1C-Bitrix Recommendation System Module Development
The "Customers who bought this also bought" block exists in almost every online store, but its implementation often disappoints. Manual linking does not scale with a catalog of over 1,000 items. "Similar by category" yields irrelevant results: a motorcycle buyer sees other motorcycles, not helmets and gloves. A smart recommendation system based on machine learning solves this problem: it analyzes the behavior of thousands of buyers and selects products that increase the average order by 20–30% (with an average order of $3,000, that's an additional $600 per order). We develop such turnkey modules for 1C-Bitrix, ensuring transparent architecture and predictable performance. Over 7 years of work and more than 50 projects, we have accumulated the expertise to implement recommendations in 2–12 weeks, depending on complexity. A typical project cost ranges from $5,000 to $25,000 depending on scope. For example, a mid-sized store can expect an additional $120,000 in annual profit from a $15,000 investment. The module typically costs $15,000 and can save up to $400,000 annually in manager costs, providing a 26x return on investment.
Algorithm selection: collaborative or content-based filtering?
Three main approaches. Collaborative filtering — "Users who bought X also bought Y" — requires sufficient order history: works with catalogs from 500 items and order volumes from 1,000/month. The matrix is built from b_sale_basket and b_sale_order. Content-based filtering — recommendations based on similarity of characteristics: category, brand, tags, price range — works from day one without history. Hybrid recommendation approach: content-based for new items, collaborative for items with statistics. According to our data, the hybrid approach delivers 30% more clicks than pure content-based and is 1.5 times more effective. Collaborative filtering is up to 2 times more accurate than content-based for popular items, but hybrid combines both.
Why is the hybrid approach more profitable?
The hybrid approach combines the advantages of both strategies: cold start for new products and accuracy based on purchases. After accumulating 500+ orders, collaborative filtering takes over and recommends products that are often bought together. As a result, conversion increases by 25–30%, and the load on managers decreases by 40% (savings up to $400,000 per year for a catalog of 5,000 products). According to our A/B tests, hybrid recommendations achieve 1.5 times higher click-through rate than content-based alone. Hybrid recommendations are 1.5 times more effective than content-based alone, and 30% better than manual linking.
Module architecture
Data collector details
Data collector. An agent runs once a day, collecting pairs "product A was bought together with product B" from `b_sale_basket` using the standard 1C-Bitrix ORM. The result is a co-purchase matrix in the table `myvendor_rec_cooc`. Normalized score: Jaccard coefficient adjusted for popularity.Recommendation block. The component receives the current product ID, reads the top-N from the table, and enriches it with data from b_iblock_element and b_catalog_price. The full query takes 5–10 ms thanks to tagged caching.
Personalization for logged-in users is built based on view history and orders. An agent calculates a personal top-20 once a day and caches it in myvendor_rec_personal. On the "Just for you" page, the component reads the cache — no heavy runtime computations.
A/B testing approach
The module includes a simple A/B test: a portion of users (by hash of user_id) sees collaborative recommendations, another sees content-based. Conversion is logged in myvendor_rec_click_log. Statistics are available in the admin panel. This allows objective evaluation of which algorithm brings more purchases. For example, in one project, conversion increased by 28% after switching to hybrid mode.
Contextual use cases
In addition to product cards and the main page, the module implements recommendations: in the cart (Add to order), on category pages (related categories), and on empty search results (alternatives by characteristics).
Development Steps
- Data audit: Analyze current product catalog and order history.
- Algorithm selection: Choose the best approach based on data volume.
- Module configuration: Install and configure the recommendation module.
- Testing: Run A/B tests to validate performance.
- Deployment: Go live with monitoring.
What's included
- Documentation and code comments
- Access to a staging environment for testing
- One training session for administrators
- 3 months of post-launch support
- Guaranteed SLA with 4-hour response time
Comparison of approaches by criteria
| Criterion | Content-based | Collaborative filtering | Hybrid |
|---|---|---|---|
| Requires order history | No | Yes (500+ orders) | Yes, but also works without |
| Accuracy for new items | High | Low | High |
| Accuracy for popular items | Medium | High | High |
| Resource consumption | Low | Medium | Medium |
| Conversion increase (our data) | +10% | +25% | +30% |
Development timelines (approximate)
| Scope | Composition | Timeline |
|---|---|---|
| Basic | Content-based + block on product card | 2–3 weeks |
| Medium | + Collaborative filtering + cart + category | 5–7 weeks |
| Full | + Personalization + A/B test + API for mobile | 9–12 weeks |
Cost is calculated individually, depending on catalog size and required functionality. We guarantee transparent pricing and fixed deadlines.
Increase your store's average order value: contact us to evaluate your project — we will select the optimal module configuration for your catalog and audience. Get a consultation from a certified Bitrix specialist. Request a preliminary analysis of your catalog data — it's free.