Classic retargeting is simple: someone visits your site, a pixel drops them into an audience, and an ad network shows them your banner for the next thirty days. It works well enough that almost every advertiser does it, and badly enough that most of them overspend on it.
The problem is that "visited the site" is a weak signal. A buyer comparing plans and a job seeker reading your careers page land in the same audience, see the same ad and cost the same to reach. AI retargeting starts from a better question: how likely is this visitor to buy, and what would move them?
Start from intent, not from a visit
Every session leaves a trail: the pages someone read, in what order, how far they scrolled, whether they came back, whether they looked at pricing or started a form and stopped. On its own each action means little. Together they separate the researcher from the evaluator from the person who is one nudge away.
ROI.me turns that trail into an intent score for each visitor. The score is rule-based and explainable: you can see exactly which actions raised it, and you can change the rules. An AI model helps interpret what a visitor was interested in, but it doesn't decide on its own who gets spend.
Audiences that update themselves
Once visitors are scored, audiences stop being static lists. "High-intent visitors who viewed pricing in the last 7 days" is a definition, not a snapshot: people enter it when they qualify and leave when they don't, including when they convert. That last part matters. Paying to show "Book a demo" ads to someone who booked a demo yesterday is the most common way retargeting budgets leak.
Audiences can also combine sources: website behavior from the ROI Pixel, stages from your CRM and purchases from your commerce platform. Where a visitor has identified themselves, identifiers are hashed before they are matched to an ad network.
Creative that matches what someone cared about
The second half of retargeting is the ad itself. Someone who spent five minutes on your integrations page has told you what they care about; showing them your generic brand ad wastes that. AI creative generation makes it practical to produce a variant for each interest, in each format a network needs, without a designer building every one.
The guardrail is your brand. ROI.me learns your colors, voice, products and claims from your site first, then generates within those limits, and nothing launches without an approval you control.
What AI retargeting isn't
- It isn't identifying anonymous people. Scoring works on behavior within your own site. It doesn't try to work out who an anonymous visitor is from device data or by joining data across companies.
- It isn't autopilot by default. Recommendations to shift budget, pause a fatigued ad or widen an audience come with reasons, and spend changes go through the approval and limit rules you set.
- It isn't a replacement for measurement. Retargeting takes credit easily. Judge it on conversions you define, deduplicated across the pixel and the ad networks, not on view-through numbers alone.
A practical starting point
If you're running retargeting today, three changes pay back quickly: exclude people who already converted, split your single "all visitors" audience by intent, and give each segment creative that speaks to what they looked at. That is the core of AI retargeting. Automation makes it possible to keep it up across every network and every week.