The Zeleny Bazaar in Almaty operates every day, in every weather. Vendors know their regulars by name. Prices are discovered through conversation, not price tags. Quality is assessed by touch, smell, and the reputation of the seller's family going back two generations. It's a social infrastructure as much as a retail one.
When a client asked us to build a B2C marketplace for food and agricultural goods in Kazakhstan, we made the mistake most outside teams make: we defaulted to the Amazon mental model. Cart. Checkout. Reviews. Done.
It took about three weeks of user research to understand why that approach was going to fail.
The Trust Problem Is Different Here
In markets where cash-on-delivery is still the dominant payment method — Kazakhstan saw ~60% COD for e-commerce in 2025 — digital trust infrastructure operates differently. Users are not extending trust to a platform. They're extending trust to a specific vendor, and the platform must not obscure that relationship.
What this meant for our architecture:
- Vendor profiles were primary, not subordinate to product listings. A buyer navigates to a seller first, then browses that seller's inventory — not the reverse
- Messaging was not a feature, it was the core loop. The WhatsApp-style chat between buyer and seller was where price negotiation, freshness questions, and delivery timing happened. We integrated Telegram bot notifications because sellers were already managing their businesses there
- Reviews were insufficient on their own. We added a "repeat order" metric — buyers could see what percentage of a vendor's orders came from returning customers. This single metric converted better than a star rating in user testing
The star rating is a Western shortcut for trust. In Kazakhstan, the question is not "is this product good?" but "can I trust this person?" Those are different questions with different answers.
The Payment Complexity Nobody Mentions
Kazakhstan has a fragmented payment landscape that no international payment library handles gracefully:
- Kaspi dominates with ~70% of digital transactions for B2C goods
- Halyk Bank processing for business accounts
- QIWI still relevant for peer-to-peer transfers in some demographics
- COD with courier verification codes
We ended up building a payment routing layer that detected the buyer's most-used payment method (with permission, from Kaspi's open banking API) and defaulted the checkout to it. For a first-time buyer, the default was COD with an option to upgrade.
This dropped cart abandonment by 34% from our baseline.
Logistics Is a Product Decision
In Kazakhstan, "last-mile delivery" is not a commodity infrastructure problem — it's a genuine competitive differentiator. The courier market outside Almaty city center is fragmented, unreliable, and expensive relative to order values.
Our decision: integrate with Яндекс Доставка (Yandex Delivery) for urban zones and build a native "seller delivers" mode for agricultural vendors who had established routes to specific neighborhoods.
The seller-delivers model required giving vendors a simple mobile interface to:
- Confirm delivery slots with buyers
- Generate a QR-coded delivery receipt that the buyer scans to release payment
- Log failed deliveries with photographic evidence for dispute resolution
This is infrastructure that Shopify, WooCommerce, and standard marketplace SDKs don't provide. We built it custom, and it became our strongest competitive moat.
What the Data Showed at Six Months
The platform went live in April 2025 with 47 vendors across three product categories (fresh produce, dried goods, dairy). By September 2025:
- GMV: 18.4M KZT monthly (~$37k USD)
- Repeat buyer rate: 61% (vs. 23% industry benchmark for CIS food marketplaces)
- Average vendor rating: 4.7/5 — but more tellingly, 74% of rated vendors had >80% repeat order rate
- COD to digital payment migration: 58% of buyers who started on COD switched to Kaspi within their first 90 days
The platform didn't succeed because we replicated an existing model. It succeeded because we studied why the Zeleny Bazaar works and translated those mechanics into a digital system — social trust, vendor personality, payment flexibility, and logistics that matched actual seller capabilities.
Lessons for Building CIS Marketplaces
- Audit the existing informal commerce in your target vertical before writing a line of code. The bazaar is not your enemy — it's your requirements document
- Chat is not a support channel. It's a commerce channel. Build it first
- Payment defaults matter more than payment options. One click to the right method is worth five options at checkout
- Seller logistics capacity is a first-class data point. A marketplace is only as strong as its worst reliable vendor
- Repeat order rate beats star rating as a trust signal in high-social commerce environments