What 50 Million Shopping Sessions Taught Us About Browse Behavior

We've been collecting and analyzing shopping session data since 2022. Over 50 million sessions across 300+ stores, covering apparel, home goods, electronics, beauty, sporting goods, and specialty retail. At that scale, patterns emerge that aren't visible at the store level but are consistent enough to act on.

What follows is a summary of the most actionable findings — the things that actually changed how we build personalization logic and how we advise stores to structure their shopping experience.

Most purchase decisions are made within the first three products viewed

The popular mental model of an e-commerce shopper is someone who browses extensively, considers many options, and then reaches a confident purchase decision after thorough comparison. This is not how most purchases happen.

Across purchase sessions in our dataset, 61% of shoppers who bought something added their first purchased item to cart within the first three products they engaged with meaningfully (defined as spending 8+ seconds on a product page or scrolling below the fold). They may have viewed more products before buying, but the item they purchased was encountered early.

The implication: what you show in positions 1-3 of any product listing, recommendation widget, or search result matters enormously. Stores that optimize for "showing the catalog" — rotating through variety to give the shopper broad exposure — are misaligned with how shoppers actually work. The goal is to get the right product in front of the shopper early, not to expose them to everything.

The "consideration set" is smaller than you think

Related finding: in sessions that ended in a purchase, shoppers engaged meaningfully with a median of 4.2 products before buying. The entire consideration set — the universe of things they were seriously evaluating — was roughly four products.

In sessions that did not end in a purchase, shoppers engaged with a median of 7.8 products. Browsing more without buying is a signal of discovery mode or decision difficulty, not of someone building toward a higher-confidence purchase.

This asymmetry has a direct implication for recommendation strategy. When a shopper is in a low-product-count, high-engagement pattern — a few products viewed, significant time per product, possibly a size check or description scroll — they're in purchase mode. Your recommendations at that point should be tightly related to what they've engaged with, not a broad catalog sweep. When they're in a high-product-count, low-engagement pattern, they're still exploring. Broader, more varied recommendations that help them discover something that resonates is the right approach.

Mobile sessions browse more but buy less per session

Mobile now represents about 65% of e-commerce sessions in our dataset but only about 42% of completed purchases. Desktop sessions have a significantly higher purchase conversion rate despite representing a smaller share of traffic.

This gap has narrowed over time — mobile checkout has gotten better — but it persists. The behavioral data explains why. Mobile sessions are shorter (median 4.2 minutes vs. 8.7 minutes on desktop), involve more page views per minute (suggesting faster, less deliberate browsing), and have higher exit rates from product pages where the visual content requires horizontal scrolling or fine detail examination.

The personalization implications are substantial. Mobile shoppers are more impulsive — they're browsing for stimulation, not conducting research. That means early-session personalization matters more on mobile: you need to hook them with the right product quickly before they exit. Desktop shoppers are more deliberate — they're evaluating. Comparison-friendly layouts, detailed product attributes, and cross-sell timing that matches longer session durations are more valuable on desktop.

Stores that serve the same personalization experience on mobile and desktop are leaving conversion on the table on both. The right signals, right timing, and right presentation are different for each.

Return visitors behave fundamentally differently from first-time visitors

First-time visitors spend more time on the site (median 6.1 minutes vs. 4.4 minutes for returning visitors with purchase history), view more products (8.3 vs. 5.1), and have higher cart abandonment rates (81% vs. 67%). They're learning the catalog.

Returning visitors who've purchased before have a much more focused behavior pattern: they come back with a clearer intent, navigate to the relevant category or search for a specific product type, spend less time browsing, and when they buy, they buy faster. The second purchase decision is easier than the first.

Treating these two populations with the same experience is a significant missed opportunity. First-time visitors need discovery — content, social proof, brand orientation, and a broad enough recommendation sweep to find something they respond to. Returning purchasers need efficiency — get me to the right product quickly, show me what's new in the categories I've bought from, don't waste my time with things that aren't relevant to my established preferences.

Price range is a stronger personalization signal than category

One finding that surprised the team and has since changed our recommendation weighting: within a session, the price range a shopper is engaging with is a stronger predictor of what they'll buy than the category they're browsing.

A shopper browsing home decor at $40-$80 is more likely to buy a different home decor item in the $40-$80 range than they are to buy a home decor item at $200. That much is obvious. But they're also more likely to buy an apparel item in the $40-$80 range than a home decor item at $200 — if a cross-category recommendation appears.

Price range is essentially a proxy for the shopper's budget state in this session. Someone browsing at premium price points is in a premium buying frame. Someone browsing at value price points is in a value buying frame. Recommendations that match the price frame of the session outperform category-matched but price-mismatched recommendations by about 30% on add-to-cart rate in our data.

The seven-minute threshold

Sessions that last between 5 and 12 minutes have the highest purchase conversion rates in our dataset. Under 5 minutes is mostly quick visits that rarely convert. Over 12 minutes starts to suggest the shopper hasn't found what they're looking for — conversion rates drop off as sessions extend beyond that point.

The 5-12 minute window is the active purchase zone. Shoppers in this window are engaged enough to have found something relevant but haven't yet reached the decision-fatigue or confusion state that long sessions suggest.

The implication for real-time personalization: the system should be most aggressive about surfacing high-confidence, conversion-focused recommendations during the 5-12 minute window. A shopper at minute 3 is still orienting. A shopper at minute 7 who's engaged with two or three products is in the window where a well-placed recommendation can close the session.

What this means for your stack

The through-line across all of these findings is that generic personalization — the same logic for every visitor, every device, every stage of the session — leaves most of the value on the table. Browse behavior is patterned. The patterns are knowable. Acting on them requires a system that's tracking session signals in real time and adjusting its outputs based on what those signals indicate.

The good news is that the signals aren't mysterious. Products viewed, time per page, price range engaged with, session duration, device type, visit history — these are all standard data points. The difference between stores that are capitalizing on them and stores that aren't is whether their personalization layer is actually using them, or just running static recommendations dressed up with first-name mail merge.