Custom E-commerce Development: From Catalog to Integrations

A client's e-commerce store with a catalog of 20,000 SKUs lagged on every request. An OpenCart installation on shared hosting with PHP 7.4 delivered a catalog page in 5 seconds — LCP was 8 seconds, half of users left before load. After migrating to PostgreSQL 15 with JSONB attributes and integrating Elasticsearch 8.x, load time dropped to 200 ms, CLS decreased from 0.35 to 0.02. That's how a proper e-commerce store is built — not a template, but an architecture where every detail affects speed and order pipeline.

Development starts not with choosing button colors, but with the business model: inventory management, pricing, logistics, marketing — everything must be aligned before a line of code. Choosing the wrong stack upfront saves time initially but turns into technical debt within a year. We perform a free audit and propose optimal solutions.

Each e-commerce store requires an individual approach: data model, integrations, load. Universal solutions don't exist. Contact us for a consultation on your project — we'll evaluate the architecture and recommend the right stack.

How to Choose a Technology Stack for an E-commerce Store

The stack is determined by three factors: catalog size, integration scenarios, and expected traffic. Typical options:

For most mid-range projects, a monolith with clear module boundaries is optimal — faster to develop, easier to maintain.

Why the Catalog Data Model Is Critical

The catalog data model determines everything else. A common mistake is creating a products table with columns color, size, material directly. This works until the first non-standard product. The correct approach: EAV (Entity-Attribute-Value) or JSONB attributes in PostgreSQL. Example schema:

products (id, sku, slug, base_price, status) product_variants (id, product_id, sku, price_delta, stock) variant_attributes (variant_id, attribute_id, value) attributes (id, name, type, filterable, sortable) 

For catalogs with heterogeneous products (electronics + clothing + furniture), we use product types — each type defines its own set of attributes. This is how Akeneo PIM works, which we connect for catalogs from 10,000 SKUs.

Search and Filtering

Standard SQL search LIKE '%query%' is unsuitable even for a catalog of 500 items — no relevance, morphology, or synonyms. Options:

• PostgreSQL FTS with ts_vector — suitable up to ~5,000 products, configure for Russian via ispell or hunspell
• Elasticsearch — full-featured: fuzzy search, field boosting, synonyms, percolator for "similar products". According to Elasticsearch Reference, aggregations allow counting documents for each filter value.
• Typesense — simpler to operate, good for small teams

We build filters via Elasticsearch aggregations: users see only those filter values that yield non-zero results. This is faceted search — the ecommerce standard.

How to Implement Faceted Search

1. Determine product attributes that will be filters (price, brand, size).
2. Create a mapping for the index with correct types (keyword, float).
3. Index products with their attributes.
4. Run aggregations for each filter in a single query.
5. On the frontend, display aggregated values and update results when filters are selected.

Cart and Checkout

The cart is where 60–80% of conversion is lost. Critical decisions:

• Guest cart stored in Redis with TTL of 30 days, merged with server-side on login. Without guest cart, registration becomes a barrier.
• Single-screen checkout statistically outperforms multi-step for B2C. For B2B, the opposite — steps are needed: billing info, delivery address, manager approval.
• Real-time shipping calculation: APIs from CDEK, Boxberry, Nova Poshta, DHL. Rates depend on weight, dimensions, origin point — computed on the fly, not stored in a table.
• Payment gateway: Stripe for international, CloudPayments / YooKassa for Russia/CIS, Fondy / WayForPay for Ukraine. Integration via webhook confirmation; order status changes only after signature verification. Stripe fee: 2.9% + $0.30 per transaction.

Order Management

Order lifecycle is a state machine. States: draft → pending_payment → paid → processing → shipped → delivered → completed. Plus side branches: cancelled, refunded, on_hold. Libraries: winzou/state-machine for Laravel, xstate if part of the logic is on the frontend. Each state transition triggers notifications (email, SMS, push), inventory updates, and logging.

Integrations

Typical set for a full-fledged store:

• 1C / MySklad / Retailio — synchronization of stock and prices via a queue (RabbitMQ/Redis Queue) with a 5–15 minute interval
• CRM (AmoCRM, HubSpot) — automatic transfer of orders and leads
• Email marketing (Klaviyo, Mailchimp, Sendpulse) — trigger sequences: abandoned cart, post-purchase, reactivation
• Analytics: GA4 with Enhanced Ecommerce + server-side GTM for accurate data without ad blockers

Performance and SEO

Catalog pages must be cached. Strategy: stale-while-revalidate — serve cache immediately, update in the background. Redis for server-side cache, CDN (Cloudflare) for static assets. Ecommerce-specific SEO:

• Canonical URLs for filter pages — without this, search engines see thousands of duplicates
• hreflang for multilingual stores
• Schema.org markup: Product, Offer, AggregateRating — affects rich snippets
• Pagination via rel="next"/"prev" or infinite scroll with SSR rendering of the first screen

What's Included in Our Work

• Technical specification with functional description and integrations
• Interface prototyping (Figma)
• Backend development: catalog, cart, orders, personal account modules
• Frontend on a modern stack (React/Vue) with responsive design
• Integrations with payment systems, delivery services, warehouse systems
• Testing (unit, integration, load)
• Production deployment and monitoring setup
• Administration documentation
• Post-launch support (one-month warranty for bug fixes)

Timelines and Stages

Typical medium-scale e-commerce store (1,000–10,000 SKUs, standard checkout):

Total: 12–20 weeks for a full project without ready-made templates. Shopify projects with minimal customizations — 4–8 weeks. The key principle: an e-commerce store launches not when everything is ready, but when enough is ready for the first sales. The rest is iterations. Contact us to discuss details.