We Used AI to Identify Browse-Abandonment Behavior Before It Costs You a Sale

Most stores watch the wrong drop-off point. They build an abandoned cart email, call it a recovery program, and move on. Browse abandonment, when someone looks at a product and leaves without adding it to a cart, happens more often than cart abandonment, and almost nobody builds a system for it. We built one for a fitness apparel and equipment client. Here is what we tracked, what we automated, where AI helped and where it didn't.

What browse abandonment is

Browse abandonment is when a visitor views one or more product pages, shows real interest (time on page, scrolling, image zooms, repeat visits), and leaves the site without adding anything to a cart. It sits one step earlier in the funnel than cart abandonment, where a shopper added an item and then left before checkout. Far more visitors browse than add to cart, so browse abandonment is a much bigger pool of lost revenue, even though each visitor in it is a weaker buying signal than someone who already has an item in their cart.

The two need different treatment. A cart abandoner already made a decision, so the message is usually "finish what you started," sometimes with a small nudge like free shipping. A browse abandoner hasn't decided anything yet. That message has to remind them what they looked at and give them a reason to come back without sounding like a hard sell.

Why we started tracking browse abandonment

For this client, we were already running Google Ads and Meta campaigns and had built several Omnisend flows for abandoned checkout and post-purchase review requests. Newsletter and campaign performance showed up in every report, and the account team kept asking a fair question: what happened to the much larger group of people who visited a product page and never bought? We didn't have a good answer. That gap was uncomfortable given how much traffic a seasonal campaign (in this case tied to a major event in their category) was driving to specific product pages.

So we added a browse-abandonment layer to the setup that already existed instead of starting a separate project.

How we identify browse-abandonment behavior

Identifying the behavior is mostly a tracking and segmentation problem, and it comes before any AI is involved:

1. Event tracking. Product view events fire through Google Tag Manager into GA4 and into the ecommerce platform's own event stream, tagged with product ID, category, and price.
2. A negative filter. We define the audience as "viewed product X" minus "added product X to cart" within a set window (we used 48 hours as the starting point, then adjusted per category based on typical consideration time).
3. Segmentation by product as well as by visit. A shopper who looked at running shoes gets a different message than one who looked at recovery equipment. A lot of generic abandonment tools fall short here because they treat every browse session the same.
4. Cross-channel matching. The same behavioral signal feeds both email/push automation (Omnisend) and paid retargeting audiences (Meta and Google), so a shopper who saw a product on-site can also see it in a paid placement a day later.

This stage needs no AI, only clean tracking and a clear definition of the behavior you're chasing. AI comes in at the next stage, once the behavior is identified.

Where AI does the work

AI handles three jobs in this setup:

• Send-time optimization. Omnisend's built-in AI predicts the best send time per recipient based on their past open behavior, instead of sending every browse-abandonment message at a fixed hour. For a client with an international audience and staggered time zones, this mattered more than the message copy.
• Product recommendation blocks. Instead of a static "here's what you looked at" email, the recommendation engine pulls in adjacent products the shopper is statistically likely to want, based on browsing patterns across the whole customer base as well as their own history.
• Ad platform audience signals. Meta and Google's own algorithms use behavioral signals (including browse events we feed them via the pixel and conversions API) to build lookalike and retargeting audiences that go well beyond a simple "visited this URL" rule. Clients notice this part least and benefit from it most, because it happens inside the ad platform and never appears in a dashboard we hand them.

All three come down to timing and pattern-matching at a scale a person can't manage by hand, running inside a system that people still design and check.

What broke and what we'd do differently

Two things went wrong on this build.

DNS and domain authentication issues delayed the launch. Getting Omnisend's sending domain verified took longer than expected because DNS access sat with a third party, and email deliverability suffers badly if this isn't right before you turn on automated sends. We now check domain authentication in week one of any automation project, before any flows are built.

Segmented push campaigns needed a narrower trigger than we first used. Our first version of "shoe viewed, not purchased" pushed too broadly and felt like generic retargeting to some recipients. We narrowed it to specific product categories with a shorter recency window, and the messages started to feel timely instead of automated.

What browse-abandonment tracking added to reporting

The direct, measurable outcome we can point to is a reporting line the account didn't have before, showing what happens to traffic between a click and a cart. Budget conversations moved from "the campaign drove X visits" to "the campaign drove X visits, and here's what we're doing with the ones who didn't convert yet." For any e-commerce brand running paid traffic, the second conversation is the more useful one.

FAQ

Is browse abandonment worth automating if my traffic volume is low?

It's worth it once you have enough product-page views per week to build a meaningful segment, in our experience usually a few hundred. Below that, the effort is better spent on cart abandonment first, since it converts at a higher rate per message.

Do I need a dedicated AI tool, or can existing marketing platforms do this?

Most ecommerce email/SMS platforms (Omnisend, Klaviyo, and similar) already include AI send-time optimization and product recommendation features. You rarely need a separate AI tool to get started. What you need is clean event tracking feeding the platform you already have.

How is browse abandonment different from retargeting ads I already run?

Retargeting ads are one channel reacting to the same behavior. A full browse-abandonment program coordinates that behavior across email, push, and ads, so a shopper gets one consistent story instead of three disconnected messages.

What's the biggest mistake agencies make setting this up?

They treat every browse session the same. A generic "you looked at something" email performs far worse than messages segmented by product category and recency.

If your store has clean event tracking but no plan for the traffic that doesn't convert on the first visit, that's usually the gap to close before spending more on new traffic. It's part of how we approach performance marketing for ecommerce clients, and it plugs directly into the AI layer we now build into most retainers. If you're earlier in the process and still deciding what to build first, our web and digital products team can help you get the tracking foundation right before automating anything on top of it.