolate the effect of AI-generated product highlights across contrasting demand patterns, one exposed to event-driven spikes and one largely insulated from them, and both returned revenue growth against a held-back control. That result gives Sweaty Betty a measured basis for deciding where AI-generated product highlights go next.

Athos continues to run Sweaty Betty’s product feed syndication across all six channels and three markets, along with the AI data-quality audits, Local Inventory Ads work, and launch feeds that sit alongside the paid search program.

What transfers to other retailers

• Test product feed changes against a held-back control group. Most feed tools apply a change across the catalog and report the aggregate move, which cannot separate the change from seasonality or a spend increase. Because the Experimentation capability runs the split inside the same platform that manages the feed, the comparison uses the same product data and the same syndication path for both arms. Sweaty Betty’s control group even retained its older manual highlights, which makes the measured lift a floor rather than a ceiling.

• Generate highlights from the product data you already syndicate. Channel Assistant reads the same product data model that populates Google, Meta, TikTok, AWin, Braze, and StoryStream. A highlight generated for Google Shopping is therefore built from the same source that feeds every other channel, so improving the product data improves every destination rather than one export.

• Measure at the segment level, not the account level. The Performance capability reports by product set. That is what let Sweaty Betty see two different results in yoga and running instead of one blended number, and it is what tells a merchandiser which segment to expand into next.

Sweaty Betty’s product feed already reached six channels before this test. What changed was how well the product data in that feed described what each product is used for, ensuring product discovery matches shopping intent

Partner with Athos Commerce to transform your ecommerce store. With proven tools and expertise in search and merchandising, we empower businesses to achieve measurable growth. Contact us today to start optimizing your store for success.