Six phases, one rollout.
Connecting to your catalog.
Your live product catalog connects to ob.session — pricing, stock, availability, product attributes — so every product the visitor sees is one you can actually sell.
Read-only access. Catalog data only: no customer data, no carts, no orders, no payments.
Mapping your store.
Every store is structured differently — categories, product attributes, where things appear on the page. Before the tracker goes live, ob.session maps your catalog and storefront to the structure it uses internally — making sure every relevant signal is captured on your specific store and nothing important is missed.
Handled by ob.session — at most a brief kickoff, no further input from your team.
Tracker deployment.
A small JavaScript tracker goes live on your website. It captures behavioral signals on your site — hover, scroll, dwell, clicks — building the training data the model needs.
Anonymous signals only. No personal information, no cart or order data.
Two install options: install the full pixel on your site (preferred for lower latency), or add the CDN URL — the tracker is downloaded from the ob.session CDN on each page open.
The same pixel will later also serve the recommendations on your storefront — no second install needed.
Training the model.
While the tracker accumulates sessions, ob.session analyzes your data — your catalog structure, your visitor behaviors, what drives engagement on your store. Once enough training data accumulates (~50K sessions), the model is fine-tuned to perform best on your specific store.
Handled by ob.session. The duration depends on your traffic volume.
A/B test launch.
The fine-tuned recommender engine goes live in an A/B test against your current setup — no recommender, an existing one, or anything else. You define the scope: which widgets, which collections, what share of traffic goes to each group.
Metrics run either on your existing A/B engine or on ob.session's — you have access to real-time dashboards to monitor every relevant metric live.
Exit criteria for the A/B test are agreed up front with your team.
Full rollout.
Once the A/B test produces a positive result, the recommender engine rolls out to 100% of traffic across your defined scope.
From there, the engine keeps learning from new sessions, staying tuned to your store's evolution — seasonal shifts, new products, changing visitor behavior. You retain access to the same real-time dashboards. Scope can expand to new widgets or collections whenever you choose.