← All case studies
Kaspi · 2025

Kaspi AI: a product listing in 5 minutes instead of 9

RoleProduct Designer
Design time1 month
TeamPO, team lead, engineering
Year2025
9 → 5 min

to create one product listing

77% → 82%

conversion to adding a product

200K

listings created with Kaspi AI in May

Problem. One product listing takes 9 minutes of manual work and three steps: pick a category, add photos, fill in the specs.

Sellers filled in the specs by hand, wrote the description and took product photos themselves. For new sellers, uploading their whole catalog from scratch was especially hard.

Solution. The seller adds a photo, and Kaspi AI fills in the rest.

From a photo, a link or a file, Kaspi AI identifies the product, category and brand, puts together the specs, writes a description and creates studio-quality photos.

Manual entry didn’t go anywhere: “Add product manually” is always right there for those who prefer the old way.

Introducing Kaspi AI
Introducing Kaspi AI
A photo, a link or a file
A photo, a link or a file
Photos added
Photos added

When AI isn’t sure, it doesn’t guess — it asks the seller.

  • Category and brand. If AI isn’t sure, it doesn’t pick on its own — it offers the seller a few options.
  • Materials and country of origin. You can’t see them in a photo, so AI doesn’t make them up — it asks for a photo of the label, the packaging or a document. This step can be skipped.

This way, the seller doesn’t have to hunt down and fix mistakes across the whole listing later.

AI suggests categories
AI suggests categories
Asking for a photo of the label
Asking for a photo of the label

Generation. Instead of an endless spinner, you see what AI is doing at every step.

“Analyzing the label” → “Putting together the specs” → “Writing the description” → “Creating photos”. Specs appear line by line, so the seller immediately sees what AI understood about the product. AI identifies the category in 7 seconds and prepares the specs and photos in 38 seconds.

Analyzing the label
Analyzing the label
Specs appear line by line
Specs appear line by line
Writing the description
Writing the description

The final say. AI creates a draft, but the seller decides.

The first idea was that AI would fill in everything on its own, and edits would only be possible after publishing. We dropped it: the seller should check the listing before buyers see it.

Any block can be edited: specs, description, photos. All that’s left for the seller is to add the dimensions and the price.

Almost done: AI double-checks what it isn’t sure about
Almost done: AI double-checks what it isn’t sure about
The finished listing
The finished listing

Existing listings. AI also improves listings that are already live.

In the product list, each listing shows its quality — low, medium or high — and the “To improve” tab collects listings that AI can polish. The seller doesn’t have to create the product all over again.

Listing quality in the list
Listing quality in the list
Medium quality → “Improve”
Medium quality → “Improve”
AI explains what’s wrong: too few photos, a short description
AI explains what’s wrong: too few photos, a short description
High quality
High quality

Result. A listing in 5 minutes instead of 9, and adding products got easier.

  • Almost 2× faster: 9 minutes per listing by hand and 5 minutes with Kaspi AI. Instead of three manual steps — one required action: add a photo.
  • Conversion to adding a product grew from 77% to 82%.
  • Sellers adopted it quickly: almost 200K listings were created with Kaspi AI in May, and the usage share grew from 17% to 27% in 4 months.
  • Kaspi shared these numbers at a public presentation streamed on YouTube.
  • AI has already improved around 500K of the most popular listings — Kaspi shared this in its investor report (November 2025).

What I learned. In an AI product, you design the wait.

This was my first AI product. Generation means pauses: without design, people see a spinner and think it’s frozen. When every step is visible, the wait becomes part of the product.

Next case study

Kaspi Gold in Apple Pay and Google Pay: 1M cards in 3 days

→