Titane Intelligence is releasing TITAN-REC 3.0, the third generation of its personalized recommendation engine for e-commerce sites. This version doubles recommendation precision on real purchases, adds five capabilities absent from the previous generation, and has already generated several tens of thousands of euros in revenue in just a few days in production.
What TITAN-REC is for
TITAN-REC is Titane's customer-to-product recommendation engine. For every active customer of a site, the model generates a top-10 of personalized product recommendations, ranked by relevance, and updated automatically every 48 hours.
Recommendations can be activated across the entire customer journey: on the homepage and the account page, in win-back and reactivation emails, or in post-purchase cross-sell blocks. The model relies on an end-to-end deep learning architecture that is 100% customizable, calibrated specifically on each site's data.
Overview — What changes with 3.0
The table below positions TITAN-REC 3.0 relative to the classic approaches on the market (collaborative filtering, RFM segmentation, category- or popularity-based recommendations) and relative to the model's two previous generations.
Migrating from the previous generation
For customers already running a previous generation, 3.0 plugs into the same connectors and the same activation surfaces: no overhaul on the site, CRM, or email platform side.
Three operational changes are worth noting. The output format gains the reasoning field, which justifies each recommended product and can be used directly in "For you" blocks and emails. The update frequency moves from weekly to every 48 hours. The top-10 per customer is now guaranteed, with minimum catalog coverage of 15% of active products across the whole base.
The five key advances
Five capabilities have been added relative to the previous generation. Each one answers a need observed in production on the first TITAN-REC deployments.
Where it actually gets activated
TITAN-REC 3.0 produces product recommendations for every customer, updated every 48 hours, usable by any tool capable of reading a standard export: CMS, CRM, email platform, retargeting, in-app. The model is deployable anywhere — the activation surfaces are specific to each customer. Here are a few concrete deployment ideas:
How it's evaluated
TITAN-REC 3.0 is evaluated on the purchases actually made after the recommendations are generated, with no data leakage (strict temporal split). The metrics tracked are:
The evaluation is stratified by customer profile to guarantee relevance across the whole base.
How it's deployed
The model runs automatically every 48 hours on cloud infrastructure managed by Titane. Recommendations are delivered in a standard JSON format usable by PrestaShop, Shopify, WooCommerce, Mailchimp, Klaviyo, Brevo, and most CMS and email platforms on the market via the Titane connectors.
No migration. No overhaul of what's already in place. Titane handles the entire process: data ingestion, training, deployment.
First production deployment
TITAN-REC 3.0 was deployed quickly with a first Titane customer. In just a few days, the model has already generated several tens of thousands of euros in revenue attributed to its personalized recommendations.
What TITAN-REC 3.0 doesn't do yet
Three areas are still to come in upcoming versions and are already the subject of early internal tests.
Visual understanding of products
TITAN-REC will soon understand products visually, directly from catalog images. This capability will unlock the recommendation of visually similar products — color, shape, style — beyond text and behavioral signals alone.
Cross-device without authentication
Recommendation consistency for unauthenticated visitors remains a focus area for upcoming versions.
Real-time activation
Recommendations that adjust continuously (beyond the 48-hour update cycle) — inference roadmap in design.
What's next
The main effort over the coming weeks is the production deployment of TITAN-REC 3.0 across our customers : generating measurable incremental revenue, then multiplying the use cases on each site (homepage, email, post-purchase, retargeting, in-app) to put the model to work across the entire customer journey.
