Boozt

Objective

Lift conversion and basket size on a fashion site where shoppers arbitrate between thousands of items, without adding production cost for every new item. The conversational assistant takes over the selection work the customer would otherwise do alone inside filters, and image generation replaces part of the photo shoots. Both push the same outcome: more orders converted per visitor, with a content cost that no longer grows at the pace of the catalogue.

The deployment

Boozt runs two online fashion department stores in the Nordic countries, Boozt.com and Booztlet.com. The company placed a conversational shopping assistant between the catalogue and the customer: instead of filtering alone through a wide assortment, the shopper describes what they want and the assistant proposes products. Reporting its second quarter 2026 results, management put a number on the assistant: customers who use it convert roughly 2.5 times more than the normal rate and spend about 8% more per order. In parallel, Boozt industrialized the production of on-model product visuals: around 10,000 AI-generated model images since April 2026, a pipeline that cuts content cost and shortens the delay between receiving an item and putting it online. The quarter itself was strong, with 13% revenue growth and an EBIT margin around 6.5%, but nothing in the sources attributes that growth to AI: the only figures tied to the system are those of the assistant and the image volume. What the case does not say: what share of visitors actually use the assistant, and whether the conversion gap measures the effect of the assistant or the stronger intent of customers who choose to use it. Management also notes that average order value on Boozt.com declined year over year, driven by a higher proportion of new customers and fewer items per basket, which coexists with the higher basket observed among assistant users.

Results Proof B

The figures come from Boozt's second quarter 2026 results, presented on 14 August 2026, which is a level A event by nature. The level retained is B rather than A because the primary document could not be reached at collection: both accessible sources are secondary accounts of the earnings call (GuruFocus relayed by Yahoo Finance for the second quarter, GuruFocus relayed by Investing.com for the first). The figures retained measure the AI system itself, conversion among assistant users, basket among those same users, volume of images generated, and not company context: the quarter's 13% growth is not attributed to AI by any source and is not counted as a result. Caveat: the conversion gap is measured between users and non-users of the assistant, not against a control group, so it blends the effect of the system with self-selection by the customers who use it.

The internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.

1. 1
   Stating the need customer

   The customer describes what they are looking for to the assistant rather than filtering alone through the catalogue.

2. 2
   Assisted selection AI

   The assistant queries the catalogue and proposes items, with descriptions and styling suggestions.

3. 3
   Visual production AI

   On-model images of the items are generated in batches and published on product pages.

4. 4
   Segmented measurement data team

   Conversion rate and basket among assistant users are tracked separately from the rest of the traffic.

5. 5
   Assortment decisions marketing

   What the assistant reveals feeds merchandising and the choice of items to promote.